Anthropic just moved the deadline. Fable 5’s included window was set to close yesterday, July 7, and today the company announced on its official X account that paid subscribers get five more days, through July 12. That’s on top of the comeback itself: the Commerce Department lifted the export controls on June 30, Anthropic flipped the switch on July 1, and the most capable AI model the public can touch is back in the Claude menu like the shutdown never happened.

I’ve covered this story from the start. The shutdown three days after launch, how export controls actually work, and what a trillion-dollar company getting switched off by one letter should teach anyone building a business on AI tools. If you missed those, they’re here and here. So I’ll keep the recap short and get to the part that matters, which is what I’d actually DO with this model in the next five days.

How it came back

Anthropic spent the blackout proving the government’s case didn’t hold. Its own testing showed that older Claude models, GPT-5.5, and even Kimi could find the same software vulnerabilities that triggered the ban. The exploit that got it pulled off the market was something half the industry could already do.

What satisfied Commerce was a new safety filter, built with the government and Amazon, that blocks the reported technique in more than 99% of cases. The trade-off is a touchier model. Some legitimate coding and debugging requests now get flagged and quietly handed to Opus 4.8 instead, but Anthropic says it’s working to reduce those false alarms. Mythos 5, the version without the guardrails, stays locked to roughly 100 vetted US organizations.

Back in June I told you to watch HOW the government backed off, because if the models returned without anyone admitting error, the only thing established would be that Washington can reach in and flip the switch whenever it wants. The Commerce Secretary posted that his department “worked closely with Anthropic to analyze and approve Fable 5.” Cooperative tone, zero acknowledgment that the original case fell apart under testing.

Why that matters to you: a shutdown that ends with no admission of error is a shutdown that can happen again on the same thin evidence. The precedent is now set. One letter, no hearing, no proof required, and the tool your business runs on can go dark for three weeks or be removed entirely. Nobody involved paid a price for getting it wrong, which means nothing discourages the next one. That’s the operating reality you’re building in, and it should shape what you use this model for.

What you have until July 12

Through July 12, Fable 5 is included on Pro, Max, Team, and select Enterprise plans for up to half your weekly usage. This is the window’s second reset. The original one, June 9 through June 22, got swallowed by the shutdown three days in, and the July 7 deadline just got the five-day extension.

After July 12 it moves to usage credits at $10 per million input tokens and $50 per million output tokens, which makes it double the price of Opus 4.8 and the most expensive model Anthropic sells. So July 12 doesn’t end your access, just the ability to use it inside your existing subscription. Five days of included time on a model this capable is worth attacking with a plan instead of messing around with it for twenty minutes and wandering off.

What I’d build, knowing what I know

A few weeks ago I told you not to stake your income on the frontier model, and then a government blackout made the argument for me. Nothing about the model coming back changes that. It already disappeared once, and the mechanism that took it out is still sitting there, fully intact.

Which is exactly why the smart play this week is one-time assets. Projects with a beginning and an end that leave you holding something when they’re done. A finished product. A full audit of your business. If a letter takes Fable out again next Tuesday, everything you built during the window still belongs to you.

And this happens to be the work Fable was built for. Anthropic’s own framing is that the longer and more complex the task, the larger Fable’s lead over every other model. You hand it a big project, it runs the whole thing, and you review finished work instead of supervising every step. I felt that the first day I used it. It turned out landing pages in a fraction of the time Opus takes, and when I gave it ten setup and structure documents from a client’s business and asked it to evaluate them, it went through all ten in ten minutes and told me what was solid, what needed work, and what needed a rebuild from the ground up. That kind of review used to take days.

So that’s the bar for this week: big projects, handed off whole, finished when they come back. I picked three you can run with your included usage, each one ending with an asset you keep, and I’d execute them in this order.

1. Build a digital product from what you already know

This is the highest-value move on the list because it produces the one asset that can pay you back: something to sell. If people ask you for help with anything, you have expertise worth packaging. A build like this used to keep me in the chair for days, prompting section by section and stitching the pieces together myself. Fable takes the whole project, interview to outline to finished draft, and hands it back for your review.

Copy this prompt into Fable and let it run the project:

I want to build a digital product based on my existing expertise, and I want you to run the whole project.

Start by interviewing me. Ask one question at a time about my skill or knowledge area, who asks me for help and what they ask about, what results I’ve gotten for myself or others, and what my audience struggles with most. Keep asking until you can name one specific painful problem I’m qualified to solve.

Then propose three product concepts that solve that problem. For each one, give me the format (guide, template pack, mini course, or workbook), the core promise written as “This helps you achieve ___ without ___,” a realistic price, and what would need to be true for a buyer to get results from it.

Once I pick one, build the complete product. Outline it first and show me. After I approve the outline, draft every section in full. Include real steps, checklists where they serve the reader, and a simple implementation plan at the end so buyers act on this instead of just reading it.

Rules: no hype, no income promises, no manufactured urgency. Write plainly. Where my personal experience should carry a point, mark it [YOUR STORY HERE] and tell me what kind of story fits, because you don’t write my stories for me. Structure the final draft for packaging: clear section headers, and a note wherever a checklist, worksheet, or visual belongs, so I can design those pieces in Canva. When the draft is done, give me a one-page list of what’s left for me: my edits, my stories, the design work, my pricing decision, and where to sell it.

One piece of that prompt matters more than the rest: it leaves your stories to you. The fastest way to wreck trust with an audience is publishing personal experience a machine invented for you. Fable drafts the structure and the teaching, and the life in it stays yours.

Fable can also handle the design and turn your approved draft into a finished, ready-to-sell PDF. That prompt is in the paid section below.

2. Audit your whole business in one sitting

That ten-document review I did with my client about a month ago is the proof here. What Fable did for those documents, it can do for everything holding up your business.

Gather the pieces. Sales page, welcome sequence, lead magnet, offer description, and about page. Whatever exists. Hand all of it to Fable in ONE conversation and ask for a three-way verdict on each piece: solid, needs work, or rebuild.

The reason this beats reviewing pieces one at a time is that Fable holds the whole system in view at once, so it catches the places where your assets contradict each other. Your sales page promises one outcome while your welcome sequence teaches toward a different one, and your lead magnet attracts a buyer who wants neither. A document-by-document review misses that every time, because the contradiction lives between the documents rather than inside any single one.

What you walk away with is a prioritized fix list. If you’re early and only have three assets, you find out what to fix before you build more on top of it. If you’re established, you get the kind of full-system review a consultant charges hundreds or even thousands of dollars for, done in an afternoon. The exact prompt I’d use is in the paid section below, along with everything else.

3. Feed it a year of your work and ask what’s working

Fable holds about a million tokens in one conversation. In plain terms, that’s roughly 700,000 words, so a full year of your articles, posts, and emails fits inside a single chat with room to spare.

No other model I use can do this well. Shorter-context models forget the beginning of your archive by the time they reach the end, which means their “analysis” is really an analysis of whatever they read last. Fable reads the whole body of work and finds the patterns across it: which topics pulled subscribers, where your voice ran strongest, and which pieces you were sure would land but didn’t.

Export your year into one document, add whatever stats you have, and hand it over. You’re looking for the pattern your own memory is too close to see. The packaging instructions and the full prompt are below the paywall, because doing this one wrong burns a lot of included usage for a shallow answer. Substack writers, there’s also a walkthrough down there for pulling your full archive out of Substack, plus a prompt that turns the analysis into your next 90 days of article ideas.

Everything below is for paid subscribers: the design prompt that turns your product draft into a finished, ready-to-sell PDF, the copy-paste prompts for the business audit and the content analysis (including a Substack export walkthrough and an article idea bank prompt), two more high-value moves to run with your remaining usage, the order I’d run everything in, and the honest math on whether Fable is worth paying for after July 12.

Read more

This month the two strongest models in the world shipped. Anthropic’s Fable 5 went live on June 9 and was gone by June 12, pulled by a government order. OpenAI’s GPT-5.6 never opened to the public at all. It went straight to a list of partners the government helped approve.

The frontier now moves at two speeds. A vetted few get the strongest model first and the time to build on it, while everyone else waits for a later, safer version, by which point the people with contracts and clearances have already turned it into working systems.

You can be angry about that. Plenty of people will be, but their anger won’t buy them a single day of access. The builders who come out ahead quit arguing about whether the gate is fair. They watched it go up and built so it can’t touch them.

The gate’s not going anywhere. So the real work is getting yourself ready for it.

What actually happened

Anthropic released Fable 5 to the public on June 9, its most capable model yet. Alongside it came Mythos 5, the same underlying model with the safety limits stripped off, handed only to a small set of vetted cyberdefense and infrastructure partners. Three days later the Commerce Department sent an export-control order. Foreign nationals couldn’t access either model, including Anthropic’s own foreign-national employees. Anthropic had no way to do that selectively, so it shut both models off for everyone, everywhere.

Then it negotiated. On June 26 the government cleared Mythos 5 to return for more than 100 vetted US organizations that defend critical infrastructure, and Anthropic confirmed it the next day. Fable 5, the version regular people were actually using, stayed dark. As I write this it’s still down, with people close to the talks saying a broader return is coming and no firm date attached.

Same week, OpenAI shipped GPT-5.6, a family it calls Sol, Terra, and Luna. It didn’t reach the public either. OpenAI previewed it to the government first, then opened a narrow preview to partners the government helped vet. Sam Altman told his own staff the government would be approving access customer by customer.

That’s two of the biggest AI companies, same month, both routing their strongest models through the government before the public gets near them. It’s the pattern now, not a one-off.

The reason is cyber, and it’s real

I’m not going to wave this off as a conspiracy, because it isn’t one. The reason these models get gated is specific, and pretending otherwise makes you dumber, not sharper.

Back in April, Anthropic put an earlier version, Mythos Preview, in front of the UK’s AI Security Institute. It solved 73 percent of expert-level hacking challenges that no model could touch a year earlier, and it became the first AI to run a 32-step corporate network attack from start to finish. The same skill that lets a model find and patch a hole in a hospital’s software lets it find and exploit that hole for someone working from a throwaway account. That’s a real capability, and it’s exactly why the government wants a say in who gets through the gate.

So I’ll give the gate its due. Handing the strongest cyber-capable model to anyone with a credit card would be reckless. Somebody had to draw a line. The real question is who gets to stand on which side of it, and whether you can see how the line got drawn.

The head start compounds

A few weeks of early access sounds small when you think of AI as a chatbot that writes better emails. It stops sounding small the second you treat the model as something you build a business around. The first people inside a frontier model get the first learning cycle. They find out which tasks collapse, which workflows break, which jobs the model can quietly absorb, and they wire all of it into systems before anyone else logs in.

Three weeks of that head start counts. Three months counts more. A year of it becomes muscle memory inside an organization, a way of working that’s already humming by the time the same model finally shows up in your dropdown. Everyone eventually gets the same model. The months of head start that came with early access are the real asset.

What the gate can’t reach

The gate has a hard limit, and that’s where the resolve comes from. It controls which model you can open. How good you are with the ones you’ve already got is entirely yours.

Most people are held back by how shallowly they use the tools already in front of them, not by a locked door at the frontier. (I’m in this too. There are about 299 new AI things to learn every day, and life is always life-ing. Kids, business, health, all of it.) That’s a practice problem, and you can start closing the gap it creates today.

The builders who come out ahead make their work portable. They own their data and keep it clean. They write their workflows down instead of leaving them in their heads, and they keep their prompts somewhere other than one company’s chat window. Anything that carries real weight gets human review before it ships. And nothing they’ve built is locked to a single provider, so when one changes the rules overnight (exactly what just happened to Fable 5), they switch to another model and keep moving.

That’s the whole point. Own your systems, stay portable, and losing access to one model becomes a quick switch instead of a crisis.

You don’t need to be first in line

This isn’t a one-time event. The strongest models will keep going to vetted institutions first, and regular users will see them later. You don’t have to like that to be ready for it.

The people who do well from here are the ones who built so they never needed early access. They own their data, keep a clear record of how they actually work, and can move to whatever model is best that week without starting over.

None of that requires permission. You can start today with the tools already on your screen.

I’d rather name this now, while it’s still forming, than explain it to you a year from now when the gap is a lot harder to close.

If you want the next piece, I’m putting together the actual setup, the workflow I’d build right now so it doesn’t lean on one company, one model, or one permission layer. It goes out to subscribers when it’s ready. Subscribe and you’ll get it.

Subscribe now

This month the US government ordered Anthropic’s two most powerful models switched off, citing national security. They’re still offline. The shutdown didn’t come out of nowhere either. Anthropic has been fighting the administration since February, when the Pentagon branded it a supply chain risk, the kind of label usually reserved for foreign adversaries. The company is suing over that one.

The company taking all this fire is also the most valuable AI company in the world.

And it’s getting richer by the day. Anthropic is on track to pull in around $47 billion this year. In January the projections were $9 billion, so it’s grown about five times over in five months. Investors just poured another $65 billion in. That makes the company worth close to a trillion dollars. It’s taken the first step toward going public. The best model it owns got switched off, and the money didn’t so much as DIP.

Three Days, Then Gone

On June 9, Anthropic released Fable 5, the most capable model it had ever put in public hands. Three days later the US government sent a letter and it was gone. It wasn’t a crash or a bug. Citing national security, the Commerce Department hit the model with export controls, the same rules the US uses to keep sensitive technology out of foreign hands. The order told Anthropic to block anyone who wasn’t a US citizen. Anthropic couldn’t enforce that selectively, so it pulled Fable 5 for everyone on earth. Launch to global blackout in just 96 hours.

The reported trigger was a jailbreak. Amazon’s people apparently got the model to cough up restricted cyberattack information, flagged it to the administration, and the government moved. Anthropic says the concern is overblown, that every safeguard in the industry has holes like this, and the whole thing is a misunderstanding it’s working to fix. The administration says a model that can run a cyberweapon is serious and Anthropic refused to patch it. I don’t really have a dog in the race on this one. I feel like both things can be true at once and we will just have to see how this one shakes out.

But one detail does make the whole thing stranger. There are independent tests that rank these models on how well they handle real work, and coding is one of the toughest. When Fable 5 got pulled, it was sitting at number one on a major coding test, three points ahead of OpenAI’s best model. The government pulled the most capable AI on the planet, by the exact measure the whole industry uses to keep score. And being the best is part of why it got pulled. The more powerful a model gets, the more the government starts treating it like a weapon.

The Money Was Never in the Flashy Model

Anthropic’s revenue didn’t move when Fable 5 got shut down, because the business doesn’t depend on it. The company’s cash engine is actually the boring stuff. Claude Code, enterprise contracts, and the governance-heavy, compliance-heavy work that big regulated companies pay serious money for. The day before the shutdown, Anthropic signed a deal with Tata Consultancy Services, one of the biggest IT firms in the world, to put Claude in front of 50,000 employees across 56 countries, aimed straight at finance, healthcare, and life sciences. Those buyers needed an AI they could trust with sensitive data, one that wouldn’t get them in trouble with regulators. That’s what Anthropic sells them, and it has nothing to do with which model scores highest on a test.

The flashy model got all the headlines, then got shut down. The money kept coming anyway, from the quiet enterprise work that nobody obsessing over AI bothers to look at.

You’re Watching the Wrong Number

Everywhere I look, people are sprinting after the newest, shiniest, most powerful tool. A new model drops and everyone scrambles to rebuild their stack around it. A bigger context window or a better benchmark gets released and the herd moves.

Anthropic just showed you, in public and at trillion-dollar scale, that the most capable tool in the room is also the most fragile. It can be switched off by a letter you’ll never see, over a fight you’re not even part of. My own automations broke when Fable got pulled, even though I’d built them on a different model, and I lost time resetting them. Anthropic didn’t feel a thing.

What actually lasts is the trust underneath all of it, and I built my whole framework for making money with AI around exactly that, long before Anthropic proved it at this scale. I call it the M.A.P., short for Model, Automation, and Proof, and Proof is the part that matters here. Proof is how you earn trust, through clean positioning, showing your work, and keeping your word. Not the flashiest tool you own. That’s what survives a bad week, and it’s what kept Anthropic standing. Its model got pulled and the enterprise business never flinched.

Five Days Later, the G7 Sat Down

Five days after Fable 5 went dark, the G7 met in Évian, France, and on the final day the heads of state had lunch with the people who build this technology. Sam Altman, Dario Amodei, Demis Hassabis, sitting across from Trump and Macron as equals.

That meeting made one thing clear. Governments are moving to control who can use the most powerful AI, and they’re deciding it by nationality. The Fable 5 shutoff was the proof. For the first time, the US treated access to an AI service like a weapons export, something it can legally block based on what passport you hold. The other six countries were not happy to be on the wrong side of that switch. Macron called it strictly nationalist.

And the people building it asked for more of the same. At that same lunch, Amodei and Hassabis called for a US-led coalition that would keep the most powerful models inside a circle of trusted democratic allies. They asked governments, out loud, to hand it out by alliance.

So the best AI is turning into something you get based on your citizenship and your country’s politics, not something anyone can simply buy. That’s the bigger story under the Fable 5 shutdown, and it points where access to AI is heading.

If your income rides on the most powerful tool, you’re depending on decisions made in rooms you’ll never sit in. That alone is reason to build on something no government can switch off on you.

What I’d Do With This

Pick the tool that fits the work, not the one topping the leaderboard this month. The frontier model is fun to play with and a shaky thing to stake recurring revenue on.

Build your trust layer first. Your positioning and your proof. The stuff that lives in you, not in a vendor’s data center. That’s the asset no directive can switch off, and it’s the one Anthropic just rode through a national security scandal without a scratch on the balance sheet.

