In partnership with
Welcome to OHMO AI. The person who wrote OpenAI's safety reports has quit, and the reason he gives is that the culture is broken. Meta is handing out the code to connect its Muse agent to hardware you build yourself, and giving 5,000 of its own gadget away to subscribers.
Capcom says where AI belongs in making its games and where it doesn't, and Anthropic is paying to train the people who make AI work inside big companies.
In today's newsletter:
The OpenAI employee who wrote its safety reports quits and calls the culture broken
Meta releases the code to build your own Muse gadget, and gives 5,000 away
Capcom will use AI to make games faster but not to make what's in them
Anthropic commits $100 million to train 10,000 AI engineers
LATEST IN AI
OPENAI
OpenAI's safety report writer quits and says the culture is broken
Image source: OHMO AI
OHMO AI: David Robinson, the OpenAI employee who led the writing of the safety reports that go out with its major product launches, has resigned. He gave his reasons in an essay in The Atlantic, and the short version is that he thinks the company's "culture is broken."
The details:
He'd been there three-and-a-half years, which by his own account makes him "among the longest-tenured employees at the company."
He wrote the piece after his departure, once he was already out of the building.
Engadget's report describes him as another OpenAI employee tasked with ensuring the safety of its models who has left.
Why it matters: Three-and-a-half years is enough to be among the longest-serving people at OpenAI, which says how few people there remember the launches those reports went out with. Those reports were his to write, and the public verdict on the culture behind them is now his too. Whoever writes the next one starts with less history behind them than he had.
PRESENTED BY ALIGNED
One workspace where buyers, sellers, and AI close deals together
Instead of running your deals across scattered email threads and forgotten attachments, Aligned brings everything into one shared deal workspace.
Buyers get a single place to evaluate, loop in their team, and say yes. Sellers get visibility into what is actually happening when they're not in the room and AI Deal Insights that help you act accordingly, before it's too late.
Best part? Your first deal room takes only minutes to stand up.
META
Meta hands out the code to build your own Muse gadget
Image source: OHMO AI
OHMO AI: Meta is providing open-source firmware, the low-level code a device runs on, and a Linux software development kit so that anyone can build their own hardware that connects to Muse, its AI agent, along with a few project ideas to start on. A cheap hobbyist board like a Raspberry Pi or an ESP32 will do, and Meta built an example of its own, a gadget called Muse Home Link that lets the agent control things around the house.
The details:
Nat Friedman, head of product at Meta's Superintelligence Labs, said in a post on X that Meta made 5,000 Home Links and is giving them away free to Muse subscribers while supplies last.
With a Home Link, community-built skills let the agent "turn on a light, control your TV, send a document to a printer, and more, depending on your setup," in Meta's words.
Meta's own pitch is to wire Muse up to whatever displays, buttons, sensors and actuators you have lying around on your workbench.
Meta describes the devices as "open source devices you build yourself," which puts the building on the subscriber, not the company.
Why it matters: The price here moved from money to effort. Meta isn't selling a box, it's handing over the firmware and expecting you to supply the Raspberry Pi and the evening, and the 5,000 free Home Links are the exception that goes only to people already paying for Muse. Anyone without a workbench is still waiting for a product.
CAPCOM
Capcom puts AI into making its games and keeps it out of them
Image source: OHMO AI
OHMO AI: Capcom has set out plans to bring AI into its "development workflows" through a major upgrade of RE Engine, the engine its games are built on. The company's line is that AI goes to work on making games more efficiently, and it has said before that AI-generated assets won't go into the games themselves.
The details:
Programmer Satoshi Ishida presented the plan at the Capcom Open Conference RE: 2026, under a project called REX that's meant to eventually turn RE Engine into a "game engine of the AI generation."
The earlier promise came in the Q&A at a shareholders' meeting, where Capcom said it would explore AI to assist with graphics, sound and programming but would not use AI-generated content in its games.
