Welcome back! The wall between the AI labs and the US government is getting thin. OpenAI hired the man who helped write Trump's AI plan, while every major lab now hands its models to Washington for testing before they ship. On the product side, OpenAI and Anthropic shipped rival upgrades to their coding agents within an hour of each other last night. Busy day, so let's get into it.
In today's Generative AI Newsletter:
Codex: How do you teach an AI agent a task without writing a single prompt?
Washington: What does it mean that every AI lab now lets the government test its models before launch?
OpenAI: Why did OpenAI put Trump's AI-plan author on payroll?
Claude Code: What happens when your coding session turns into a page your whole team can watch?
OpenAI added a feature to Codex that retires the prompt for repetitive work. It's called Record and Replay.
You do a boring task once on your Mac, like filing an expense report, while Codex watches over your shoulder.
It turns that demo into a skill it can rerun later with new dates or files, and you can open the skill and edit it.
This is a different way to teach an agent.
Instead of describing a chore in a paragraph and hoping the model gets it, you show it once and it keeps the recipe. For anything that's easier to demonstrate than explain, that's the way to do it.
The fine print keeps it from being a today thing for most people. It runs only on macOS, you need OpenAI's Computer Use turned on, and it isn't in the EU yet.
The direction is the real headline though. The prompt was always a workaround for not being able to show the computer what you mean.
Special highlight from our network
Anthropic ships new Claude capabilities almost every week. Skills, Connectors, Cowork, vibe coding. Most professionals are still using Claude for one-off chats, missing the modes that actually change how work gets done.
Outskill's 2-Day Claude AI Mastery Workshop is a live, end-to-end walkthrough taught by mentors who condensed 800+ hours of research and real-world testing into one weekend.
You'll walk away knowing:
How to use Claude's three modes: Chat, Cowork, and Code
How to set up Skills and Connectors that automate your Notion, files, and desktop
How to vibe code working apps and dashboards alongside 10+ tools that pair with Claude
Saturday and Sunday, 10 AM to 7 PM EST. Free for the next 48 hours.
The investors and engineers who backed it more than two decades ago, when three Falcon 1 rockets had failed in a row and the company was nearly out of money. They own a piece of one of the most valuable companies on the planet today.
For all the years SpaceX stayed private, that piece was a club only venture capitalists and insiders could join. The public got its first shot last week.
Equity crowdfunding changed that. You can now own a stake in a private AI company before it ever lists. That company is GenAI Works.
Real revenue, up 2.5x YoY in 2024-2025.
Nvidia, Google, Oracle, IBM and 300-plus partners behind it.
Built on the world's largest AI community, 14 million people.
GenAI Academy launched and hit 30,000 students in its first month.
Our last raise pulled in $520K, and those backers got the first seats.
You can take one of the next seats from $1,000. Lock in before July 1 and Phase 1 adds bonus shares, 5% at $1,000 rising to 25% at $25,000, with an extra 10% if you've backed us before. After July 1 that same check buys fewer shares.
12 days left at this price.
In making an investment decision, investors must rely on their own examination of the issuer and the terms of the offering, including the merits and risks involved. Genai Works, Inc. has filed a Form C with the Securities and Exchange Commission in connection with its offering, a copy of which may be obtained here.
OpenAI is hiring the man who helped write Trump's AI playbook.
Dean Ball, a primary author of the White House's AI Action Plan last year, starts July 6 leading a new OpenAI team called Strategic Futures.
The job is to shape frontier AI policy, both in Washington and inside the company.
The same administration that switched off Anthropic's two best models last week is the one whose former policy author OpenAI put on payroll.
OpenAI is an insider now, and its rival, Anthropic, keeps getting squeezed.
Ball's team will work the heavy questions, from catastrophic risk and AI that improves itself to the hit to jobs and how the labs deal with governments.
He argues the labs, not regulators, will end up making the big AI governance calls, so what happens inside a company like OpenAI matters more than people think.
This is the second marquee hire in a week, after OpenAI poached Gemini's co-lead from Google. While Anthropic spent the week fighting Washington, OpenAI spent it stacking talent and political cover ahead of its IPO.
It isn't only Anthropic. After the government pulled Mythos last week, the part worth knowing is that every major US lab now hands new models to Washington before release.
Google, Microsoft, xAI, OpenAI and Anthropic have all agreed to let the Commerce Department's Center for AI Standards and Innovation evaluate their models for security risks.
The reviews are voluntary for now, aimed at things like cybersecurity and bioweapons. The Anthropic fight is what's pushing them toward mandatory.
The White House used a single jailbreak to force Anthropic's Mythos offline, and it's weighing a formal process where a model needs a government sign-off before it can launch.
The bar the White House keeps reaching for is a model that can't be jailbroken at all, and researchers across the field say that cannot be guaranteed with today's AI.
So you get a gate no model fully passes, run by the same officials who decide which company is in favor this week. Anthropic found that out the hard way. OpenAI hired the guy who helped write the rules.
Claude Code now makes Artifacts. Anthropic shipped its own coding-agent upgrade the same night, within an hour of OpenAI's.
While Claude is working, you can ask it to turn what it's building into an interactive page, like a walkthrough of a pull request or a live project dashboard, and get a private link to send your team.
The page is built from the whole session, the code, the connected tools and the chat, and it keeps updating at the same link as the work continues.
A teammate opens it later and sees the current state with version history. It's in beta on Team and Enterprise plans.
The two launches landed the same night and aim at the same problem from opposite ends.
OpenAI is teaching the agent to learn your tasks, while Anthropic is turning the agent's work into a page your team can read.
Whoever makes the coding agent feel less like a black box keeps the enterprise seats.
Actively gives every sales account its own AI agent. Instead of one rep reactively juggling 200 accounts, you get 200 agents, each one researching its account around the clock, flagging risks, drafting outreach and telling the human what to do next. It's the agent-for-everything idea pointed straight at revenue teams.
How it works:
It runs a dedicated agent on every account across the funnel, for SDRs, account execs and leaders, not one narrow use case.
The agents work from one shared, always-updated picture of your accounts, so reps stop stitching context across a dozen tools.
Teams at Ramp and Samsara use it to cover far more accounts than their headcount would allow.
It's enterprise only, so there's no free sign-up. You book a demo and their team helps you deploy it.
Who it's for: revenue and go-to-market leaders who want every account worked, not only the loud ones.
Accenture had its worst day in years: The stock fell about 18% to its lowest price since 2016, after it cut its outlook and warned that AI is starting to eat the consulting work that pays its bills.
Anthropic is funding 1,000 AI jobs: It's putting $150 million into a Claude Corps fellowship that places young people in full-time roles at nonprofits adopting AI.
Amazon wants to sell chips, not only cloud: It's reportedly in talks to sell its own AI chips to other companies, taking aim at Nvidia.
The meter is coming for AI: OpenAI, Anthropic and others are pushing customers toward pay-per-use pricing as the cost of running models climbs.
Learn more about AI from the experts building it
📸 Follow us on Instagram for fast, visual AI updates in 30 seconds.
🐝 Subscribe to our Atlas newsletter — trusted by 3M+ subscribers — to stay ahead of AI news across tech, education, and business.
📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators.
🐦 Follow us on X for breaking AI news and real-time industry updates.
🧠 Learn how to build your next AI application with practical resources and expert guidance.
🚀 Explore investment opportunities in the future of AI and join our community-backed growth journey.