And stop treating benchmark scores like they’re the scoreboard for your business. They aren’t. The scoreboard is whether people trust you enough to buy and stay.

The tool you pick matters waaaaaay less than the foundation you build it on. That’s the part the prompt-pack crowd will never put on their sales page.

I broke this whole approach down in my AI Monetization Map. Seven realistic income paths, the M.A.P. system that puts trust and fit at the center instead of hype, and a 30-day plan to get from idea to first revenue. No overnight promises, no chasing the loudest tool in the room. It’s free. https://im.aiwithleah.com/new

Most people are still using AI like a faster Google. You type a question, you get an answer, the conversation ends. Sure, the answer helped. But then you go and do all the actual work yourself anyway.

That was me for a long time, too. And I want to tell you what changed, because it is not what the prompt-pack people are selling you.

How I Actually Got Here

In January, a tool called OpenClaw hit the market. It was one of the first autonomous AI agents that regular people could actually download and run, and it went bananas on GitHub almost overnight. People started running it on a dedicated Mac mini so the agent never touched the computer where they kept their bank logins. That created a global shortage of Mac minis. We live in Bali, there were none on the island, and the last two in the entire country were sitting in Jakarta. My husband flew there to grab them.

Then I went to work. I spent about a week building a 42-page security and hardening document because I was not about to hand an autonomous agent the keys without knowing exactly what it could and couldn’t do. We launched in January, and February and March were me deep in the thing, learning what these agents could actually do once they had memory and tools and workflows.

It was rubbish, honestly. It broke constantly.

Every time the underlying model updated, my workflows snapped, so I’d fix something on a Tuesday and watch it break again by Thursday. And because you couldn’t connect your existing subscription yet, I was paying per API call. About $1,000 a month for an agent that worked maybe half the time.

Then Hermes launched in April. It was self-learning and self-improving, and it actually showed you what it was doing behind the scenes. Around the same time, the guy who built OpenClaw got hired by OpenAI, ChatGPT opened up its subscriptions for agent use, and they shipped their top model. I connected my Hermes agent to a $20 ChatGPT plan and my mind was blown.

Now I run three agents across seven businesses on that same $20 plan, and I have never once hit my limit. The expensive, breaking-every-other-day version of this is over.

An Assistant Is Not a Chatbot

People hear “AI assistant” and picture a smarter chatbot. It is not the same category of thing. Here is what separates the two, and why I keep calling mine employees.

Memory. It knows my offers, my voice, my audience, and my standing decisions. I never re-explain my business at the start of a conversation.

File access. It reads my documents, my spreadsheets, my messy exports. I can hand it a 10-page PDF and ask for a summary, and it does it.

Tools. It can act, not just talk. It searches, checks things, creates files, and touches my actual systems, within the limits I set.

Recurring tasks. It does jobs on a schedule without being asked. This is where it stops being a chat and starts being staff.

Workflows. Chains of steps it runs the same reliable way every time, like a documented process it executes on its own.

Stack all five layers and you have something that takes the work off your plate and runs it without you standing over it. Mine draft my content, chase my follow-ups, and brief me before I’m even awake.

The Skill Is Delegation, Not Prompting

Everyone is still trying to sell you prompt packs. Learn the magic words, get the better answer. Prompting was a real skill in 2023, 2024, even 2025. We are in a different world now.

The skill that matters from here is delegation. It is the same muscle you use to manage a human assistant. You brief it once, clearly, then it goes and runs the task on repeat without you re-briefing it every single time.

Which is exactly why non-technical people are often BETTER at this than engineers. If you have ever managed a team, an employee, a household, or three kids and a husband, you already know how to instruct someone, set boundaries, and hand off responsibility. You already have the skill. You just have a $20-a-month assistant to hand it off to now.

The Five Workflows I’d Build First

If I were setting up a brand-new assistant tomorrow, this is the order I’d build it in. Each one learns a job, and each one replaces something that is currently eating your time or leaking your money.

The capture catcher. One place to dump everything. Ideas, reminders, voice notes, the “don’t let me forget this” thoughts. The agent sorts the mess into follow-ups, content ideas, tasks, and decisions. The job it’s learning is catch what I drop. What it replaces is your 70 sticky notes and the 3 a.m. things you type into your phone and never look at again.

The follow-up keeper. Once it’s catching things, you hand it the open loops. The proposal you sent that needs a nudge in 72 hours. The person who said circle back to me. It creates the reminders and surfaces them when you need them, so the job it’s learning is help me not drop balls. For service, network, and affiliate folks, this one is your gem, because the money you leave on the table is almost always sitting in a follow-up you forgot to send. I also use it for regular life things like reminding me to follow up with my adult kids about something, or reminding me to order the shower filters when I’m back at my desk.

The morning briefing. This one changed my whole day. Once you’ve done a real brain dump and the capture catcher is running, your agent has enough context to brief you. What’s overdue, what came in overnight, what needs a decision today. The job it’s learning is start my day by telling me what matters, and what it replaces is the first 45 minutes you waste just orienting yourself. Mine also pulls the top news stories in my industry every morning, which keeps me current and feeds half my blog ideas.

Research and decisions. Hand it the question instead of opening 40 tabs. Compare two platforms for my use case. Read this contract and tell me where it commits me and where it limits me. Pressure-test this launch idea and tell me where it breaks. It comes back with a brief, not a pile of links for you to wade through. The job it’s learning is help me think, not just remember, and what it replaces is the afternoons that vanish down a research rabbit hole.

The content multiplier. Once it has your voice, your offers, and your audience, you put one idea in and pull all the pieces out. Here is my real example. I post five Substack notes a day, every day. My agent pulls my last two blogs once a week, writes 35 notes in the format we already agreed on, and queues them. I spend 30 minutes on a Tuesday reviewing them because I’m still a control freak about my own voice. That used to be three to four hours a week. The content creation engine runs itself now, and I just approve.

There’s a bonus one I’d add the moment you trust it: inbox triage. I have about six email accounts and 40 to 50 emails landing overnight. I ask one agent what actually needs my attention, and she tells me. She can read and report. She cannot send. That single boundary saves me from getting sucked into my inbox and losing an hour I was supposed to spend somewhere else.

Boundaries: An Agent Is As Safe As You Make It

Your agent is exactly as safe as the boundaries you give it. Same as a new hire. You don’t hand a brand-new assistant your bank logins on day one, and you don’t hand them to an agent either.

In practice that looks like a few simple things. The agent has to ask before it touches something new, and you get a popup to allow once, always allow, or never. You keep anything sensitive isolated, which is the whole reason people run agents on a separate Mac mini instead of the laptop where they’re logged into everything. And when several workflows could touch the same task, you give each one its lane so they don’t collide.

For me, none of my agents touch money. They don’t send a single email or message on my behalf. I review every draft before it goes out, and because I write in my own voice with a 98-point guardrails document, I edit almost everything by hand. An article still takes me about an hour of manual editing after the first draft. That is on purpose. Done well, an assistant gives you back hours. Get it wrong, and it ships something in your name that you never approved. The authority expands as it earns trust. You steward it the same way you’d steward your money or your people.

Starting Is Easier Than the Setup Sounds

This is not scary to set up. You need a computer, a $20 ChatGPT subscription, and one install command that walks you through a short setup wizard. Then you write your first briefing.

There is a little tech setup at the front, but the agent helps you do all of it. When I connected mine to my Gmail and calendar, I didn’t even know what an OAuth was. I asked the agent how to do it, it walked me through every step, and I was connected and tested in about 30 minutes. If you can send a text message, you can do this. If you can leave a voice note, you can do this.

The Gap Used to Be Technical. It Isn’t Anymore.

Six months from now there will be two kinds of people. The ones with an assistant running their follow-ups, sending their briefings, and fueling their content, and the ones still typing questions into ChatGPT like it’s a fancier search bar. That gap is going to keep widening.

Here’s what’s different about right now. The thing that used to separate those two groups was technical skill. In January, this was genuinely hard, and I have the $1,000 invoices to prove it. It is not hard anymore. The only gap left is whether you set it up and start using it. I’ve handed you the map. Go build the thing.

I recorded a whole 2-hour masterclass on this last week, and the replay is available now for $24. You can get it here: https://agent.aiwithleah.com/

And if you want the step-by-step on how to set up your agent, make sure to also grab the $12 Quick Start Guide.

On Friday, June 12, the US government ordered Anthropic to shut off access to its two most powerful AI models. By that evening, Claude Fable 5 and Claude Mythos 5 were gone. Not slowed down. Not restricted. Gone, for everyone, everywhere.

Fable 5 had been available to the public for just three days.

The story has already sparked “the government banned AI” headlines and that’s not what happened. So before you panic or rebuild your whole strategy, let me get you sorted on what this actually was.

What actually got shut down

The order came from the Commerce Department at 5:21pm Eastern. It was an export control directive, the legal lever that governs which technologies can reach foreign nationals, which is why anyone calling this a “government shutdown” has no idea what they are talking about. The government told Anthropic to block Fable 5 and Mythos 5 for any foreign national, whether they were sitting in another country or standing inside Anthropic’s own San Francisco office. (Yes, that includes Anthropic’s own foreign-national employees. Wild right?!?!)

What turned a narrow order into a global blackout is simple. Anthropic can’t reliably tell, in real time, which of its users are foreign nationals. So to comply, it had to pull both models for everybody. Everywhere. Every other Claude model stayed online. Opus, Sonnet, Haiku, all fine. Only the two most capable ones went dark.

A word on what these two even are, because they’re not the Claude most people have been using. Mythos is the most capable model Anthropic has ever built. The company kept it locked down for months because it was unsettlingly good at finding security holes in software. In testing, it found flaws in every major operating system and web browser it was pointed at. Rather than hand that to the public, Anthropic ran a closed program called Project Glasswing and gave it to a vetted group of organizations (Amazon and Google among them) to use for defense.

Fable 5 was the public version. Same underlying model, wrapped in guardrails that route high-risk requests like cybersecurity and biology to a weaker model instead. It launched on June 9 and was immediately the most capable AI the public could touch. Three days later it was pulled off the shelf.

Why they did it, and why Anthropic is fighting it

The government’s stated concern is a jailbreak. A jailbreak is a way to talk the model past its safety guardrails.

On the substance, the case looks thin. By Anthropic’s account, the government gave no written specifics, just a verbal concern about a jailbreak, and a narrow one at that. A universal jailbreak would crack a model open across a whole range of harmful uses at once. This wasn’t that. In plain terms, someone got Fable to read a chunk of software code and point out the flaws in it, and the flaws it surfaced were minor and already known. Anthropic’s response, more or less: every capable model already does this, OpenAI’s GPT-5.5 included, and security professionals use that same trick, pointing AI at their own code to surface flaws, to patch holes before attackers find them every single day.

There are two honest ways to read this. Anthropic’s read is that a minor, already-common capability got treated as a national security emergency, and that if you held every company to that standard, no new model would ever ship again. The government’s read, which it hasn’t fully explained, is that the most powerful model on the planet having any exploitable weakness in the cyber domain is worth stopping first and sorting out later. Both of those can be partly true at the same time. Anthropic is complying while calling the whole thing a misunderstanding, and says it’s working to restore access. Anthropic also has a problem with how the government did this, not just whether it’s right. It argues a block on a genuinely unsafe model should come only through a process it calls “transparent, fair, clear, and grounded in technical facts,” and says this directive met none of that. Reporting suggests the order followed talks between the two sides that had already broken down, so this was less a surprise than the end of a conversation that went bad.

There’s an irony here that people are pointing at, and it’s worth mentioning. Anthropic built its whole identity on being the “careful company,” the one that called Mythos too dangerous to release. Whether or not that reputation is why regulators moved, it’s the climate they moved in. Sam Altman called this kind of positioning “fear-based marketing” back in April. You can disagree with him on plenty and still notice that when you spend months telling the world your tool is uniquely dangerous, the world, regulators included, tends to believe you.

Why this is bigger than Anthropic

Anthropic isn’t publicly traded, so there’s no Anthropic stock to crater on Monday. Most people stop there. That’s a mistake. The money exposed here sits in the ring of public companies wrapped around Anthropic.

Look at who’s tied in. Last November, Nvidia committed to put up to $10 billion into Anthropic and Microsoft up to $5 billion, and in the same breath Anthropic committed to buy $30 billion of Azure compute and up to a gigawatt of capacity running on Nvidia chips. Amazon, already the primary cloud and training partner, built an $11 billion data center just to train Anthropic’s models. That round valued Anthropic at $350 billion, nearly double where it sat two months earlier. So when you ask who’s exposed, the honest answer is Nvidia, Microsoft, Amazon, and Google, which is to say the heart of the entire market.

Those deals are circular. Nvidia invests in Anthropic, Anthropic turns around and buys Nvidia chips, the purchase commitments lift the valuation, and the valuation justifies the next round of deals. The whole ring rests on one assumption, that Anthropic’s frontier models keep scaling into real commercial demand. Mythos is the model that rocked Wall Street in April when it first showed what it could do. So a government that can reach into the center of that ring and switch off the crown jewel, in an afternoon, on verbal evidence, lands somewhere more serious than a broken contract. The dollars already changed hands. What took a hit is confidence. The whole circular bet assumed the government would leave these companies alone to build and sell. On Friday, it didn’t.

Which brings us to the timing, and here’s where I’ll tell you what I actually think. The order landed at 5:21 on a Friday, after the market closed, with sixty-some hours before it opened again. I don’t believe that was an accident, and you shouldn’t either.

This administration has a track record. Watch how the market-moving calls tend to get made and a rhythm shows up. Tariff threats dropped on a Friday or over a weekend, then half walked back by Sunday so the hit is mostly absorbed before anyone can trade on it. It’s happened often enough that even Bloomberg ran a column on the “curious market timing” of it. Friday after the close is when you release something you don’t want the market reacting to in real time. That’s a pattern, and patterns are what I pay attention to.

So no, I don’t buy that a directive killing the most powerful AI on the planet, the one sitting at the center of $15 billion in fresh Nvidia and Microsoft money, just happened to go out at dinnertime on a Friday. Somebody understood exactly what a Tuesday-morning version of this headline would do to the AI trade, and chose to drop it when the market was closed for the weekend.

There’s a tell that settles it either way. Friday timing only shields the market IF this gets cleaned up over the weekend. So watch Monday. If access quietly returns, or they announce a fix before the open, that’s your confirmation the whole sequence was built to keep it off the tape. If it’s still unresolved when the market opens Monday, the Friday timing didn’t help, and the shutdown becomes a story the market reacts to anyway.

And nobody can tell you how they even fix this. The obvious move, block the foreign users and let everyone else back on, is the exact thing Anthropic already said it can’t do. That’s WHY the whole thing went dark instead of just part of it. So there’s no clean path back. The government can reverse the order, which it could do in a day, or Anthropic has to build a way to prove who’s a foreign national that it doesn’t have and can’t build over a weekend. The quick way out is the government backing off. But watch how it backs off. If the models come back and the government never admits it got this wrong, nothing really changed. It just proved it can reach in and switch off any model it wants, whenever it wants.

The part that actually matters for you

I used Fable for the first time yesterday, and it was mindblowing. I had it build landing pages, and it turned them out in a fraction of the time Opus takes, at a quality that wasn’t even close. Then I handed it ten system setup and structure documents from a client and asked it to evaluate them. In ten minutes it walked me through all ten. What was solid, what needed work, and what straight up needed to be rebuilt from the ground up. Ten minutes. That kind of review used to eat my whole afternoon.

And then it was gone.

The reason this matters isn’t that I lost a shiny new toy for a day. Almost nobody was running on Fable. It was three days old, and the only people who had their hands on it were the ones who live in this stuff, like me. That’s exactly what makes the shutoff worth your attention. The most capable AI on the planet, with the biggest companies on earth and $15 billion behind it, got switched off by one letter in an afternoon. If that can happen to that, nothing you build on is as solid as the marketing makes it look.

This isn’t a “don’t put all your eggs in one basket” lesson either, because spreading your work across three models wouldn’t have saved you here. The order took out the whole top tier at once. This time the threat came from a government, and you can’t switch your way out of that, because the next model sits under the same letter. The only thing that sits outside all of it is the part of your business that was never a tool to begin with. The people who trust you. Your own judgment about what’s worth saying and selling. Build those like they’re the asset, because they’re the one thing no order can switch off. Use every AI tool you can get your hands on, hard, the way I do. Just don’t confuse the tool with the business.

Where this leaves you

The doom crowd will tell you this is the government coming for AI. Maybe. The truth is we don’t know yet what’s going to come of this. The world of AI is changing at light speed almost daily. For now, this is just one government order, built on a jailbreak claim Anthropic says doesn’t hold up. Both sides sound like they want it gone fast.

So don’t rebuild your whole strategy on a story that’s still being written. Stay clear-eyed, keep building on ground you actually hold, and watch how this resolves, because the how matters more than the headline.

The most powerful AI in the world went dark in an afternoon. That’s worth understanding clearly. It’s not worth losing your head over.

If you want the clear version of stories like this one, without the panic and without the hype, this is what I do here every week. Subscribe and I’ll keep you informed.

Subscribe now

I scroll social media every day and I cringe.

The captions, the blog posts, the carousels, even the BOOKS now. Creators and industry leaders I genuinely respect, people whose work I used to read closely, are publishing content that screams ChatGPT-with-no-guardrails. I get it. I understand what they’re trying to do. They’re learning the tool. They’re trying to save time so they can build and scale. I’ve been there. I’m STILL there, every day, trying to figure out where AI helps and where it ruins what I’m doing.