RE Engine is Capcom's own proprietary technology, and REX is billed as a major improvement project for it, not a new engine.
Why it matters: The line Capcom is drawing runs between AI that helps its programmers make the graphics and graphics an AI made, and it sits inside the build pipeline where only Capcom can see it. A player gets the promise and no way to check it on screen. The saved hours stay Capcom's, and the reassurance is all the buyer gets.
PRESENTED BY ATLASSIAN
State of Product reveals what AI still hasn’t solved
80% of product professionals say AI helps them ship faster, but customers aren’t seeing value any sooner. Atlassian’s State of Product 2027 explores the gains AI is delivering and the challenges that remain, from decision-making that hasn’t kept pace to gut instinct overriding customer evidence.
ANTHROPIC
Anthropic puts $100 million into training 10,000 AI engineers
Image source: OHMO AI
OHMO AI: Anthropic is spending $100 million on an academy to train the people who put AI to work inside other companies. The Claude Frontier Academy aims to turn out 10,000 "frontier deployed engineers" by the end of 2027, drawn from firms in Anthropic's Claude Partner Network, and the company says they'll be trained to the same standard of skills as its own engineers.
The details:
The first cohorts come from Accenture, Bain, Capgemini, Commonwealth Bank of Australia, Deloitte, McKinsey, Morgan Stanley and Novo Nordisk, among others.
Steve Corfield, Anthropic's global head of business development and partnerships, said the academy "trains people the way our own engineers learn," and that graduates should "set the standard for how AI gets built inside a business."
Anthropic announced it in a press release on Friday and frames the problem it's aimed at as the enterprise AI talent gap.
Anthropic bills the academy as the first of its kind.
Why it matters: Not a scholarship, a channel. The money buys reach as much as relief: each graduate carries one vendor's method into every client they are posted to next, and no rival's gets taught alongside it. An engineer outside that network fits the shortage but not the description.
AI GUIDES
Get a cited research report that only draws on the sources you trust
Ask ChatGPT a big question and it pulls from whatever the web turns up. Run Deep research with your own list of sites and you get a report, with citations, built from the sources you would have checked yourself. It costs one menu tap before you start: open Sites, then Manage sites, add the websites you trust and choose whether to restrict to them or just prioritise them.
Copy this, swap in the bracketed parts, and paste it in.
/Deepresearch
Research this question: [the question you want answered, in one or two sentences].
Use only the sites I have set under Manage sites: [site 1], [site 2], [site 3]. If something is not covered there, say so rather than filling the gap from elsewhere.
Before you start, show me your research plan so I can edit it. In the final report, give me [a short summary / a comparison table / a list of options] followed by the detail, and cite the exact page for every claim.The site list fixes where the answers can come from, and the plan check lets you steer the search before it runs. You can also interrupt it mid-run to narrow the focus or change which sources it is allowed to use.
COMMUNITY
Featured Reader
Every newsletter, we feature how a reader is using AI to work smarter, save time, or make life easier.
Today's reader is Stefan K. from Karlsruhe, Germany:
“The Steuerbescheid used to go straight into the folder. I read the final number and if it was not shocking I trusted the Finanzamt had done the rest.
Now both files go into a ChatGPT Project called Finanzamt. The ELSTER return as a PDF and the Bescheid as a PDF. I ask it to compare every line and list each figure that differs with the euro amount and which document it came from. This year it found my Werbungskosten were 1,400 euros lower in the Bescheid than in what I filed. One receipt line had vanished. I sent the Einspruch within the month and got 380 euros back. It cannot tell me if the Finanzamt is right. Only that the numbers do not match. Every Bescheid gets compared now before it gets filed.”
Want to be in the next newsletter? Tell us how you use AI here.
AI Meme
Go Deeper
AI Guides: Practical AI guides, prompts and tools.
Instagram: The AI page they pray you’ll never find.
Facebook: For the ones who never left Facebook.
