Issues About
Issue 006 · Friday, June 19, 2026
This Week

The model I wrote about last week went dark in a standoff with Washington. And a quieter study revealed what actually makes someone good with AI. The two stories belong together.

About this Notebook

I'm building toward a for-profit AI venture that funds Gospel work. Each Friday I share the news, tools, and patterns I'm watching, with builders in mind. Join me as we figure out what kind of company God might use to change how Gospel initiatives are funded.

— Tommy Lee, by Friday each week

Last week I wrote to you about Claude Fable 5, the most powerful model Anthropic had ever handed the public. I said the most interesting thing about it was the restraint, the way they built real power and then built in real limits.

Well. This week that same model went dark. The U.S. government ordered Anthropic to block all foreign access, and rather than comply halfway, Anthropic pulled Mythos and Fable offline worldwide. The model I was admiring seven days ago is now sitting behind a wall, caught in a fight between a company and a government.

And underneath all that noise was a much quieter study that I think matters more for people like us. So let me walk you through both, and what I am taking from this strange week.

— Tommy

What's Changing

Building blocks. What is new, and what it could mean for builders thinking about Gospel impact.

Anthropic & the U.S. Government · June 15 to 19

The model I wrote about last week just went dark, caught in a standoff with Washington.

Just days after launching Fable 5 to the public, Anthropic pulled both Fable and the more powerful Mythos offline worldwide. The trigger was a U.S. "export control directive" ordering the company to block access for all non US citizens, even those inside the country, and even its own foreign national employees. Rather than carve out a partial block, Anthropic suspended access for everyone.

The week got messier from there. Bloomberg released the actual letter from Commerce Secretary Howard Lutnick warning against distributing to "foreign persons." The New York Times obtained internal messages where employees felt the lab was being "unfairly targeted." The Washington Post reported the access list had "ballooned," including a South Korea firm with suspected ties to China. The move was traced partly to Amazon, an Anthropic investor, which flagged a Fable vulnerability to officials. By June 19, Anthropic said it was "very confident" access would return "in coming days."

Sources Anthropic · Bloomberg (Lutnick letter) · Washington Post · Semafor
Builder angle

Here is the lesson that hit me hardest. Last week I was admiring the bleeding edge. This week the bleeding edge got switched off by a government overnight. If we are building tools meant to serve a pastor in Jakarta or a founder in Nairobi, we cannot anchor that work to the most powerful model of the moment, because that model is now a strategic national asset that can vanish for reasons that have nothing to do with us. The wiser foundation is a stable, broadly available tier that will reliably be there next year. Build on the ground that does not move.

Anthropic Research · June 17

The quiet study of the week: what makes you good with AI is not coding skill. It is knowing your own field.

While the standoff grabbed the headlines, Anthropic published research analyzing 400,000 Claude Code sessions. The finding: a person's expertise in their own field matters more than their coding ability. Lawyers, managers, and scientists with no coding job title nearly matched professional software engineers on coding tasks, finishing within just seven points of them.

The study also showed users made about 70% of the planning decisions while Claude handled around 80% of execution. And experts pulled far more from each prompt: an expert drew roughly 12 actions and 3,200 words from Claude where a beginner drew about 5 actions and 600 words.

Sources Anthropic Research
Builder angle

This is the most hopeful thing I read all week, and it is the quiet answer to the loud story above. We do not need the most powerful model to do real work. We need people who deeply understand the problem. Our cohort understands ministry, cross cultural work, and the daily reality of marketplace leaders better than any engineer in Silicon Valley. That understanding is now the scarce ingredient, and the AI handles the execution. The barrier was never "can you code." It is "do you know the work." And our people know the work.

Pew Research · June 17

Half of Americans now use AI chatbots. Most of them do not trust it.

Pew surveyed more than 5,000 U.S. adults and found adoption and trust moving in opposite directions. About half of adults now use a chatbot and a quarter do so daily, up from a third of the public in 2024. Yet nearly 40% expect AI to make society worse over the next 20 years, and just 16% think it will make things better. The under 30 crowd uses it the most and trusts it the least.

On platforms, ChatGPT leads at 44% of adults, Gemini at 24%, and Claude at just 6%, even though Anthropic is a constant subject of industry conversation.

Sources Pew Research Center
Builder angle

That trust gap is not a warning for us. It is an opening. People are leaning on AI more while believing in it less. A Gospel centered product, built openly, with clear values and a real relationship behind it, can be the trustworthy option in a sea of tools people use but do not trust. Trust is exactly what the Church has to offer a skeptical world. In a market where most people are uneasy, being the name people trust is a real and lasting advantage.

One Idea Worth Stewarding

Teach our cohort to build their own tools, because expertise now beats coding skill.