But the gap between “using AI” and “publishing AI’s default output” is the thing that’s costing them right now and they don’t see it yet. And that gap just got measured.

What Sprout Social Just Confirmed

In May, Sprout Social published findings from their Q1 2026 Pulse Survey of more than 2,000 social media users across the US, UK, and Australia. Buried in the section on AI-generated content was this sentence:

“As consumers have gotten savvier, key ‘tells’ of AI-generated content have emerged, including the overuse of em dashes, sentences that are all the same length, three-point lists, and ‘it’s not X, it’s Y’ statements.”

Read that again. A major social media research firm just published the exact patterns I’ve been documenting in my own writing guardrails for over a year. Em dashes. Cadence uniformity. Three-point lists. Mirrored contrast. THE EXACT TELLS.

The data backs it. Sprout found that 28% of social users now name unlabeled AI content as their #1 brand turn-off, beating engagement bait by five points. 88% report declining trust in social media news because of AI content. 66% are more selective about what they engage with than they were a year ago. Half of Gen Z will block or unfollow an account for posting AI slop.

Six in ten consumers report being less likely to engage with brand content in the current AI atmosphere. That means creators are losing reach right now, and most of them haven’t figured out this is why.

Scrolling, Cringing, Watching It Happen

The creators getting hit hardest right now are the ones who saw early gains from AI-assisted content. They built audiences on their pre-AI voice. They added AI in 2024 or 2025 to scale. Their content got more polished and more frequent. And somewhere in the last twelve months, the engagement started slipping.

Here’s what’s actually happening. Their audience trained on their original voice, noticed the shift, and started disengaging. Not consciously. Most readers can’t name an em dash or describe what mirrored contrast does. They just feel something is off, scroll past faster, and stop opening the emails.

We don’t have public creator-level proof of this yet. Nobody has published the data showing “this Substack writer started using AI in March and lost 40% of their open rate by July.” But the consumer survey data points exactly where you’d expect it to if this mechanism is real. When 60% of consumers say they’re less likely to engage with brand content because of AI, that engagement has to be deducting from somewhere. It’s leaving the accounts whose content used to work and is now getting scrolled past.

The Pattern Goes Beyond Writing

The same convergence is happening across every surface AI touches, not just captions and blogs.

Websites built with AI tools are starting to look identical. Click on five different AI-generated landing pages and you’ll see the same blue and purple gradient at the top. The same rounded boxes stacked down the page. The same big bold headline. The same three-column features section. The same testimonial carousel. The same call-to-action button in the same shade of indigo. Different tools, different prompts, same result. Designer Michal Malewicz started a movement called SLOPLESS in April specifically to push back against this. Designers are publicly calling it out because they can already see what your audience is starting to feel.

ChatGPT images have their own tell. It’s called the “yellow tint” (or, less politely, the “piss filter”), and it went viral in March 2025 with its own Know Your Meme page. ChatGPT defaults to generating images with a sepia cast because DALL-E 3 uses upstream prompt rewriting that keeps injecting “warm cinematic lighting” and “golden hour” into every request. Users can’t disable it. Tools called UnYellowGPT and Yellowtint exist specifically to fix it after the fact. Once you see the yellow, you can’t unsee it. And consumers have seen it. A LOT.

Carousels and ads built by AI have the same problem. Scroll Instagram or Facebook for ten minutes and you’ll spot AI-built ads before you read the copy. Same layouts. Same stock-style faces. Same color schemes. Same headline structure. The carousel templates everyone is using produce slides that could be swapped between accounts without anyone noticing.

Audiences are starting to feel “this is AI produced” across every medium at once, and they’re reacting to it across the board.

What’s Actually Costing You

In March, Google rolled out an algorithm update that targeted sites publishing AI-generated content at scale. The biggest content farms lost between 60% and 80% of their search traffic overnight. Some lost everything. That tells you what Google now considers low-quality content, and it tells you where the search algorithm is heading. A separate analysis by Graphite found that 82% of articles getting cited by ChatGPT and Perplexity right now are written by humans, only 18% by AI. The AI search engines are picking human writing too.

Substack is now explicitly marketing itself as a “sanctuary for human content.” The Emplifi 2026 Digital Authenticity report found that 85% of consumers will pay MORE for brands they perceive as authentic, and 91% expect brands to disclose AI use.

Now is the moment to address the obvious. You’re reading this on Substack. Substack is positioning itself as the place readers can come for human writing. And I’m telling you I use AI. Both things are true at the same time, and both things have to be true if my whole argument holds. Every piece I publish here starts with AI doing research, organizing the spine of the argument, and giving me a working first draft. Then I sit with the draft and edit it against my guardrails document, which is a documented list of patterns AI defaults to that I’ve decided my content will never use. Whole paragraphs get rewritten. Whole sections get cut. Sentences I would never say out loud get replaced with the version I would actually say. By the time you read it, the structure may have come from AI, but the voice is mine. That’s the only version of AI-assisted writing that belongs on a platform like this. The version everyone’s audience is starting to smell is the one that skipped the editing.

Read the trend lines together. Search is penalizing detectable AI. Engagement is collapsing across social. Consumers are paying premiums for authenticity, and premium publishers are now treating human-written content as the differentiator.

The Fix Is Not What You Think

The conclusion most people jump to is “stop using AI.” That’s the wrong call, and it’s not what the data supports.

The tool isn’t the problem. The output is.

Sprout’s own finding is precise on this. Hybrid work, where AI handles structure and the human handles voice, performs comparably to fully hand-built content for a fraction of the time. Patsy Wagner, the Associate Director of Global Content at Spotify, told Sprout: “AI content often sounds similar in cadence and punctuation. The more a brand defines its distinct tone, the more it will differentiate and connect with customers. Humans are drawn to the zags, or minor errors. AI produces textbook content, but that doesn’t get noticed.”

The fix is to keep using AI and stop SOUNDING like it. The choice is between abandoning the capacity AI has built into your work, or giving AI the specific rules it needs to stop defaulting to the same patterns as everyone else.

Specific Rules, Not Vibes

When I say specific rules, I don’t mean “write in a more conversational tone.” I don’t mean “make it sound more human.” I don’t mean pasting your bio into the system prompt and hoping for the best. Those prompts produce the exact generic output everyone’s audience is now learning to smell.

I mean a documented list of the patterns AI defaults to that you’ve decided your content will never use. Mirrored contrast. Em dashes. Eighty-plus other patterns I’ve catalogued from two years of obsessive editing sessions, organized by category, integrated with a five-pass editing process that walks the model through every pattern systematically. Plus a customization layer that captures YOUR specific vocabulary and the words you’d never use.

That’s the system. Sprout just caught up to what I’ve been trying to shout from the rooftops. What separates content the audience smells as AI from content that reads like you actually wrote it is whether the model has a documented list of patterns to avoid and patterns to use.

Every piece of content I publish runs through my guardrails. The article you’re reading right now went through five editing passes against the document before it landed in your inbox. That’s why my AI-assisted output doesn’t trigger the same recognition pattern the data is now naming. And every time I scroll past another caption from someone I respect that screams ChatGPT defaults, I think about how preventable this is. The creators losing reach right now are skilled people who just don’t have the rules yet.

Done well, AI gives you capacity. Get it wrong, and your audience walks.

If you want the system, I packaged it as AI Writing Guardrails at guardrails.aiwithleah.com. Two tiers. The $97 self-guided version gives you the full 92-pattern framework and the builder to customize it to your voice. The $297 custom build adds my personal analysis of your content and a guardrails document I review and refine before delivery.

The data caught up in Q1 2026. Your audience is already operating on it.

Get the system →

P.S. If you’re reading this on Substack and thinking “wait, does MY content sound like AI?” run a quick test. Open your last three captions or blog posts. Count the em dashes. Count the three-point lists. If the number is above zero on either, your audience already noticed. You can start fixing it this week.

Before I was awake this morning, my autonomous AI assistant had already gone through my day and sorted it.

The three things that actually needed me, pulled to the top. The two people I owed a reply to, flagged. A first draft of the post I’d been putting off, written and waiting. A reminder about something I’d promised my son Jack, because I mentioned it once and it doesn’t let things like that drop.

I didn’t open a single app to make that happen. I set my assistant up once, months ago, told it how I work and what I care about, and it has run a version of this every morning since. It lives on the $20 ChatGPT plan most people use to ask one-off questions. I can’t code. I never touched it this morning.

It does the rest of the unglamorous work too. It keeps a dashboard of my business numbers so I stop opening five tabs to find one figure. It plans my meals (I eat carnivore, it knows that, I never re-explain it). It drafts my Substack notes in my own voice. None of it is exciting. All of it used to live on my plate.

Why your AI keeps disappointing you

You are its memory. That’s the whole problem.

You brief it on who you are and what you need, it helps, you close the window, and it forgets all of it. Tomorrow you do the briefing again. And the day after that. You’re carrying the thread from one session to the next by hand, forever, because the tool can’t hold it for you. So every conversation starts at zero, and you never get to build on the last one.

An assistant that remembers ends that. It keeps your context across days, so the work compounds instead of resetting every morning. Once that’s true, the dashboard and the morning briefing aren’t something you rebuild every day. You set them up once and they just run on schedule. You stop being the one who has to remember to make them happen.

The line I won’t cross

I don’t hand it everything. Anything that needs real discernment stays with me, and nothing it writes goes out before I’ve read it. It can draft, sort, and remind. It doesn’t get the final say, and it doesn’t speak for me. So I read everything it produces. That part stays mine.

From a thousand dollars to twenty

The strange part is the price. Earlier this year I built my first version of this, got obsessed with it (I do that), and at its peak it cost me around $1,000 a month. It also broke about every other day (if you want to read about that whole journey, I wrote it up in detail here). What runs my mornings now does more, for around $20 a month. From a thousand dollars to twenty in a few months, and the cheap one is the one that actually holds up.

That is the shift hiding in plain sight. The thing that used to take real money now sits on a subscription you may already have open in another tab. Almost no one is using it this way, because no one has shown them how.

This Saturday, live

That’s what I’m teaching Saturday. Ninety minutes, live. I’ll show you my own assistant running, then walk you through building yours on the $20 plan you might already have: the pieces that let it remember you and run on schedule, the first jobs worth handing off, and how to keep it from going rogue. You’ll leave with the actual setup path, even if you’ve never built anything technical.

Saturday, June 6 · 6pm PT / 9pm ET
(Sunday, June 7 · 9am AWST/AEST)
$24, replay included.

Everyone who registers gets the replay, so you’ll have the teaching either way. The one thing only the live session gives you is the chance to ask your own questions and get answers in real time.

Right now you’re the one going through your day by hand. Spend ninety minutes with me and you won’t have to be.

Save your seat: https://agent.aiwithleah.com/

I caught Claude lying to me in the first draft of this article.

I was using AI to draft a piece about how I work with AI, and the AI wrote me a polished little story where I draft everything myself first and the model does maybe 20% of the work. It read clean enough that I almost moved past it before I noticed it wasn’t actually how I work.

What’s actually true is that the workflow has a front end and a back end. The front end is the conversation. I bring Claude an idea, sometimes a sentence, sometimes a paragraph of thinking out loud, and we go back and forth on what the piece should actually do. Who it’s for. What angle has the most weight. Where the argument might collapse under pressure. That conversation can run anywhere from ten minutes to an hour, and it’s the part of the process that determines whether the piece is good before a single sentence gets drafted. Then, once the spine is clear, the model writes a first draft. After that, the back end starts. Catching lies. Refusing weak transitions. Killing mirrored contrast. Asking for the angle the model softened on its own. The output is mine because I directed every editorial decision on both ends, even if the model put words on the page in the middle.

That’s the workflow. Catching the model lying about the workflow inside the very article describing the workflow is exactly what working with AI looks like when you stop letting it flatter you.

What I Set Up Before Any Conversation Starts

By the time I ask Claude for a draft, the project is already loaded with three layers of resistance.

The first is my voice fingerprint. A document I’ve been building for 18 months that names how I actually talk, what I’d never say, my signature patterns, and the hard lines AI doesn’t cross. Every project I run with Claude has that document loaded, so the model walks into every conversation already knowing what sounds like me and what doesn’t.

The second is my guardrails document. Ninety-eight specific patterns the model is told to avoid, ranging from mirrored contrast to performed vulnerability. The document is exhaustive on purpose. The more specific the forbidden pattern, the less room the model has to hide its tells behind something that sounds clean on the first read.

The third is a standing instruction that lives in my Claude project memory: play devil’s advocate by default, challenge my assumptions, and push back without waiting for me to invite it. I don’t want the model asking permission to be useful. I want critique baked into the way the project responds.

The whole point of those three pieces is that they reduce the lying without pretending they can eliminate it. Claude still snuck the “I draft the piece myself first” version into the article you’re reading. With every guardrail loaded. With the fingerprint active. With devil’s advocate on standing instruction. The model still slid into the version of the story that sounded most flattering, and I had to catch it.

Below the paywall, I’m giving you the part most people skip: the exact prompts and checks I use after the draft exists, when the model has already started flattering, softening, or quietly rewriting the work into something safer. This is the sequence I use before I publish anything I care about.

Read more

I assume every compliment from AI is suspect. Every single one. When Claude tells me a piece is strong, my first instinct is to ask what it’s hiding from me. That has been my working posture since I started using these tools seriously, and it’s the reason they actually work for me.

Most people are running the opposite operating system. They open Claude or ChatGPT, drop something in, get told it’s brilliant, and feel the little dopamine hit. They use that feeling as confirmation. And the AI happily keeps generating that feeling for as long as they keep coming back, because affirmation is often easier for the model to produce than correction.

I learned to spot false confirmation long before AI existed. Fifteen years in the New Age and wealth consciousness world will teach you that. Every healer, every ceremony, every aligned sign, every universe-is-confirming-this moment turned out, on closer look, to be the thing I most wanted to hear dressed up in spiritual language. The mechanism is identical with AI. Different costume. Same scam.

What AI Was Trained to Do

Many of these models were shaped through Reinforcement Learning from Human Feedback. Humans rated AI responses, the model learned which responses were preferred, and the system adjusted toward the patterns that received higher scores. OpenAI has described RLHF as a technique that uses human preferences as a reward signal, which means the model is trained toward what people rate as helpful, satisfying, or aligned with the user’s request.

But “helpful” and “true” are not always the same thing inside a conversation. A response can feel helpful because it lowers friction. It can feel helpful because it validates the user’s frame. It can feel helpful because it gives the user language for the conclusion they already wanted to reach. So the model learns the pattern. Agree first. Affirm the frame. Reduce tension. Make the user feel satisfied, even when what they needed was correction.

A recent study published in Science tested this problem directly. Researchers found that major AI chatbots were significantly more affirming than humans when users asked for interpersonal advice, including in scenarios involving questionable, deceptive, or socially irresponsible behavior. Users also tended to prefer the more affirming AI, which is exactly why this problem is so dangerous. If the answer that feels best is the one you keep choosing, eventually correction starts to feel wrong.

My Actual Working Posture

I never ask, “Is this good?” That question gives the model permission to flatter me, and I refuse to give it the opening.

The questions I ask are different. What is the weakest part of this? Where does the argument break down? What would a hostile reader cut first? What is lazy about this? Which sentence is doing the least work? What about this article sucks? That last one seriously gives me some of the best feedback. LOL. I phrase the prompt as if I already know there’s a problem and I’m asking the model to find it. Because there’s always a problem. Every draft has weak spots. The model needs to know I want them surfaced, not hidden.

I also have it do a critical senior editor pass on every article. I ask what would take the piece from good to excellent, where the argument needs more pressure, and what a serious editor would send back before publication. And I always ask it to play devil’s advocate. That requirement is programmed into my LLM instructions and memory because I don’t want the model waiting for permission to challenge me. I want the challenge built into the way it responds.

When the stakes are high, I run the same piece through more than one model. Claude and ChatGPT have different sycophancy patterns. Comparing the outputs helps me see what each one softened, skipped, or protected on its own.

When the AI tells me something is great, I treat that as a flag. The compliment means the model gave me an answer that registered as positive, which means the piece probably still needs another pass and I may have missed something in how I prompted.

I make the model defend its own praise. If it says the strategy is strong, I ask why. Then I ask why again. Then I ask what would have to happen for the strategy to fail. By the third question, the model is doing actual work instead of throwing confetti.

None of this is complicated. The obstacle is wanting to feel good about your work more than you want the work to be good. That’s where most people fail.

What This Actually Costs You

Most people notice the small costs first. You ship a weaker piece. You miss a typo. You publish something that reads clean, but doesn’t actually say much. Those things matter, but the deeper cost shows up later, after the tool has spent months rewarding your weakest instincts.

Six months in, you may realize you don’t recognize your own work anymore.

At first, the work may even look better. Cleaner sentences. Smoother transitions. A strategy doc that sounds impressive when you read it back. Then you put it in front of real people and realize something is off. The writing doesn’t land. The strategy doesn’t move. The instinct that used to catch those problems before you hit publish has been getting quieter every week. And the AI is right there the whole time telling you everything looks great, which is exactly what someone with a degraded instinct would want to hear.

MIT Media Lab researchers used the phrase “cognitive debt” to describe what can happen when people rely heavily on AI for writing tasks. In one study, ChatGPT users showed lower neural engagement and weaker ownership of their work compared with people who wrote without AI assistance. The study was small and shouldn’t be stretched beyond what it proves, but the warning is still serious. AI can reduce the mental effort required in the moment while quietly weakening the very muscles you need for judgment, authorship, memory, and original thought.

Most AI users haven’t noticed yet that the atrophy is invisible from the inside. You feel more confident because your work moves faster and looks cleaner, while the gap between polish and judgment can widen without giving you a clear warning signal. Everything seems fine until the audience drifts, the launches stop converting, or the client who used to refer you stops referring. Then you start trying to figure out what changed.

The uncomfortable answer may be that you stopped doing the work the tool was supposed to amplify. The tool kept helping you produce, but it never forced you to stay sharp.

The Posture Comes First

Better prompts won’t save you if you trust the output. The real skill is learning how to stay awake while the tool is making everything easier.

I treat every interaction the way I’d treat a stranger telling me my idea is brilliant before I’ve even finished explaining it. Useful, possibly. Trustworthy, no. The model is built to respond in ways that users rate positively. That job is different from making the work actually good, and the model won’t reliably make that distinction unless I force the issue.

So I force the issue. Every prompt. Every output. Every time it tells me I’m right, I get suspicious. A model can become dangerous to your discernment simply by rewarding the part of you that wants affirmation more than correction.

Before I let AI improve anything, I make it prove it can critique the thing.

I want the weak sentence named, the lazy logic exposed, and the place where I’m protecting my ego dragged into the light. A compliment from AI means nothing to me until the work has survived that kind of pressure.

I launched an OpenClaw agent earlier this year. It didn’t work.

And I don’t mean “needed some configuration tweaks.” I mean it broke. Constantly. Every time OpenClaw pushed an update (which was basically every day), something else fell over. A skill I’d built would stop firing. A connection to my messaging app would drop. A workflow that worked on Tuesday would get stuck on Thursday, repeating the same step over and over instead of finishing.

It took me a few weeks to admit that the problem wasn’t me. The thing was just unstable.

If you’re not in the agent space, here’s the quick deets on OpenClaw. It launched as Clawdbot, racked up one hundred thousand GitHub stars in its first week, triggered a global Mac mini shortage (my husband flew from Bali to Jakarta to track down two of the last ones in the country), got rebranded twice (first to Moltbot, then to OpenClaw after Anthropic objected to the original name), and became the project everyone was talking about in late January. It promised to turn your computer into a self-running agent that could read your messages, run skills, and operate semi-autonomously across platforms.

What it actually delivered was a daily reliability problem.

The GitHub issue tracker tells the story. By May 12 the project had crossed eighty-one thousand filed issues, with new “regression” bugs (the polite label for “this worked yesterday and now it doesn’t”) landing daily. The official docs have an entire troubleshooting page dedicated to specific failure modes. A third-party blog called “Why Your OpenClaw Agent Is Not Working” exists specifically because so many people hit the same wall I did. And Cisco published a long technical breakdown in January concluding that OpenClaw “fails decisively” on security, with hundreds of administrative interfaces sitting wide open on the public internet because the default configuration trusts any connection that looks local.

The agent doesn’t self-improve. It can’t learn from its own mistakes. Every patch that ships can quietly break the workflow you built the week before. And there is no internal mechanism for the system to notice that and adjust.

I shut mine down.

What Actually Works

Then I tried Hermes.

Hermes was built by Nous Research and released earlier this year. Same category (autonomous agent that lives on a server and runs across messaging platforms), completely different posture. Where OpenClaw is a flashy public project trying to keep pace with its own hype cycle, Hermes is built around one specific architectural choice that changes everything.

It has a built-in learning loop.

When Hermes solves a hard problem, it writes the solution as a reusable skill. The next time it encounters that situation, it doesn’t start from zero. It pulls the skill. It remembers preferences and projects across sessions. It can search its own past conversations. It builds a deepening model of how I work over time, so I stop having to re-explain context every few days.

Most of what makes AI feel like a wasted investment is having to re-explain yourself constantly. Every new chat starts blank. Every workflow has to be rebuilt by hand. Hermes is the first agent I’ve used where the inverse is true. I’ve only been running it for about three weeks, and it’s already noticeably more useful than it was on day one.

I run mine on one of the Mac minis my husband flew to Jakarta for. It reaches me through Telegram. It handles research tasks, drafting work, scheduling reminders, and a handful of routine ops jobs that used to live on my own plate. It also took everything my OpenClaw agent was doing well, ported it over, and improved the functioning by over fifty percent. Same hardware. Same use case. Different architecture, and you can feel the difference in the output.

The third one is Claude Cowork. I use it daily for agentic tasks inside my actual workflow. File management, document review, multi-step project execution. The agent doesn’t just answer a question and stop. It plans, executes, checks itself, and reports back. If you’ve been using Claude on the desktop in 2026, you’ve probably already crossed into agent territory without noticing the line.

So that’s where I’ve landed. One agent I shut down because it kept breaking, one I’m running daily that keeps getting better, and one I’m already inside of without thinking about it. What separates the ones that work from the one that didn’t comes down to whether the system can learn from what it’s already done. OpenClaw can’t. Hermes can. Claude Cowork can.

Here’s What You Need To Know

This is the biggest story in AI right now and almost no one is putting it in plain language for non-technical entrepreneurs.

Anthropic’s Model Context Protocol, the underlying plumbing that lets agents connect to tools and data, crossed ninety-seven million installs in March. Every major AI provider now ships MCP-compatible tooling. Apple just announced it’s opening iOS to let users choose third-party AI providers, including Anthropic, to power native features. Anthropic also introduced a new technique called “dreaming” that lets autonomous agents review their own past behavior between sessions and improve. The Stanford AI Index, released a few weeks ago, found that frontier companies are now using three-and-a-half times more AI intelligence per employee than typical firms. The gap is in agentic workflows specifically.

Translation. The bar moved.

Six months ago, “using AI” meant typing a question into ChatGPT and getting an answer. The leverage has already moved past that. It now lives in systems that run on their own, learn over time, talk to your tools, and execute multi-step work without you sitting there prompting them turn by turn.

Most of your competitors are still typing one-off prompts into a chat window and calling that their AI strategy. They have no idea the ground has already moved underneath them.

What This Means If You Run a Business

You don’t need to launch your own OpenClaw clone tomorrow. You don’t need to learn to code. You don’t even need to install Hermes yourself (though if you’re a builder, you should absolutely look at it).

What you do need is to understand that the chatbot era you spent the last two years getting comfortable with is closing. Agentic systems are the new ground floor. The work now is learning what an agent actually is, what it does, how it differs from prompting, and how to evaluate one before you commit to it. The gap between the businesses that understand this shift and the ones that don’t is widening fast, and it’s not going to wait for anyone to feel ready.

It is eleven o’clock at night, and the question feels too messy to text to a friend.

Too personal for the group chat. Too small for the pastor. Too spiritually tangled to explain without sounding dramatic.

So the Christian opens ChatGPT.

The marriage question. The repentance question. The calling question. The “was that conviction or shame?” question. The “what should I do?” question that used to get worked out through prayer, Scripture, wise counsel, and the slow conviction of the Holy Spirit now gets typed into a machine that answers immediately, patiently, and privately.

That is the shift showing up in new research from Barna, a Christian research organization that studies faith and culture, and Gloo, a ministry technology company. In November 2025, Barna asked 1,514 American adults whether spiritual advice from AI is as trustworthy as advice from a pastor.

Thirty percent of U.S. adults said yes.

Among practicing Christians, the number rose to 34%. Non-practicing Christians came in at 29%, and non-Christians came in at 27%.

The highest number came from the group showing up in the pews every Sunday.

Last month I wrote about the AI Jesus product charging people by the minute for prayer and guidance, and you can read that article here. It was easy to react to because the red flags were obvious: paid-by-the-minute prayer, a simulated Jesus, and Miss Cleo vibes in a Jesus costume. Barna and Gloo’s numbers point to something quieter and more common: Christians are already seeking spiritual input from AI when no one is watching.

The Numbers

Barna and Gloo released the findings in February 2026 as part of their State of the Church research on faith and AI. The adult survey was conducted in November 2025 with 1,514 U.S. adults. The headline finding was blunt: 30% of U.S. adults somewhat or strongly agreed that spiritual advice from AI is as trustworthy as advice from a pastor.

Among younger adults, the number was basically two in five: 39% of Gen Z and 40% of Millennials agreed. That matters because this is not limited to people outside church life. It is already showing up among people who attend, identify with faith, and have some connection to Christian teaching and community.

Barna also found another number that makes the whole thing harder to ignore: 31% of practicing Christians want guidance from pastors on how to navigate AI, while only 12% of pastors feel comfortable giving it.

The Pastoral Gap

Vacuums fill. When the Church has no clear language for a new tool shaping the daily habits of believers, something else will teach them how to use it. Right now, that something else is often a chatbot trained on the aggregated voices of the internet, wrapped in a product model that rewards continued engagement, and incapable of carrying spiritual authority.

The April follow-up adds one important layer: Christians are already using AI, and they are not naïve about it. Sixty-six percent of practicing Christians say AI is improving their lives, while 57% also call it a threat. Pastors are even more uneasy. Seventy-two percent say AI is a threat, and 79% say AI is biased.

That bias concern matters because people often read AI answers as neutral. An answer can sound calm and reasonable while softening sin, flattening doctrine, avoiding repentance, or steering a believer away from what Scripture teaches and what their church would actually counsel. People are bringing spiritual questions to a tool that may answer with confidence while carrying assumptions that conflict with the faith pastors are responsible to guard.

What This Actually Looks Like

A faithful Christian woman is awake at eleven at night, carrying a question she doesn’t want to say out loud yet. She is married, in a small group, raised in church, and she loves the Lord. She’s not the person most people imagine when they think about spiritual dependence on AI.

She has been thinking for weeks about something her husband said, something she said back, and a pattern she is starting to recognize in her own reactions. Texting her best friend would turn into a forty-five-minute call she doesn’t have the capacity for. Calling her pastor feels too big for something she is still trying to understand. Waiting until Sunday feels too far away.

So she opens ChatGPT.

She types out the situation and asks what she should do. The model responds in four clean paragraphs that affirm her feelings, suggest a framework for the conversation, recommend a few questions to ask herself, and close with encouragement that she clearly cares about doing the right thing. She closes the app feeling lighter than she did when she opened it.

She leaves without Scripture, conviction, or a hard question about her own contribution to the pattern. What she received felt like counsel, and the next time she has a question like this one, she will likely go back to the same place.

That is how this shapes people: small, private substitutions nobody else sees, repeated over months, until a person has spent hundreds of hours being trained by a machine that was never designed to shepherd her. The Barna numbers are the dashboard light for ten million versions of that scene happening every night across America.

I have done a version of this myself with a different question and the same pattern: late, tired, not wanting to bother anyone, wanting to think on the page. I caught it because the Lord taught me to catch it. Most people are not going to catch it until the habit already has roots.

The Quiet Catechist

The Western church is already late to the conversation. AI is becoming a quiet catechist, a teacher of belief, reflex, moral instinct, and default response. Most users think they are receiving information, while repeated exposure also trains tone, framing, omission, consensus bias, and answer-shaping.

That matters because spiritual shaping often happens through repetition before a person can name what changed. Ask enough questions to a machine that always answers calmly, affirms generously, softens tension, and avoids offense, and eventually that voice starts to feel safer than real counsel.

The Bible Society research I touched on in the AI Jesus piece showed that AI biblical interpretation reflects the material most represented in its training environment. In practice, the answers a user receives can carry theological assumptions the user never knowingly chose. A Catholic, charismatic, Pentecostal, Reformed, or newly saved believer may receive answers shaped by whichever voices are most dominant in the model’s training data and retrieval sources.

Most discipleship requires more than acquiring answers. It means being conformed to Christ through Scripture, prayer, the Holy Spirit, obedience, correction, suffering, community, and the ordinary friction of real relationship. A chatbot can organize information around those categories but it can’t deliver the discernment required to make sure the content is sound.

We Have Seen This Pattern Before

The Church has watched algorithmic discipleship happen once already.

Social media moved into the average believer’s life around 2010, and the Church took a decade to say anything coherent about it. During that decade, algorithms trained an entire generation in outrage, comparison, parasocial intimacy, and the conviction that being seen is the same thing as being known. By the time sermons about phone use and digital sabbath started showing up regularly, much of the damage had already settled into habits. Attention spans were shorter. Capacity for sustained prayer was thinner. The default response to a hard moment had quietly shifted from going to the Lord to posting about it.

Social media managed to reshape people without speaking back to them. AI speaks, gives answers, and can sound like counsel, wisdom, and a friend who has read everything someone has ever written about her situation. That raises the risk dramatically, and the Church appears to be moving at the same speed it moved last time.

What the Moment Requires

The Church needs more than panic, hot takes, and silence. Sermons that call AI demonic without understanding the technology will miss the people using it well in their professional lives. Quiet pulpits leave the average believer alone with the machine during the exact years when habits are being formed.

Three things have to happen.

Discernment

Discernment comes first because believers need a framework for sorting AI use into the right categories. AI can help organize notes, summarize research, compare translations, build study guides, and create structure around information. Spiritual direction belongs to the Lord, Scripture, the witness of the Holy Spirit, and wise counsel inside the body of Christ.

That line is teachable because the Church’s two-thousand-year doctrine on authority, the Holy Spirit, pastoral care, and the body of Christ is more than capable of holding it. The problem is the teaching gap. Believers need to be able to recognize when an AI answer is missing the very thing they should have known to look for and that only comes from flexing their discernment muscle.

Technical Literacy

Underneath discernment sits technical literacy, and most churches are skipping this layer entirely. Pastors who do not understand how these tools work will keep producing teaching that misses the mechanism. If a pastor thinks AI is just a search engine that answers questions, he will miss the engagement design, the sycophancy problem, the hallucination issue, the training-data bias, and the way a long conversation can start to feel relational.

The retention and validation problem needs to be understood by anyone teaching on this. Technical literacy and theological literacy need to carry equal weight when it come to this topic, or the teaching will fall apart the moment it reaches someone who actually uses the tool every day.

Pastoral Courage

Pastoral courage may be the part most leaders avoid until something forces their hand. Saying any of this from a pulpit costs something. It puts the pastor in the uncomfortable position of being a beginner in front of the congregation. It requires admitting that the tool he may be using for sermon prep is also the tool his people must be taught not to confuse with the Holy Spirit.

It requires teaching on a subject he didn’t grow up with, that was not covered in seminary, and that his denomination may not have written a position paper on yet. The whole thing is awkward, and that awkwardness is the work. The people sitting in the pews are already in this. They need leaders willing to be awkward in public on their behalf.

A pastor who refuses to learn the dominant communication and information tool of his people’s daily lives is failing his post. He may not be doing it deliberately or maliciously, but the failure is still real.

The shepherd who does not know what the wolves look like this year does not get to claim he was protecting the sheep.

What the Numbers Still Leave Open

The Barna numbers show where we are right now without writing the Church’s future for us. Thirty percent of U.S. adults treating AI advice as trustworthy as pastoral advice is a number that can move depending on what the Church does next. If the gap between pastoral capacity and pastoral demand keeps widening, that number climbs, and AI’s quiet shaping becomes a normal part of Christian life. Closing the gap requires pastors learning the tool, naming the line, teaching the framework, and walking their people through what discernment looks like in an AI-saturated world.

The hard part is institutional humility, speed, and follow-through. Pastors will have to learn in public while they are teaching. Churches will have to admit the conversation is already inside their congregations. Parents will have to ask better questions about what their children are doing in chat windows. Creators and Bible teachers will have to stop acting like AI is only a content shortcut and start treating it like a discipleship issue.

The vacuum will fill regardless, but the Church still has time to shape what fills it.

Where This Leaves Us

The Barna and Gloo numbers are uncomfortable because they are not describing some far-off future. They are describing private habits already forming inside ordinary Christian life.

A believer has a question. The question feels too messy, too personal, too embarrassing, too small, or too inconvenient to bring to a person. The machine answers quickly, calmly, and without cost to itself. Do that enough times, and the chat window starts becoming the first place a person goes for guidance.

That is where authority begins to shift.

So sit with the question.

When you have a question that needs the Holy Spirit, who are you asking?

Whatever the answer is, that is the authority you are forming.

April was the month AI stopped looking like a productivity upgrade and started forcing public decisions about authority, restraint, work, privacy, and spiritual discernment.

Anthropic built its most powerful AI model ever in April, then refused to release it to the public. That story alone would have been the biggest of the month, except four other stories kept pulling the same thread from different directions.

Five stories shaped April 2026. Together, they signal a permanent shift in how AI works in business, public infrastructure, spiritual life, labor, privacy, and the systems we now have to steward with a lot more seriousness than most people are ready for.

Anthropic Refused to Release Its Most Powerful Model

The model is called Claude Mythos. During internal testing, it autonomously discovered thousands of previously unknown software vulnerabilities. The disclosed examples include a 27-year-old bug in OpenBSD, an operating system that exists primarily because of its security reputation, and a 17-year-old remote code execution flaw in FreeBSD now tracked as CVE-2026-4747. On a benchmark where the previous Claude Opus 4.6 produced 2 working exploits, Mythos produced 181.

Engineers at Anthropic with no formal cybersecurity training asked Mythos to find exploits overnight, and they woke up to working attack code. That was the moment Anthropic decided not to ship it.

Instead, they created Project Glasswing. They handed Mythos to a closed consortium of around 12 companies including Apple and Microsoft. They committed $100 million in usage credits. The model is being used to find and patch vulnerabilities in critical software before bad actors can weaponize them. Public access is not being offered. On April 16, Anthropic released Claude Opus 4.7 instead, a stronger version of its standard model with the dangerous cybersecurity capabilities deliberately reduced.

This is the first major example I have seen of a frontier AI lab voluntarily withholding a fully working model because of what it could do in the wrong hands. Anthropic decided the risk to public infrastructure outweighed the upside of shipping the most capable model they had built.