Put the two big stories side by side. The frontier model is unstable, switched off by forces outside our control. And the expertise study says the real power was never the model anyway, it was the person who understands the problem. So the most valuable thing we can do is not build one product and hand it down. It is teach our marketplace leaders to build the specific tools their own ministries and businesses need.

Imagine a short workshop at the São Paulo conference in August or the South Africa conference in September: "Build the tool your ministry needs in a weekend." No coding background required. Each leader brings one real problem, learns to describe it clearly, and walks out with something working. A donor tracker. A cohort onboarding flow. A sermon prep helper. Whatever their context actually demands.

Hypothesis worth testing: if we teach 25 cohort leaders to build their own AI tools, we do not get 25 tools. We get 25 builders who each teach others, and the capability multiplies across the network the way the Gospel itself has always spread, person to person.

Tools Worth Knowing

Two tools from this week's reading, with steps you can follow and why they matter for our work.

Google Antigravity Build a Real App

Turn a plain English description into a working, hosted web app with login and a database. This is the "build the tool your ministry needs" idea made real, and it runs on broadly available models, not the frontier tier that just went dark.

  1. Download Google Antigravity from antigravity.google/download, install it, and sign in with your Google account.
  2. Create a new empty folder and open it in Antigravity. A clean folder keeps the agent focused.
  3. Give it a two line prompt describing your app. Example: "Build a cohort tracker with login, members, cities, notes, and a dashboard. Use React, Firebase Auth, and Firestore. Plan first."
  4. Review the plan it produces before letting it build. If it adds features you do not need like billing or teams, tell it to cut scope.
  5. Ask it how to install Firebase, follow the steps, then have it deploy so you get a real, shareable URL.
Why it matters for us: a country director could build a simple cohort dashboard. A finance lead could build a donor tracker. No engineer required, just clear thinking about the actual need. That is exactly what the expertise study says works.
Google Workspace Studio Automate Meeting Prep

Automatically generate a prep brief before every meeting and drop it into a calendar block, so you walk into calls ready instead of scrambling. Useful for anyone running back to back cohort and partner meetings across time zones.

  1. Go to Google Workspace Studio (studio.workspace.google.com) and create a new flow using the "Based on a meeting" trigger.
  2. Set the trigger to run 10 to 15 minutes before the meeting starts.
  3. Add an "Ask Gemini" step that writes a short prep brief: context, open items, a suggested agenda, and questions to ask.
  4. Add a "Create a doc" step using the meeting title plus "Meeting Prep" as the name and the Gemini output as the body.
  5. Add a "Block time" step that places the doc link in a prep block on your calendar. Test it with a fake meeting before turning it on.
Why it matters for us: you lead a lot of meetings across a lot of time zones, from city directors to donors. Automated prep means you show up present and ready for each one. Start with one recurring meeting type, then expand.

Quick Hits

Faster signals from the week, with a builder lens.

Access
OpenRouter launched Fusion, an API that pools several models into one answer. A trio of DeepSeek, Kimi, and Gemini nearly matched Fable's deep research score for about half the cost. Notable timing given the Fable restrictions, and a reminder that frontier level work no longer depends on one model.
Talent
OpenAI hired Noam Shazeer away from Google, the co author of the 2017 "Attention Is All You Need" paper that shaped modern AI. In 2026, top researchers keep landing at either Anthropic or OpenAI. Where talent goes is always a signal.
Surprising
Midjourney revealed a full body medical scanner. The image company built a device that scans your body in 60 seconds using ultrasound, with plans to put them in spas in 2027. A reminder that AI companies are moving into the physical world in unexpected ways.
Reader Story
A retired birder with no coding background built a real web app using AI tools, to help birders find and evaluate rare sightings. He described features in plain English and kept iterating until it worked. This is the expertise study playing out in real life, and it is already here.
Money
OpenAI reported $3.7B in Q1 cash burn against $5.7B in revenue. Both tripled from a year earlier. The frontier labs are spending enormously to stay ahead, which is exactly why a lean, focused, values driven builder can still win in a specific niche.

What I'm Reading

Sources I rely on, with builder lens.

Daily News
The Rundown AI. 5 min/day, 2M readers, free. therundown.ai
Builder
Lenny's Newsletter for product and growth. Latent Space by swyx for technical builder perspective.
Sources
Direct: anthropic.com/news, blog.google, openai.com/news. Skip the commentary, read the source.

I keep sitting with the contrast this week. The most powerful model in the world got switched off by forces no one in ministry can control. And in the same week, the research said the real power was never the model. It was the person who understands the work.

That is strangely freeing. We are not waiting on the frontier. We are not at the mercy of an export order. The thing we already have, deep understanding of the people and places God has called us to, is the very thing that matters most. So let us build on that. If this issue gave you something to chew on, hit reply and tell me. I read every one.

Tommy
TLEE's AI Notebook ● Issue 006 ● June 19, 2026