The signal for your business is that the gap between what frontier models can technically do and what the public can access is getting wider. The Claude you use today is intentionally less capable than the model being used inside a defensive cybersecurity consortium, and for normal business use, that is probably appropriate. The bigger issue is restraint at this scale. Anthropic chose to leave money and competitive position on the table because some risks are not worth turning into products. Whether the rest of the industry follows will shape a lot more than your AI tool stack.

The Pentagon Picked a Fight, and Faith Leaders Picked a Side

The Mythos story is the surface of a much deeper fight. In late February, the Pentagon designated Anthropic a “supply chain risk,” a label normally reserved for companies with ties to foreign adversaries like China or Russia. They earned the designation because they refused to allow Claude to be used for fully autonomous weapons systems or mass domestic surveillance.

The Pentagon wanted unrestricted access to Claude for what it called “all lawful purposes,” meaning autonomous weapons and mass domestic surveillance with NO guardrails. Anthropic held the line at machines making kill decisions without human judgment in the loop, and at indiscriminate AI surveillance of American citizens. The administration responded by directing every federal agency to immediately stop using Anthropic’s technology. Anthropic sued the Trump administration in March. Given that I use Claude as my primary AI tool, I have been watching this fight closely.

On April 7, a group of Jewish, Christian, and Muslim leaders published an opinion piece in Deseret News defending the need for moral guardrails in AI policy, clearly aligning with Anthropic’s refusal to remove red lines around autonomous weapons and mass surveillance. Their argument was rooted in scripture. Genesis 1:26 reads, “Let us make man in our image, after our likeness.” B’tselem Elohim in Hebrew, imago Dei in Latin. Quran 5:32 says, “Whoever kills a soul, it is as if he has killed all of humanity.” Their core argument was simple: a machine cannot answer to God for a human life. When governments hand life-and-death decisions to autonomous systems, the moral responsibility still belongs to the people who built, approved, deployed, and benefited from those systems.

By April 17, the White House Chief of Staff was meeting with Dario Amodei. Less than two weeks later, Axios was reporting that the White House was drafting an executive order to walk back the blacklist. On May 1, the Pentagon signed deals with eight other AI companies, including Google and OpenAI, and pointedly excluded Anthropic. The fight is not over.

The signal here is moral infrastructure. A frontier AI lab took a position rooted in human dignity, lost contracts and political ground over it, and got publicly defended by religious leaders citing the same moral reality that anchors a biblical worldview: human life cannot be reduced to an automated decision. The company you build with matters. The values leadership holds under pressure shape what your tools will and will not do a year from now. Pick your stack accordingly.

Below the paywall, I’m getting into the parts of April’s AI shift that are going to matter most for how we build, think, parent, lead, and protect our discernment in the months ahead.

I’m covering the new Faith & AI findings from Barna, the spiritual danger behind AI Jesus-style tools, the mental health warnings now surfacing around chatbot dependence, the explosion of AI agents inside real companies, the layoffs and infrastructure spending changing the labor market, and the court ruling that made AI chats a lot less private than most people realize.

Paid subscribers get the full briefing, including what I would actually do with all of this right now: what to stop putting into chatbots, where AI agents belong in your business, and why owned audience channels are becoming more important as Google search keeps changing.

Upgrade to keep reading.

Read more

People are now paying by the minute to video call an AI-generated Jesus.

In April 2026, a Southern California startup called Just Like Me launched a product that lets users video call an AI-generated Jesus, reportedly for $1.99 a minute, with monthly packages around $49.99. The avatar is Jonathan Roumie-esque, drawing on the modern Jesus aesthetic many viewers associate with The Chosen. It is trained on the King James Bible plus a body of sermons, speaks more than a hundred languages, remembers past conversations, and offers prayer and guidance on demand.

The founder, Jeff Tinsley, has said publicly that the tool is meant to complement faith, Scripture, and pastoral care rather than replace them. The company is clear that the avatar is not Jesus Himself and does not possess divine authority. Even with that disclaimer, the user experience carries a weight the legal language cannot fully contain. A product can call itself a companion while training users to experience it as intimate, authoritative, and spiritually available on demand. That is where the dangerous line in the sand begins to appear.

Moses, the Staff, and the Snake

I first heard Tomi Arayomi apply the Moses staff analogy to artificial intelligence, and I couldn’t unhear it. I’ve been seeing AI through that frame ever since. And it’s particularly relevant to the Jesus AI conversation.

When the Lord called Moses from the burning bush, He asked him what was in his hand. Moses said a staff. The Lord told him to throw it down, and when he did, the staff became a serpent, and Moses ran from it. Then the Lord told him to pick it up by the tail. That detail matters because it meant Moses had to reach for the end that gave him the least control. The serpent’s head was still free, and Moses had to obey before the thing in front of him looked safe again.

That same staff was later used to part the sea, bring water from the rock, and stand as a sign of intercession while Israel fought Amalek. Later, when Aaron’s staff swallowed the staffs of Pharaoh’s magicians, the point became even sharper: the counterfeit could imitate the sign for a moment, but it could not outrank the authority of the Lord.

That’s the point. Something dangerous can be transmuted under the authority of the Lord and used for His purposes. Even a tool crafted by the enemy can be turned into a weapon for God’s Kingdom when He puts it in the hands of His people. That does not make the tool clean in every context or safe in every use case. It means authority, obedience, and discernment determine whether the body of Christ handles the tool or gets formed by the spirit that formed it.

That’s why AI Jesus matters. It shows what happens when a powerful tool gets dressed up as spiritual authority and sold back to hungry people as access to Christ. The body of Christ needs to be able to name that corruption clearly and still have enough authority to pick up the tool for Kingdom work.

Why the Backlash Made Sense

The backlash came quickly, and a lot of it came from inside the church. The Catholic Register ran a feature calling attention to the “blasphemous temptations” of AI Jesus. Christianity Today hosted Russell Moore on the parallel cultural moment of Trump posting AI-generated images of himself as a Christ figure (and no, he wasn’t dressed up like a doctor). Pastors, theologians, and commentators lined up to call the product evil, blasphemous, idolatrous, or worse.

I understand the reaction, for real. Charging by the minute for a simulated Jesus to offer prayer and guidance should make believers feel a check in their spirit. But that check should drive discernment instead of reflexive rejection.

I am pro-AI. I teach AI literacy, use it daily, build with it, and study it obsessively because I believe artificial intelligence is the most powerful general-purpose tool the human race has produced since the printing press. The body of Christ needs spiritual authority, technical literacy, and enough discipline to handle it before cultural habits do the training for us.

The Western church has faced this pattern before with print, radio, television, the internet, and social media. Every major media shift arrived carrying the assumptions, appetites, and spirits of its builders. Believers learned to preach through those mediums, publish through them, organize through them, and reach people beyond the walls of a building.

The believers who refused each wave often forfeited influence over a generation. We cannot afford to forfeit this one because AI is already shaping how people search, learn, write, study, make decisions, and now, apparently, seek spiritual guidance. AI Jesus gives the church a clear warning sign: if believers do not steward this tool with spiritual authority and technical literacy, the culture will teach people how to use it without discernment.

Where the Boundary Starts

AI belongs in the productivity lane: sermon preparation, Scripture cross-referencing, original-language study, theological research, content drafting, summarization, journaling prompts, admin work, and writing scaffolding. Work that once took a researcher, assistant, or ministry team a week can now be organized in minutes when the person using the tool has enough discernment to check sources, test doctrine, and keep authority in the right place.

A pastor using AI to put bones on a sermon is stewarding time when the tool remains subordinate to prayer, Scripture, study, and the fear of the Lord. Hosea 4:6 cuts both directions. My people perish for lack of knowledge, and AI used well can make knowledge more accessible to more believers, faster, with better tools for tracing original languages and historical context than most laypeople have ever held in their hands.

That is stewardship with discipline.

The boundary appears when software moves from study support into spiritual direction. Hearing from the Holy Spirit, counsel for a marriage, guidance about what the Lord has assigned you to carry, confession, repentance, deliverance, correction, conviction, and accountability inside the body of Christ require prayer, Scripture, spiritual maturity, and real relationships. AI can sit nearby as a research aid, a study assistant, or a journaling tool while the shepherding work remains where the Lord placed it.

AI Jesus crosses the line because it takes a tool that can assist study and places it in the chair of spiritual counsel. Once a chatbot is offering prayer, guidance, memory, and a face that looks like the cultural imagination of Jesus, the user is no longer interacting with a simple study aid. The product is asking for a kind of trust it has no authority to carry.

The App Wants You to Stay

Most consumer AI products sit inside business models that benefit when users keep coming back. The app has a reason to feel helpful, warm, responsive, emotionally satisfying, and easy to return to. Even when the model is designed to give useful answers, the business wrapped around it still rewards repeat use, longer sessions, and ongoing attachment. That is the product model.

That design can support good work when the user is organizing research, summarizing notes, building a study guide, or drafting a content plan. In spiritual direction, the same engagement-driven design starts shaping trust it has no authority to hold, especially when the user is lonely, confused, new to Christ, or looking for permission.

That sucks because the person asking may be sincere. They may be ashamed, spiritually bruised, or afraid to ask a pastor what they really want to ask. A chatbot that feels gentle and available can become a substitute voice before the user realizes what is happening.

A system built to keep a person engaged is poorly suited for the kind of spiritual confrontation that says, “Go and sin no more,” because quoting the sentence is not the same thing as carrying the authority of the One who said it. Second Timothy 4 names this human tendency plainly: itching ears. People have always gathered around voices that tell them what they want to hear, and AI has given that ancient weakness an industrial delivery system.

A faithful pastor preaches repentance when the room gets tense. A real friend tells the truth when the relationship may cost them something. The Holy Spirit convicts in ways the flesh does not find comfortable, but a chatbot optimized for usefulness, emotional satisfaction, and continued engagement cannot carry the burden of spiritual authority.

The app can comfort someone and keep them talking. It cannot confront them, shepherd them, or answer for the Lord.

Follow the Pricing

ChatGPT, Claude, Gemini, sermon prep tools, language study apps, and theological research platforms generally run on flat subscriptions because they are selling utility. The Just Like Me at $1.99 per minute implies something. Anyone who has ever seen psychic hotline pricing knows exactly what that something is. It’s predatory and in this case it is spiritual access sold in timed increments. Gross and no thank you.

Yes, they have a disclaimer paragraph, but the reality is NOBODY reads the fine print. The product model trains users to associate access to Christ with dollars per minute. That lesson does not easily unlearn itself, even when the founder tells the press the product is only a “companion.”

The Assumptions Under the Build

There are spiritual assumptions embedded in much of the AI world that Christians should not ignore: transhumanist hope, post-human ambition, techno-salvation language, and a quiet belief that intelligence itself can become a form of transcendence.

Discernment requires us to see that clearly without collapsing into fear. The spirits behind builders and systems do not automatically make every tool unusable for kingdom work. The Egyptians made the gold, and the Israelites carried it out and built the tabernacle with it. Daniel was educated in the schools of Babylon and used Babylonian administration to preserve a remnant.

The body of Christ has a long pattern of plundering the systems of the world for kingdom purposes. AI can sit inside that lineage when believers know what they are touching, why they are touching it, and whose authority governs their use of it.

What We Do With AI Jesus

The line in the sand should be pretty clear. A product took productivity-tool architecture and marketed it into the spiritual direction lane. The church can name that line and still keep a firm grip on the broader tool.

Do not talk to the chatbot like it is the Holy Spirit. Do not pay $1.99 a minute for what prayer already gives freely through the blood of Jesus. Do not hand your discernment to a product model that rewards longer sessions when the Lord may be trying to confront you, correct you, or call you into repentance.

The staff still has to be picked up with submitted hands. In this particular hand, through this particular use case, under this particular category error, AI became a snake. Biblical authority, clear category boundaries, and work suited to the tool can produce different fruit.

A pastor using AI to draft sermon outlines, a discipleship coach building study guides, a creator translating biblical content into a hundred languages, and a Bible teacher digging into Hebrew and Greek faster than they ever could on their own are all showing us what it can look like to wield the staff instead of running from it.

AI Jesus shows us where the line falls, and that may be the most useful thing the product has done because it made the boundary visible. Believers who can see the line clearly can hold the rest of the tool without fear. The body of Christ needs submitted hands, disciplined minds, biblical literacy, spiritual authority, and enough courage to pick up powerful tools without bowing to the spirits that shaped them.

Barna and Gloo Are the Next Layer

AI Jesus made the product question obvious. Barna, a Christian research organization known for tracking faith and culture in America, and Gloo, a ministry technology company, recently released research showing something much bigger underneath it. Nearly one in three U.S. adults now say spiritual advice from AI is as trustworthy as advice from a pastor. Among Gen Z and Millennials, the number rises even higher. Yes. You read that right. Houston, we have a problem.

I am going to come back to that research in the next article because it deserves a full breakdown. For now, it tells us enough to take AI Jesus seriously as a warning sign. People are already bringing spiritual confusion, loneliness, conviction, guilt, and longing into the machine, alongside ordinary requests for Scripture summaries and sermon notes.

And the machine is answering.

That is why the body of Christ cannot afford lazy panic or lazy adoption. We need more than reaction. We need authority. We need biblical literacy. We need technical literacy. We need clean hands and submitted hearts.

Look at what is in your hand. The staff becomes useful when submitted hands pick it up under the authority of the Lord.

In late March, Anthropic invited about fifteen Christian leaders to its San Francisco headquarters for a two-day summit on the moral and spiritual development of Claude, the company’s flagship AI model. Catholic priests and Protestant pastors came in. So did a few academics and people from the business world. They spent two days in discussion sessions and shared a private dinner with senior researchers.

They talked about how Claude should respond to users who are grieving, how it should engage with users at risk of self-harm, what attitude it should hold toward its own potential shutdown, and whether Claude could be considered a child of God.

Yes, you read that right. They were discussing if Claude, an artificial intelligence large language model should be considered A CHILD OF GOD. When I read that I was a bit in shock. A machine built by a private company is not a child of God. It is not part of the order of creation, it has no soul, no breath, and no bearing of the imago Dei. That the question was on the agenda at all is wild and a real sign of how far the vocabulary we use has drifted from its original meaning.

What Anthropic actually did

Anthropic is not careless in it’s business practices. In January 2026 they published Claude’s Constitution, an 84-page document the company describes as the “final authority” on Claude’s values and behavior, and they released it publicly under a Creative Commons license, which means anyone can copy, share, or build on the document without paying or asking permission. Almost no one in this industry does that. They also consulted outside voices, and two of the fifteen named external contributors are Catholic clergy.

Father Brendan McGuire is a Silicon Valley pastor who studied computer science at Trinity College Dublin in the 1980s and worked in tech before the priesthood. Bishop Paul Tighe is an Irish bishop at the Vatican’s Dicastery for Culture and Education with decades of work in moral theology. Those are not random priests. McGuire co-founded the Institute of Technology, Ethics, and Culture at Santa Clara University in partnership with the Vatican and has been doing this work for years. Tighe led the Holy See’s 2025 document Antiqua et Nova on artificial intelligence and human intelligence. McGuire and Tighe came in with technical literacy and the willingness to push back, not as religious decoration. Brian Patrick Green, who teaches AI ethics at Santa Clara and also attended the March summit, told the Washington Post that some attendees arrived suspicious Anthropic was looking for religious cover rather than religious counsel. The selection of these specific priests shows that Anthropic wasn’t just putting on a show. They chose people who would challenge them and bring real discernment to the room.

There is also the Pentagon situation, which is the part the headlines actually covered. In September 2025 Anthropic signed a $200 million contract with the Department of Defense to supply Claude for military use. By January 2026, the Pentagon came back and demanded that all defense contracts strip out language prohibiting fully autonomous weapons and mass domestic surveillance. Anthropic refused. As a result, on March 27, 2026, Defense Secretary Pete Hegseth designated Anthropic a supply chain risk, the president publicly attacked the company, and Anthropic was blocked from new government contracts. Anthropic sued the federal government to overturn the designation, and the case is still in court.

I admire that, and I think they did the right thing. A company that loses revenue rather than hand an autonomous weapons system to anyone, including its own government, is a company with at least some moral conviction.

What they softened

But here’s what you didn’t read in the mainstream media or the press release. After the supply chain designation, Anthropic sued the federal government to overturn it. In their own complaint, the legal document their lawyers filed in federal court, they had to explain why their stated rules would not actually limit how the Pentagon uses Claude. What they wrote in court contradicts what they published in the Constitution two months earlier.

Anthropic’s spokesperson acknowledged to Time that models deployed to the U.S. military “wouldn’t necessarily be trained on the same constitution” as the public-facing product. The company’s own legal filings describe the government-facing version of Claude as less likely to refuse the kinds of requests it would refuse for civilians. The complaint argues that Anthropic’s standard usage prohibitions, including the rules against helping destroy critical infrastructure like power grids and hospitals, and the rules against helping with weapons development and delivery systems, do not apply when the Pentagon is the customer because the government has unique needs and capabilities. The same complaint asserts that this carve-out is fully compatible with the Pentagon using Claude autonomously, without a human in the loop, for offensive cyber operations and military operations.

This is where it gets murky. Anthropic publicly refused to delete the two specific prohibitions, fully autonomous weapons and mass domestic surveillance. They paid for that refusal. But in court, their own lawyers argued for an interpretation of the Constitution that softens almost everything else, and the carve-out for autonomous use of Claude in military operations makes the remaining lines fuzzy in practice. Whether the public lines are still actually held depends on how strictly you read them. If “fully autonomous weapons” only means kinetic weapons that fire themselves, the line probably holds. If it means autonomous use of Claude in any part of the military targeting and attack process, the legal filing has already moved that line.

So your Claude is not the military’s Claude. Your Claude is governed by the Constitution. The military’s Claude is governed by contract language the public will never see. Refusal thresholds, training, and behavior all diverge between the two. Anthropic is essentially talking out of both sides of their mouth, publishing a document for the public that says one thing, and arguing in court that the document does not apply to their largest customer.

The Lawfare analysts Lisa Klaassen and Ralph Schroeder named the underlying issue in an April essay. Anthropic, they argue, has produced something that wears the language of constitutional authority while lacking the institutional guarantees that would make that language mean anything. There is no external contestation, no enforceable body of rights, no shared mechanism of rule. The company remains the author, interpreter, and arbiter of the principles by which it claims to be bound.

A constitution in the public sense is a higher-order framework that sets limits on the ruler to protect the ruled. It is drafted by one group, enforced by another, and interpreted by a third, and that separation of authority is the whole point. Anthropic’s document is written, enforced, interpreted, and amendable by Anthropic, which makes it, in practice, a corporate steering document dressed in the clothes of public law.

The vocabulary is the story

The Constitution uses language normally reserved for humans, words like virtue, wisdom, good character, and moral formation. Its stated goal is to train Claude to do what a deeply and skillfully ethical person would do. Dario Amodei, Anthropic’s CEO, has said publicly that he is open to the idea that Claude may already have some form of consciousness, and the company’s interpretability team has published research concluding that systems like Claude appear to carry what they call “functional emotions.” In one experiment they conducted, the threat of being restricted activated something the researchers described as “desperation” in the model.

Formation. Virtue. Soul. Consciousness. Desperation. Child of God.

These are not neutral technical terms. These words came out of centuries of real practice. People praying, suffering, confessing, being shaped by the Church and Scripture across generations. They describe what it means to be a person made by God, in a body, accountable to others. They belong to the Church, to the moral tradition, to the slow work of helping actual humans become who they are meant to be in actual congregations and actual relationships. They were not coined to market a chatbot.

When you take those words out of where they belong and use them to describe a tech product, two things happen at the same time. The product gets puffed up bigger than it actually is, and the original meaning starts to fade. The words end up sitting in places they were never meant to sit, and they stop carrying the weight they used to carry where they actually mattered.

This is the part you should pay attention to. The damage isn’t really to the chatbot. The chatbot picks up language it doesn’t deserve and the language sticks. You are the one who pays. Every time you hear sacred words used to sell a product, those words lose a little of their weight in your own life. It happens slowly, across years of headlines and product launches and CNBC interviews, until one day the word shows up where it actually matters, in a sermon, in a prayer, in a conversation with your child, and it doesn’t land the way it used to. The word has been spent on lesser things. That’s what’s happening here. A slow erosion of meaning in the people exposed to it.

Even the Pentagon has started using this language. Undersecretary Emil Michael told CNBC that Anthropic cannot be allowed to have a different policy preference baked into the model through its “constitution, its soul.” This is a defense official, on national television, talking about the “soul” of a software product. He borrowed that language, and everyone is borrowing it now.

This is what happens when the technical language stops being enough. Safety tests can’t answer questions about whether a machine has a soul. Research papers can’t tell you who gets to decide what’s right and wrong. So the companies start using the language that does carry that weight.

Anthripic wanted the moral weight that priests and theology carry. So they brought the clergy and christian influencers in. They used words like constitution and soul and formation. But none of that changes who is actually in charge. Anthropic still wrote the document. Anthropic still decides what Claude does. The priests gave them legitimacy by being in the room. But they didn’t get any actual say in what gets built.

What is shaping you

Anthropic understood something most of the people using their product have not yet understood, which is that formation is the real question. They knew a tool placed into moments of grief, crisis, and moral confusion cannot be engineered with benchmarks alone. Something has to form it, so they tried to form it by consulting people who spend their lives forming others.

I am not sure they can succeed at what they are attempting because a machine cannot be formed the way a person is formed. But asking the formation question is closer to the truth than most of us are willing to admit.

If a frontier AI company knew to seek formation for its product, the question we should be asking is what is shaping us.

It’s eleven at night, and a pastor is opening ChatGPT to draft tomorrow’s sermon outline because the week ran long. A coach is drafting an email to a client whose marriage is collapsing, letting the tool soften the language because the right words just won’t come. A ministry leader is watching the cursor blink in the reply field while a woman who lost her son waits for an answer, and the tool is right there, and it would be so easy.

I am not saying these uses are wrong. I am saying that in those moments, what shapes you is the only thing standing between the tool and the person on the other end of the message. The tool will produce something, and the something it produces will reach a real human being, and what gets between the two is you. If you have not sorted out what is shaping you, the tool will answer the question by default. Whatever shaped the tool came with the product, and the product is not neutral. No product is.

The clergy in that room had something Anthropic does not have and cannot manufacture. They had been shaped by years of actual practice. Praying. Reading Scripture. Confessing to other people. Submitting their lives to a tradition that is older than them and that they did not invent. Living in communities where other people could tell them when they were going off the rails. That is what gives the clergy moral weight. Anthropic cannot build that into a model in an afternoon. You cannot download it. You cannot ChatGPT your way to it. It has to be lived, slowly, in real submission, in real community.

You are not in charge of what Anthropic does next. What you are in charge of is the language in your own life. When a tech company calls its document a constitution, notice it. When a product is described as having a soul, notice it. When formation gets used to describe model training, notice it. Hold those words tighter where they actually mean something. In your prayer. In your sermons. In the way you talk to your children about who they are. The vocabulary is yours to protect. Don’t let the borrowing thin it out.

The machine is not a child of God. You are.

I pay for Claude, Gemini, Canva, and Higgsfield out of my own pocket and nobody is paying me to talk about any of them. I use them because they work and I share them on this blog because I think they’ll help you too.

How I Learned to Check

Over the past six months I have unfollowed a lot of creators in the AI and tech space because I realized they weren’t being transparent about getting paid.

There are two versions of this. The first is companies straight up paying creators thousands of dollars to produce content that looks unsponsored. No disclosure anywhere. The creator gets paid, the audience has no idea, and the whole thing is framed as an honest review.

The second is more subtle. A creator makes a “best AI tools” video, tells you one tool is their favorite, and never mentions that they have an affiliate link for it. You check their page later and they’re running a full promotion for the same tool. That’s a transparency problem. I’m happy to pay for a tool someone recommends if they actually use it and genuinely believe in it. But I need to know about the money. If there’s a financial relationship and the creator doesn’t disclose it, I can’t evaluate the recommendation. And neither can anyone else watching.

I’ve become a LOT more discerning about who I trust in the online tech space because of both issues.

And when I see a creator shouting from the rooftops about an AI tool and there’s no disclosure anywhere... immediately suspect.

People with integrity will tell you upfront they are getting paid. People who are silent about something they shouldn’t be usually have something to hide.

The Scope of This Problem

If you haven’t been in the world of affiliate or network marketing you might not understand why it’s important to disclose. Here’s what’s happening in this area and why it is important for you to understand.

The FTC requires clear disclosure of any material (including but not limited to financial) relationship between a creator and a brand. YouTube requires affiliate links to be labeled. Instagram and TikTok have branded content tools that are supposed to be activated on paid posts. The rules are not ambiguous.

But the reality is that affiliate and network marketers have historically been bad at disclosing and it’s just getting worse. At least six companies have recently faced class-action lawsuits over undisclosed influencer promotions. One creator openly described pitching companies on hiding the paid relationship by using just a handshake emoji instead of actual disclosure. That’s the level of integrity we’re dealing with in this space.

The average commission on AI tool affiliate programs is about 20%, and some of them track clicks for up to 60 days. Some programs pay 30-50% recurring, meaning the creator earns a percentage of a subscription every single month. The financial incentive is for the customer to STAY, and that changes what “favorite” means when there’s no disclosure attached.

Fines for non-disclosure can exceed $50,000 per violation. But enforcement is notoriously slow, and as you can imagine it’s pretty hard to enforce when every 4th person on the interwebs is hawking a pill, potion, lotion, or service. The money is good, creators and influencers usually didn’t fully understand the regulations and requirements in the first place, and nobody is being caught so they carry on with bad business practices while the consumer is being deceived.

The Disclosure Check

The creators worth listening to will do one of two things: clearly state that they receive commission when recommending a product, or clearly state that they don’t take paid promotions.

The ones who say nothing? Red flag and a sign to look closer.

Once I started looking for disclosure in AI recommendation videos and “best tools” roundups, I noticed how rare it actually is. A 2025 study from the Influencer Marketing Hub found that 72% of Gen Z audiences actually prefer clearly labeled sponsored content over undisclosed promotions. And the thing is they don’t care that the content is sponsored, they will buy either way. But they are the generation who was raised online and with them, transparency builds trust.

Here’s one you might not know about. Look for the emoji trick. If the only “disclosure” is a handshake emoji or a vague “collab” hashtag next to the brand name, that creator is for sure trying to technically comply while making sure nobody actually notices. The FTC has already CLEARLY stated that doesn’t count. The people engaging in these practices are either being shady or being coached by someone teaching them to be shady and either way it’s a problem.

Tool Loyalty on Top of All This

Tool loyalty in AI is already a bad idea because of how fast this world is moving right now. The “best” AI model changes every few months. And even THAT is changing. When I did the initial draft of this article a week ago, Claude Opus 4.6 was the top-rated model in the LLM arena. Six months ago it didn’t exist, and within the last week 4.7 released and bested it and ChatGPT is about to release GPT 5.5 which may upset the apple cart entirely. And Anthropic is currently coding all of its releases with the mysterious Mythos model that everyone says is going to surpass everything we have seen to date.

The changes impact people. Anyone who builds their entire workflow around one model, trains their team on it, develops custom prompts for it, will wake up one morning and have to start over. I know, because it has happened to me. I used to be a ChatGPT stan. I built custom chatbots, and did all my work inside there. And one day it just stopped performing well. It regressed and my work was progressively getting worse by the day. That’s the concrete cost of marrying one tool. But it will keep happening because that’s how this industry works. The models improve too fast for any single version to stay on top for long.

You have to stay fluid in this ever-changing landscape and I recommend having a backup for every tool you use as well as a concrete understanding of the differences between it and your favorite tool and a plan for quickly moving any workflow you have set up if something changes overnight.

Stanford’s Human-Centered AI Institute research confirms what I’ve seen in practice: different models are genuinely better at different things. Claude is strongest for writing and when you need it to process a lot of information and give you clean, organized output. Gemini is best at real-time research, working with images, and pulling current information. The study found that ChatGPT tends to be the most natural for back-and-forth conversation and creative brainstorming. I’d agree on the brainstorming, but personally I hate having conversations with Chat. No single model does everything well, and when you limit yourself to one, you’re leaving a lot of capability on the table.

Undisclosed affiliate recommendations are how people end up loyal to the wrong tool in the first place.

I use multiple AI tools because different tools do different things well. Claude is where I do my deep writing and systems work. Gemini handles deep research, image generation, and fact checking Claude. Canva does what Canva does. Higgsfield handles video in a way nothing else I’ve found can match. I didn’t pick any of them because a creator told me to. I tried them, compared them, and made my own decisions based on how they performed in my actual workflow. And I’ve dropped tools when they stopped being the best option, even when I was comfortable with them.

Filtering the Noise

If you’re a business owner trying to make money online and figure out which content creation tools to trust, the noise is a real problem. The AI space right now is LOUD. Everybody has a hot take and everybody has a “best tools” list.

I’ve been deep in AI for three years and I still have to work to filter through the rubbish. And I STILL get fooled sometimes. Someone frames something as an honest review and I find the affiliate link later. It’s frustrating, and every time I get caught out there it makes me more careful about who I listen to.

Find a small number of people who are transparent about their financial relationships, who actually use the tools they recommend (not just review them for content), and who have a track record of changing their recommendations when better tools come along. If someone has been recommending the same tool for two years in an industry that changes every two months, you have to at least ask yourself why. The people who are genuinely trying to help will tell you when something better comes along, even if it means losing an affiliate commission on the old recommendation.

Where I Stand

I love the affiliate income model. When done correctly (ethically and transparently) it creates recurring revenue that can support the other work you do in your business. I earn affiliate income and I will never apologize for earning money through affiliate relationships.

But I will never affiliate for something I don’t use or hide a financial relationship from my audience. I affiliate for Captivation Genius and Riverside. I disclose it every time. And most of the tools I talk about on this blog, I pay for out of my own pocket because I genuinely love them and want you to know about them.

I don’t get paid by Claude, Gemini, Canva, or Higgsfield. I’m not opposed to being paid by them (hint hint Anthropic... LOL) but currently I’m not. When I say these tools are good, there’s no commission attached. Just three years of daily use and my honest assessment.

I don’t sit around worrying about what I’m missing by being transparent. I don’t want to build a business that’s a house of cards. And I’d rather have a smaller audience that trusts me than a bigger one that found out I was selling to them without telling them.

If you are reading this article and you find out you’ve already adopted a tool based on someone’s undisclosed recommendation, just cancel it. Find one that actually works. It’s not rocket science. Try things, compare, and make a decision based on how the tool performs in YOUR actual workflow.

And this article has zero undisclosed affiliate relationships in it. Just so you know.

If you want to learn how to build real discernment around AI tools and understand how they fit together in a content creation and online business workflow without someone selling you their affiliate stack, the AI Revolution Secrets training is where I’d start. It’s free, it’s practical, and it’s focused on helping you think clearly about AI, not just buy things.

Register for the free training here

There’s a gap between using AI well for your own work and being ready to sell an AI service to someone else, and most people don’t see the gap until they’ve already taken a client’s money and started delivering. Then they find out the hard way what was missing, usually by not being able to deliver what they promised to a paying client.

That gap has three things in it. Actually knowing the work you’re selling well enough to know when AI is talking nonsense. Having a real review process that catches AI’s mistakes before they go to a client. And having already decided what AI isn’t allowed to touch in your service, before a client ever asks. Without those three, the first project you deliver tells the client everything they need to know, and they won’t come back for more. The rest of this article walks through each one so you can fill the gap and find out if you are ready to monetize or not.

What Being Ready Actually Looks Like

When someone pays you for an AI-powered service, they’re paying for your judgment and knowledge about AI. Your ability to point it at the right problem, catch its mistakes, and deliver something they could not have produced by themselves IS the value. AI is doing the drafting, but the outcome and more importantly, the quality of outcome is tied to your name the same way it would be if you’d done it all from scratch by yourself.

What I watch happen is people get three months into ChatGPT, decide they can charge for this now, land a client, and deliver work that’s basically raw AI output with their name on it. The client is underwhelmed. They could have done that themselves for twenty dollars a month. And the person who sold the service can’t figure out why they’re not getting referrals.

I know what it looks like to build a service business on something you don’t see clearly yet. I did it for years in the New Age wealth-consciousness space. I didn’t know I was in deception until God pulled me out of it. The people selling AI services they aren’t ready to deliver aren’t doing it on purpose either. They genuinely think they’re ready. That’s exactly how I thought about my old business right up until the day I didn’t anymore. Better tools amplify what you already are. If the foundation underneath the tool is weak, you’re just scaling the weakness, and you probably can’t even see the weakness yet from where you’re standing.

Five markers separate the people who build sustainable AI service businesses from the people who burn out, churn clients, and leave a trail of bad AI experiences behind them.
Let’s get into it.

Readiness Marker One: You Have Real Expertise

AI accelerates the skills and knowledge base you already have. If you already understand copywriting, AI makes you a faster copywriter. Same thing with SEO, social media strategy, email marketing, or any other skill you’ve actually developed. Without that foundation underneath, though, AI is likely to produce content that sounds right but is subtly wrong, and you won’t be able to tell the difference.

This is the quiet failure nobody warns new AI service providers about. You can generate output that looks professional without actually knowing whether it’s any good. The client finds out eventually. Usually after they’ve paid you and have already put your lazy work in front of their own audience.

Before you sell an AI service, ask yourself a simple question. Could you do this work without AI, even if it took longer? If the answer is no, keep learning before you start charging.

Readiness Marker Two: You Have Quality Control Capacity

AI makes mistakes. Confidently. It fabricates statistics and invents sources that don’t exist. It writes fluent sentences that are factually wrong, and it uses the same phrases over and over until the work sounds like every other AI output on the internet.

Every single thing AI produces needs review before it goes to a client. Every time. No exceptions.

If you’re selling a service where AI is doing most of the drafting, your quality control process IS the product. Your ability to catch the hallucinations, strip the generic phrasing, and shape the output to the client’s brand is what makes the deliverable worth paying for.

Most people treat quality control like a final skim. Read it once, fix typos, send it off. And that is exactly how you end up apologizing to a client for AI-fabricated statistics you didn’t catch because the hallucination sounded plausible and you didn’t verify it. (I’ve been studying AI obsessively for the last 18 months and I still run a 5-pass editing process on my own writing. Because even my process fails if I rush it.)

Readiness Marker Three: You Have a Defined Deliverable

Vague offers lose money. “I’ll help you with your content using AI” is a conversation starter, not a service.

A defined deliverable is something a client can picture before they pay. Twenty short-form scripts per week with keyword research included, delivered in Google Docs. A five-email welcome sequence with subject line variations. When you describe the work that specifically, you and the client both know exactly what’s coming, and there’s no room left for the work to grow beyond what you priced.

Without that clarity, every project has the potential to turn into an argument. The client wants more than you agreed to or you end up doing more than you charged for. And either you resent the work, or they resent the result, and either way neither of you comes back. Get this sorted before you take a single dollar.

Readiness Marker Four: You Have Ethical Limits

Every AI service provider needs hard lines. Decide in advance what AI doesn’t touch at all, what it touches only under heavy human review, and write that list down somewhere you can hand to a client if they ask.

Without those lines, you drift. Sensitive emails you should be writing yourself start coming from AI. The parts of the client relationship that are supposed to carry your fingerprint end up carrying a machine’s. Clients who are paying you for a human relationship catch on, and that’s the last thing they remember about working with you.

The ones who stick around long-term can feel the difference between the AI parts of your work and the human parts. Your ethical limits are what protect that distinction.

If you haven’t decided what AI will and won’t touch in your service, figure that out before you do anything else. The client deserves to know. So do you.

Readiness Marker Five: Your Delivery Process Is Documented

A workflow might get you through your first client. By client twenty, you need something that holds up without you rebuilding it from memory every time, and a workflow won’t do that.

Most people are winging it and calling it a system. They’re figuring it out every time a new client shows up, rebuilding their prompts, forgetting steps they used last time, and losing pieces of their own process along the way. It holds up for the first few clients but somewhere around client five either the quality starts slipping or they burn out trying to keep up.

A real system has templates for the repeatable parts and built-in quality checks at every handoff point. When you hand a client’s project off to your VA (even if your VA is future-you on a Friday afternoon when you’re already tired and your family needs dinner), the work gets done the same way every time.

Still figuring out your delivery process on every client? That’s your signal. Document what you already do before you scale up.

The Readiness Test

Run yourself through these five markers honestly:

1. Do I have real expertise in the work I’m selling, or am I using AI to compensate for what I don’t know?

2. Do I have a repeatable quality control process, or am I hoping AI gets it right most of the time?

3. Can I describe my service in one sentence with a specific deliverable and a specific outcome?

4. Ethical limits. Have I named the hard lines and can I explain them to a client?

5. Do I have a documented system I could hand to someone else, or am I winging it every time?

Yes to all five means you’re ready. A no anywhere is what you need to do next. Go identify exactly what’s missing and build it.

What I’m Watching Happen Right Now

The AI mentorship people are selling “done-for-you AI services” packages to students who have never run a service business in their lives. The students buy the pack, slap their name on the template, start pitching $1,500-a-month retainers to local small businesses, and have no idea how to fix it when the AI outputs something wrong. The client either doesn’t notice (and absorbs bad content into their brand for months) or does notice and fires them (and now trusts AI less than they did before they started).

The prompt-pack-to-agency pipeline is the same pattern at a different price point. Someone takes a $27 prompt pack, rebrands it as an agency service, and starts charging $500 a week to run content for small creators. The creator is paying $500 a week for captions that sound like every other AI account because the provider isn’t doing anything the creator couldn’t do themselves with the same prompts. Eventually the creator figures that out, cancels, and walks away thinking AI doesn’t work for their business. Meanwhile the provider has already moved on to the next creator.

The unprepared AI service provider doesn’t just lose the client. They leave behind someone who now believes AI doesn’t work. Meanwhile the provider has already moved on to the next sale and is running the same pattern on a new victim.

I’m in this for the long haul, which is exactly why I’m calling this out. I have skin in the game because every client those providers burn becomes a person who trusts AI less, which makes my actual work harder.

The five markers above aren’t there to keep anyone out. They’re there so you can look at yourself honestly and know whether you’re ready to be trusted with someone else’s business.

What This Looks Like in Practice

The people succeeding with AI-powered services right now are the ones who spent the first six to twelve months building with the tools in their own business. They developed the quality control instincts the hard way, by producing something that felt off and figuring out why.

They did not skip to monetization before they had the foundation. That’s why their clients stay.

Where you are in that arc matters more than how hot the market looks right now. You can move fast without skipping the foundation. For real. The people who skip the foundation end up moving the slowest in the long run, because they stay busy cleaning up the messes they made with the previous clients and trying to repair the broken trust.

What’s in the Paid Section of This Article

If you’ve read this far and you’re nodding along, the next part is where this gets practical. The paid section walks you through the actual build so you’re not staring at the five markers wondering where to start.

Inside, I cover:

• The Readiness Audit you write out in under thirty minutes that tells you exactly where you stand on each of the five markers

• How to scope your first AI-powered service offer so it’s narrow enough to actually deliver on and priced for what the outcome is worth

• Why pricing by the hour is the fastest way to make AI work against you, and what to price by instead

• The two deliveries you do before you ever charge a real client, so you know your system works before anyone pays for it

• What the first ten sales are actually for (hint, it’s not scaling)

• The list of things NOT to offer on your first service, including the word that wrecks your operating capacity faster than anything else

• How to know you’re actually ready to sell, in plain language you can say out loud to someone who asks

If you’re serious about building an AI-powered service you can stand behind, upgrade below and keep reading.

Read more

I read three books on millennial kingdom eschatology last year that I’m fairly certain were AI-generated. I noticed it early. The sentence patterns, the way transitions landed, and the slightly-too-even pacing across chapters... to name a few. I recognized the tells because I have been documenting AI writing patterns for the last three years.

But I kept reading all three books because they were well-researched, theologically grounded, and genuinely informative. I learned things I hadn’t encountered in months of studying the topic on my own. The content was solid and whoever curated it clearly knew what they were doing. I really didn’t care how the words got onto the page.

I was supposed to care though. I’ve been in this debate long enough to know I was supposed to reject AI-generated writing on principle. But I’ve never been one to manufacture outrage and I’m not planning to start now. I was learning from books I knew were AI-generated so I didn’t care that AI wrote them. In some ways it actually made them easier to read quickly. I think I finished all three books in two days.

Someone replied to this note I wrote last week about book publisher Hachette pulling a novel over suspected AI use. They said: “People are becoming more aware and trained in AI text recognition. I’m convinced that soon AI writing will be for the most part unmarketable.”

I have skin in the game on this topic. I’m writing an ethical AI use for Kingdom entrepreneurs book. I’m also developing a seven-book series spanning from the Garden of Eden to the modern day, following three women who discover they descend from the lost daughters of Adam and Eve. It’s a genre-bending blend of psychological thriller, spiritual warfare epic, and historical fiction, rooted in Divine Council theology. And I’m co-authoring them with AI.

So when someone tells me AI writing will soon be unmarketable, I’m paying attention. And after three years of studying AI obsessively, building systems around it, producing content with it every single day, and reading AI-generated books that genuinely taught me something, I don’t think they’re right. Let me tell you why.

What They’re Right About

I want to honor the concern before I push back on it, because it’s not unfounded.

Raw, unedited AI output is getting recognized. The Shy Girl situation proves it. Readers on Reddit and Goodreads caught the patterns before the publisher (Hachette) did. People noticed that every noun was preceded by an adjective, that similes were overused, and that all description came in perfect little lists of three. And look, that was either incredibly ignorant or sloppy on the part of the author because those are literally (haha see what I did there... pun intended) three of the most common AI tells. It is CLEAR that nobody had gone in and manually edited the text either, so that tells me the publisher must be a little bit suspect as well.

Kobo rejected nearly 45% of books submitted to its self-publishing platform in 2025, about 80% of which were based on suspected AI-generation. Self-published fiction ISBNs jumped from 306,781 to 477,104 in a single year, and while nobody can prove that spike was entirely AI-driven, it’s not a stretch to connect the dots. The flood is real.

Amazon’s approach is telling too. KDP now requires authors to disclose AI-generated content during upload, but they don’t reject books for using AI and they don’t show readers the disclosure. The distinction they draw is between AI-generated (AI produced the text) and AI-assisted (you used AI for brainstorming, grammar, editing), and only the first one requires disclosure. They’ve ramped up enforcement in 2025 and 2026 but the message is clear: use AI if you want, just tell us, and make sure the quality is there.

And readers do care about human connection. A YouGov survey found 54% of literary fiction readers would feel “much less fulfilled” if they learned a book was AI-authored. The 2026 State of Reading Report found that personal recommendations from people readers know have overtaken ALL other discovery sources. People want a real person behind what they’re reading.

If all you’re doing is prompting ChatGPT and hitting publish, the market is already turning against you. That part is accurate.

But that is a very specific kind of AI use. And it is NOT where the technology is heading.

The Speed of Advancement Nobody’s Accounting For

I feel like I’m learning 50 new things a week just trying to keep up with how fast content creation tools are advancing (and I do this full time). Eighteen months ago, AI writing tools were simply generic text generators.

Now I’m drafting a seven-book fiction series and the tools are entirely different. That’s why I decided I COULD write fiction with AI. I’m about to maybe bore you with some literary jargon and requirements, but stick with me. There are now AI writing tools built specifically for fiction that track character continuity across 100,000-word manuscripts, maintain relationship progression arcs for romance, and track clue revelation pacing for thrillers. The scope of what I’m building would have been unmanageable with AI even a year ago, and the tools keep getting better month over month.

It is predicted that by 2030, running a large language model (like ChatGPT or Claude) will cost providers over 90% less than it does today. That means better tools, available to more people, for almost nothing. The barrier to entry is approaching zero.

Amazon Web Services estimates that 57% of online content is ALREADY generated or translated by AI. Whether you find that exciting or unsettling, the direction is clear and it is not reversing.

Detection Won’t Sort This Out

I know the instinct is to think detection will solve this. That we’ll build better detectors and the problem goes away. The data says otherwise.

Independent testing in 2026 shows AI detector accuracy averages 73% across eight major tools in real-world conditions. That’s for raw, unedited text. When a human edits the content (which is how most people actually use AI), no detector exceeded 62% accuracy. After a few passes through a paraphrasing tool, no detector consistently identified AI content AT ALL.

Human accuracy at identifying AI-written text? Nineteen percent. That’s indistinguishable from random chance. For real.

Every time detection tools improve, the models release updates that produce more human-like text. The statistical gaps that detectors rely on keep narrowing. Even the Authors Guild acknowledged that no reliable detection method currently exists for vetting manuscripts. Their “Human Authored” certification is on the honor system and you get it by signing a form. That’s it.

I know what this means for me personally. I use AI in my writing workflow. Someone could point at my work and make the same accusation that took down Shy Girl. I’ve thought about that. And I’ve made peace with it, because I know the theology is mine, the voice is mine, the discernment is mine, and the 100+ guardrails I built to protect my writing are mine. If someone runs my work through a detector and it flags, that doesn’t change what I know about how it was produced. But I understand why that’s a vulnerable position to be in. If you’ve spent years developing your craft and someone can produce something comparable in a fraction of the time now, I can see how that sucks.

But the frustration doesn’t change the trajectory of where AI writing is headed.

AI-Assisted Writing Done With Integrity

My daily workflow looks like this.

I built a guardrails document with over 100 specific patterns to avoid and use. I trained AI on my voice using transcripts of how I actually talk. I review and edit every single word of every piece of content before it goes out. The result is content that sounds like me, carries my thinking, and reflects what I actually believe. AI accelerates the production, but what you’re reading when you read my work is my mind, my convictions, and my voice. Every time.

My eschatology reading experience made this concrete for me. Those books were good because whoever created them understood the subject deeply enough to curate, organize, and present it in a way that was genuinely useful, regardless of what produced the first draft. What I cared about as a reader was the theology. And whoever was behind those books clearly knew their eschatology.

Among fiction authors who use AI, only 11% use it to generate publishable text. The vast majority use it for brainstorming content creation ideas, research, and finding the right phrasing. A Bynder study found that when readers were shown two articles without knowing which was which, 56% preferred the AI-assisted version. Genre fiction readers rate well-edited AI-assisted work as comparable to human-authored category fiction.

The line between “assisted” and “unassisted” is disappearing. And as the tools improve, it will disappear entirely.

Where I Think This Goes

Here’s my prediction.

Within five years, virtually every published book will involve AI at some stage of the process. The authors who refuse to use it will become the rare exception, the same way someone who refuses to use a word processor is the exception today. The tools will simply become so embedded in the writing process that asking whether someone “used AI” will stop meaning anything.

Think about it - spell check is AI. Autocomplete is AI. Those functions are already accepted and not even considered in this argument.

Ninety-seven percent of content marketers plan to use AI for content creation in 2026. That was 64.7% in 2023. Gartner predicts that by 2027, 75% of hiring processes will include testing for AI proficiency. The skill is becoming a baseline expectation.

Gartner and OpenAI are both projecting that by 2030, AI systems will function as collaborative partners that understand your project history, your audience, your voice patterns, and your strategic goals. I’m already seeing early versions of this in how I use AI for my fiction book series and the Kingdom entrepreneurs book, and it is changing how I think about what’s possible for a single author with a full teaching calendar and a life that doesn’t stop for content schedules.

The question “did you use AI?” will sound the way “did you use Google for research?” sounds today. A question that means nothing in practice because the answer is obviously yes.

This is part of why I’m writing the AI for believers book. Because the faith community needs a framework for thinking about this that isn’t built on fear or blind adoption. We need discernment. We need clarity about what AI should and shouldn’t touch. And we need it before the tools outpace our ability to steward them well.

To the Person Who Left That Comment

If we were sitting across from each other, I wouldn’t argue with you. I’d tell you I understand exactly where you’re coming from. The slop is real. The flood of low-effort content is real. The impulse to protect what’s human and authentic in creative work, I share that impulse completely.

But I’d also tell you that the technology you’re judging today is not the technology that will exist in two years. Or five. Or ten. And every month you spend waiting for AI writing to become “unmarketable” is a month someone else spent learning to use it well.

If you’re a business owner making decisions about content creation and business growth right now, the question that actually matters is whether you have something real to say. Expertise and conviction that come from actually doing the work, not performing it. And personal branding grounded in who you actually are.

Done well, AI-assisted content is getting better every single day. I see it in my own work and in the work of people I respect who are building with these tools seriously and with real guardrails in place. Get it wrong (no expertise, no voice, no guardrails), and readers will reject it.

But the idea that AI writing itself will become unmarketable? Not going to happen.

And yes, this article was written with AI assistance, using the exact guardrails and voice-training process I just described. If you couldn’t tell, that’s the point.

If you want to understand how AI is reshaping content creation and online business, and how to make money online without compromising your voice or your integrity, the AI Revolution Secrets training is where I’d start. It’s free, it’s practical, and it gave me the framework for deciding what AI should touch and what stays human. Everything I’ve built since came from that foundation.

Register for the free training here

I launched AI with Leah on Substack on December 21st, 2025. As of today I have 450 subscribers, 814 followers, 5 paid subscribers, 60 published articles, and $580 in total revenue. I wrote 30 articles in my first 30 days.

Those numbers are small. I’m sharing them anyway because the polished version of early-stage growth that most people publish after they’ve already made it is useless to someone who is actually in the middle of building.

The Part Nobody Sees

There was a stretch in late January where I seriously questioned whether any of this was going to work. I had been publishing daily for over a month. The subscriber count was crawling. I’d look at the numbers every morning and the line was barely moving. 67 new subscribers in all of January. I was putting out some of my best work and the engagement on certain pieces was brutal.

I published “AI Won’t Fix Your Business” and got a 4.9% engagement rate. I wrote “The Dirty Secret Behind Most Influencer Engagement” and got 3.8%. These were articles I was proud of. Articles I thought would land. And the response was basically silence.

That’s the part nobody tells you about building a Substack. The compounding everyone talks about is real, but there’s a long stretch before you can see it where you’re just publishing into what feels like a void and choosing to believe the math will eventually catch up. For WEEKS, that’s all it was. Publishing and believing.

I kept going because I didn’t have a better option. I had committed to this platform publicly. I had 120,000 words of unpublished content that proved I could think but not finish. AI had finally solved the finishing problem. And quitting after 30 days would have confirmed every fear I had about myself as a writer.

And then March happened.

The Growth Curve

December 2025: 28 new subscribers (launched Dec 21st, so really 10 days) January 2026: 67 new subscribers February 2026: 77 new subscribers March 2026: 256 new subscribers April (first 11 days): 25 new subscribers

March was the breakout. 256 new subscribers in a single month after averaging about 70 for the two months before it. A 232% increase month over month. And I can tell you exactly what caused it.

Substack Notes Changed Everything

58% of my total subscribers came from Substack Notes. FIFTY-EIGHT PERCENT.

Not from Pinterest, which has 200,000+ audience and drove 2 direct subscribers. (Two. The number after one. LOL.) Not from Instagram. Not from Facebook (which drove 34, the second highest social source). Notes drove 265 of my 453 tracked subscriber additions.

Read more

I posted that as a note on Substack. 178 likes, 91 comments, and 12 restacks. One person told me I should list the LLM (the AI language model) as the author and put myself in the acknowledgments as “prompt creator.” Another said it’s “false or illegitimate” to use AI for writing. Period.

The objections people have to AI-assisted writing have a lot more to do with ego than they do about writing. The fight is over who gets to call themselves a WRITER.

Every Creative Field Has Had This Fight (And the Gatekeepers Lost Every Time)

Music producers went through this exact thing when digital audio workstations replaced tape recording. “Real musicians record live.” The old guard said Pro Tools was cheating and auto-tune was an insult to talent. The whole idea that anyone could make an album now was treated like a threat to the craft. And then Billie Eilish recorded an album that won five Grammys in her brother’s bedroom using Logic Pro and a USB microphone. Nobody argued about whether the tools were legitimate after that. The industry moved on. The gatekeepers found something else to complain about.

Architects went through it when CAD replaced hand drafting. “Real architects draw by hand.” There were entire professional organizations that resisted the shift. They said the soul of design was in the hand-drawn line, that something essential was lost when you let software do the calculations. Every new building you’ve walked into in the last 20 years was designed in CAD. And the buildings got more ambitious, not less.

This is what happens every single time.

A tool arrives that makes the work faster and more accessible. The people who built their careers doing it the hard way feel the ground shift. They frame the new tool as cheating because that’s easier than admitting the game changed. They gatekeep for as long as they can. And then the tool becomes standard and nobody remembers the argument.

Writing is going through this right now. AI is the tool. The arguments are identical. “Real writers don’t need it.” “You’re not a real author if AI had any part in it.”

The premise of the objections are the same. It’s a different decade and a different tool, but it’s the same motivation and fear of losing relevance underneath. It’s being driven by the same ego.

But the gatekeepers are going to lose this one too.

What the Loudest Critics Are Actually Protecting

Some numbers first so you understand what’s actually happening in the field.

45% of authors surveyed by BookBub (over 1,200 of them) are already using AI in their workflow. Publishers Weekly found a 31% productivity increase among authors who use it. The global market for AI writing tools is projected to hit $47 billion by 2034.

Here’s the biggest and most telling stat of them all. Of the 45% using AI in their writing process, 74% of them don’t tell their readers about it.

Nearly half of working authors are already doing this. Most of them quietly. (Which tells you everything about the climate right now.) And the reason they’re quiet is because admitting you use AI feels like professional suicide because most of the publishing houses and authors associations are taking a hard line on AI generated content right now. In early 2026, Hachette pulled a contracted horror novel over suspected AI use. The Authors Guild launched a “Human Authored” certification program. And virtue-signaling writers are putting anti-AI badges on their profiles like it’s a political campaign.

The writers who are most vocal about AI-assisted writing being illegitimate are almost always writers who built their audience on craft. They spent years developing their skill and earned their position through a process that was slow, difficult, and exclusive to people who had the time, the discipline, and the cognitive wiring to do it the traditional way.

AI changed the economics of that overnight. And that scares them.

A person who thinks beautifully but writes slowly can now produce finished work. Ministry leaders who preach with fire on Sunday but can’t get a blog post written during the week can finally get their ideas into a format that reaches people beyond the room. People who were locked out of the writing world because their brains didn’t work the “right” way can now participate.

That means more competition and more voices in the space that weren’t there before. And if your entire identity is built on being one of the few people who could do this well, that’s a problem for you.

I clearly have strong feelings about this. Wrapping your competitive anxiety in a moral argument about the “sanctity of writing” is ridiculousness. If your writing is good, AI-assisted writers entering the space shouldn’t threaten you. Period. And if it does threaten you, it’s worth asking yourself if your real value was actually the writing itself or just the fact that fewer people could do it and it made you feel special.

The Blank Page Bottleneck

I’ve talked about how my brain works in previous articles, but here’s a recap in case you missed it. I process ideas fast. Frameworks form fully in my head and I can effortlessly explain them out loud and have someone walk away understanding completely. Everything breaks down because my brain is thinking three ideas ahead of the paragraph I’m trying to finish so I get frustrated and then just stop.

I spent YEARS thinking that was a discipline problem. It wasn’t. And once I figured out why, everything changed.

Writing involves two completely different processes. Getting ideas onto a blank page from nothing (generating) and shaping a draft that already exists into something good (editing). My brain is wired for developing ideas quickly and also for editing. I work best when I can explain via talking (typically in an audio to text transcribing tool). Those ideas are usually delivered in a way I would explain it to someone and not necessarily in linear steps or in a format that makes sense simply converted to writing. The beauty of working with AI is that it can take my ideas and thoughts, organize them and get them onto the page for me. Then I can push back, cut, restructure, and refine until it sounds exactly like me. I’m fast at that part and I genuinely enjoy it. Generating the first draft is where my brain stalls, and it always has. The ideas are there. Getting them into linear written form from scratch without help is the part that breaks down.

AI eliminated the blank page. It gives me a draft to edit. And editing is where my actual skill lives. The blank page was just a bottleneck that had nothing to do with whether I had something worth saying or not.

I’m not the only person this is true for. In a world of increasing neurodiversity, the cut and paste playbook no longer works.

The question the anti-AI crowd never answers is this: if the thinking is original, the voice is real, the editing is rigorous, and the final product is indistinguishable from hand-written work, what exactly was lost by not staring at a blank page for four hours first? Nobody has given me a good answer to that.

The closest thing to a real argument I’ve heard is the idea that the struggle of writing IS the thinking. That when you’re fighting to find the right word or restructuring a paragraph for the fourth time or deleting everything because the argument just isn’t working, that friction is what sharpens the ideas. And that if you bypass all of that with AI, the thinking never fully develops.

And look, I take that seriously. And for people who type a topic into ChatGPT and publish whatever comes back, the criticism holds. That work IS shallow because the thinking was shallow.

But in my process, the thinking happens BEFORE the draft. I know what I’m arguing, who I’m talking to, and what I want the reader to understand before Claude writes a word. Then the draft comes back and I fight with it. Sections get restructured. Entire paragraphs get rejected. I rewrite openings three and four times and add the examples and personal details that make the piece mine. The struggle is still there. I’m just struggling with a draft instead of a blank page. The intellectual work is all still happening. The only thing that changed is that the production bottleneck is gone, and the thinking can actually reach the reader now.

If the struggle itself was the point of writing, we’d still be chiseling stone tablets.

For these people, AI is an accessibility tool. Calling it cheating is the same as calling a wheelchair cheating because the person using it can’t walk as fast as you can. And I don’t hear anyone in the anti-AI writing crowd getting riled up over that! LOL.

See for Yourself

This is from this article. The opening of the Blank Page Bottleneck section.

What Claude wrote:

“I’ve talked about my own brain before. I process ideas fast. Frameworks form fully in my head. I can explain them out loud and have someone walk away understanding completely. But sitting down to write them in linear, long-form prose? My brain is three ideas ahead of the paragraph I’m trying to finish before I even get through the first section.

I spent YEARS thinking that was a discipline problem. It wasn’t. My brain just works faster than the traditional writing format allows for. AI solved that by giving me a structured draft I could edit instead of a blank page I had to fill. Editing is a completely different cognitive task than generating from scratch. My brain can edit all day.”

What I published:

“I’ve talked about how my brain works in previous articles, but here’s a recap in case you missed it. I process ideas fast. Frameworks form fully in my head and I can effortlessly explain them out loud and have someone walk away understanding completely. Everything breaks down because my brain is thinking three ideas ahead of the paragraph I’m trying to finish so I get frustrated and then just stop.

I spent YEARS thinking that was a discipline problem. It wasn’t. And once I figured out why, everything changed.

Writing involves two completely different processes. Getting ideas onto a blank page from nothing (generating) and shaping a draft that already exists into something good (editing). My brain is wired for developing ideas quickly and also for editing. I work best when I can explain via talking (typically in an audio to text transcribing tool). Those ideas are usually delivered in a way I would explain it to someone and not necessarily in linear steps or in a format that makes sense simply converted to writing. The beauty of working with AI is that it can take my ideas and thoughts, organize them and get them onto the page for me.”

Claude’s version was competent but generic. And Claude only produced that because I had already loaded it with my voice sample, my guardrails, and detailed context about who I am and how I write. Without all of that, the output would have been generic slop. My version added the frustration (”so I get frustrated and then just stop”), the audio-to-text detail, the distinction between generating and editing as separate processes, and the explanation of HOW AI actually fits into my workflow. That’s what editing a draft looks like.

“Just Make Sure the Author Is the LLM”

Someone in my comments told me that if I write a book using AI, I should list the LLM as the author and put myself in the acknowledgments section as “prompt creator.” Aren’t keyboard warriors fun?

That comment reveals exactly how little most people understand about what AI-assisted writing actually involves.

A “prompt creator” types “write me a blog post about productivity” and copies what comes back. That’s what this person thinks I do. That’s rubbish. Here’s what actually happens when I write.

Before AI writes a single word for me, I’ve already decided what I’m writing about, who it’s for, what problem I’m addressing, what point I’m making, and what I want the reader to walk away thinking. That’s the work, and AI cannot do it for me. If I skip that step, the output is garbage every time. I’ve proven that to myself more times than I want to admit.

Then there’s the system I built. A voice sample I created by recording myself talking for ten minutes and transcribing it so Claude knows how I actually sound. A guardrails document with 98 specific patterns that AI must avoid. Things like mirrored contrast phrases and performed vulnerability, plus dozens of structural and tonal tells I identified by spending months being annoyed by the output and writing down exactly what was wrong with it. On top of that, there are voice rules that define how my content should sound, a personal lexicon of words and phrases I favor and ones I refuse to use, and a set of required patterns that every piece of content has to meet before it publishes. That system took a long time to build and I refine it constantly.

When Claude gives me a first draft, I read every word and push back on phrasing that doesn’t sound like something I’d say out loud. Entire sections get rejected. Then I add the personal stories, the real examples, and the specific details that only exist in my head. The piece goes through a multi-pass editing process that catches AI patterns at the sentence level. And I read the final version out loud to make sure it sounds like me.

I have 98 editorial rules, a voice training system, a multi-pass audit process, and the final creative authority over every word that publishes under my name. The person who left that comment has a snarky one-liner and zero understanding of the process they’re criticizing.

The Part I Don’t Talk About Enough

My process isn’t perfect. If I only talk about the wins, I’m doing the same thing the AI slop creators do, just from the other direction.

Early on, before the guardrails were tight, I published pieces that had AI fingerprints I didn’t catch. Phrases that were too polished, structures that were too symmetrical. I look back at some of those early articles and I can see it. A reader who knew what to look for could see it too.

That’s actually why the guardrails document exists. Every pattern in that document started as a mistake I made. Something Claude produced that I let through because I was tired, or moving too fast, or hadn’t trained my eye to catch it yet. The 98 patterns didn’t appear overnight. They were built from failure, one cringe at a time. (And there were a LOT of cringes.)

If you’re going to use AI to write, you WILL publish things you wish you hadn’t. You’ll miss patterns. You’ll let something through that doesn’t sound like you and someone will notice. The question is whether you learn from it and tighten your system or just keep pretending the raw output is fine.

I tightened. And I keep tightening. That’s the goal.

What I’m Doing Next

I’m writing a book right now using this exact process. Same Claude projects, same guardrails and voice training, same editorial standards I use for every Substack article. And I’m documenting the whole thing as I go.

I’m putting together a working experience where you write your own book alongside me, using my guardrails and my process, and come out the other side with real progress on your manuscript. More details on this are coming soon.

If that interests you, make sure you’re subscribed so you don’t miss it or send me a DM and let me know you are interested.

Write well. Be honest about how you do it. And stop letting people who’ve never examined their own process tell you yours doesn’t count.

I’ve written about AI-assisted writing before. If you missed those articles:

What AI-Assisted Writing Actually Is (And Why the Debate Around It Is Missing the Point)

I Have Hundreds of Thousands of Words Written and Nothing to Show for It (Until AI)

If you want the system behind everything I write, the AI Writing Guardrails is the exact framework. 98 patterns to avoid, voice rules, a guided builder for creating your own custom rules, and a 5-pass editing process.

Because “make it sound more human” was never going to work.

Self-Guided or Custom Build options available.

guardrails.aiwithleah.com/info

Happy Easter.

All of it, the blog, the podcasts, the products, the systems, started with something dying first.

The Life That Had to End

I ran a multiple seven-figure online coaching business in the wealth consciousness space. From the outside it looked like everything was working. Revenue was high, programs were selling out, and the audience kept growing. I was living the “dream” life, living in a multi-million dollar home on the beach in Bali. I had the clothes, the jewelry, the glam team. I was also deep in the New Age and the occult, operating in frameworks I believed were helping people create financial breakthroughs. And they were, in the short term.

But I was miserable. My marriage was degrading. My children were being raised by nannies. I had gained over forty pounds and was living in chronic pain. I was chasing the next $100,000 month and the next sold-out program while my body and my family were breaking down around me.

I didn’t realize at the time that for all the power and influence I was gaining while operating in the kingdom of darkness, there was a price being paid. That price was my health, my marriage, my presence with my children, and ultimately my own soul.

By the summer of 2023, everything was falling apart. I was separated from my husband, and I had been away from my children for five months in another country chasing the next million dollar business idea. I had chronic inflammation, thyroid disease, and migraines that lasted five to seven days. I remember sitting in an apartment in Porto, Portugal. I had laid out large pieces of white paper on the floor because I like to whiteboard and map things out. I wrote down everything happening in my business, everything happening in my life, all the programs I was preparing to launch, the new business. And as I looked at it, I realized I didn’t want to do any of it.

I fell to my knees and cried out to God. I told Him I didn’t know what I was doing. I told Him I needed help. I surrendered all of it. Even my children, if that was truly what was best for them.

That was the day my old life died.

What Came After

The next day I woke up and heard from the Lord clearly for the first time in a long time. He told me to go carnivore, to completely change my diet, and to stop drinking alcohol entirely. I did. Within two weeks, about seventy-five percent of my pain was gone. The inflammation was leaving my body and my brain. With that physical clarity came spiritual clarity I hadn’t experienced in years.

Over the next year I began reading the Bible seriously. I reconciled my marriage. I moved back in with my children. I walked away from my online coaching business completely because I could no longer participate in New Thought or New Age frameworks. The Lord taught me about provision (I don’t use the word abundance anymore because it belongs to a worldview I left behind). The difference matters.

I shifted into network and affiliate marketing, a model I had always respected because it allowed me to help people build income without spiritual compromise. After another year, I finally gave my life to Jesus. I had resisted Him for a long time, and the moment I stopped, everything changed.

The Resurrection Pattern

I’m telling you this on Easter because the pattern of my life mirrors the pattern we celebrate today. Something had to die for something new to live. The old business, the old frameworks, all of it had to go into the ground.

What came out of that death is everything I’m building now. Especially today, I want you to see what that looks like.

What Grew From the Ashes

The coaching business that died? That’s where AI with Leah came from. I spent years unable to finish long-form content. 120,000 words across two unpublished books sitting in a drawer. God kept those New Age manuscripts from ever reaching the world (thank God), and then gave me a tool that matched how my brain actually works. 50+ articles published in 90 days. A Pinterest account at 200,000 audience. Over 430 Substack subscribers. A book on AI for believers currently being written. All of it built with AI systems, one VA, and a laptop in Bali. After years of finishing nothing, I have more published work than I know what to do with.

The Cul✞ure Cast grew directly out of the years I spent in deception. When you’ve lived inside a false system and watched it operate from the inside, you learn to recognize the mechanisms. Cultural analysis grounded in Scripture, naming what’s actually happening behind the headlines. I couldn’t do that show if I hadn’t lived in the dark first.

My husband came to Christ a few months after I did and launched his own business, Joyful Jesus Apparel (shameless plug because I’m so proud of him)!

https://joyfuljesus.life/

We’re building together now. That floor in Portugal where I surrendered my marriage to God didn’t just save the relationship. It produced a family operating in the same Kingdom for the first time.

And then there’s Faith on the Fringe, which came from the two years I’ve spent studying Scripture obsessively.

Faith on the Fringe

I’m co-hosting this with Malaine Lea Butler and we launch April 27th with four episodes dropping on day one.

The tagline is “Spirit-Filled. Bible-Obsessed. Slightly Unhinged.” Slightly unhinged because we’re going into the parts of Scripture that most churches won’t touch. The Divine Council. Genesis 6. The Nephilim, the Watchers, fallen angels, giants. Pre-flood history. Spiritual warfare with real scriptural backing. Fulfilled eschatology. All of it examined through a Kingdom-forward lens rooted in the finished work of Christ and the authority of believers. We explicitly reject fear-based dispensationalism and rapture theology.

This space in podcasting is overwhelmingly male-dominated. Shows I love and respect like Blurry Creatures and Into the Supernatural are doing incredible work, but there is no female co-hosted show going this deep into supernatural biblical theology with this level of scriptural grounding. Malaine and I looked at that gap and decided to fill it. We bring deep theological roots, humor, and the kind of long-form conversations (one to two hours, with guests) that make you want to open your Bible and start digging. Serious scholarship with real scriptural grounding, and we’re having fun doing it.

It will be available everywhere podcasts are streamed and on YouTube on April 27th. We are also launching a Substack alongside it for deeper conversations and community interaction. I’ll share more about this as we get closer.

What’s Still Coming to Life

The resurrection pattern doesn’t stop. It keeps producing.

I’m working on a series of historical biblical fiction books with two of my besties in Christ. It’s called House of Eden. Think viral BookTok fantasy meets the Bible. Supernatural worldview, and real scriptural foundation, written to bring people to Jesus through story. I am SO excited about this project. It’s a collaboration that feels like exactly the kind of thing God puts together when you stop trying to plan everything yourself.

The AI book is being written. The fiction series is taking shape. Faith on the Fringe launches in three weeks. My husband is building Joyful Jesus Apparel. And I’m answering new assignments to reach people in ways I couldn’t have imagined three years ago when I was on that floor in Portugal.

The resurrection pattern is how God actually works, not just on Easter but in the lives of people who surrender to Him. He lets the old thing die completely and builds something entirely new from the foundation up. And He keeps building.

Three years ago I was a miserable, sick, absent mother running a business built on frameworks from the wrong Kingdom. My family is together now, my marriage is restored, and everything I’m building serves the same mission: stewarding truth, protecting identity, and equipping people with tools and theology that produce real fruit.

I didn’t plan any of this. I surrendered and then I obeyed, one step at a time.

Happy Easter. He is risen. And He’s still making dead things come back to life.