AI ‘co-workers’ are training in billion-dollar fake offices
Anthropic and OpenAI are pouring billions into teaching AI how to use the same apps you do — Salesforce, Zendesk, even Excel — by letting models practice inside simulated workplaces.
The setup:
Companies like Turing build reinforcement learning gyms — cloned apps filled with mock tasks.
AI models “play office” by sending fake Calendly invites, updating Salesforce leads, or tweaking Excel models, while human experts check if they did it right.
Contractors now range from NASA scientists to radiologists, paid up to $250/hour to teach AI how real professionals work.
Why it matters: Training data scraped from the internet can only take models so far. RL gyms let AI learn workflows step by step — a path to turning chatbots into actual knowledge workers. One OpenAI exec privately said they expect the “entire economy” to become an RL machine.
The cost: Anthropic may spend $1B on RL gyms this year alone, while OpenAI projects $8B in data costs by 2030. Rival startups like Scale and Surge are racing to cash in by selling simulated apps and expert labor.
The bottom line: The new frontier of AI isn’t just smarter models — it’s AI that can navigate the same software you use at work. If it works, “virtual collaborators” won’t just brainstorm, they’ll do your job beside you.
Supported by Metronome
Flat-rate subscriptions can’t keep up with AI. As teams roll out AI-powered features, companies like Miro and Airtable are moving to seat-based credit pools, where each seat adds credits into a shared pool that all users can draw from. This model balances predictability with flexibility, lowers adoption barriers, and supports rapid pricing iteration. It can also raise challenges around forecasting, revenue leakage, and customer visibility. This post breaks down why pooled credits are gaining traction, what to watch out for, and how modern infrastructure makes the model viable at scale.
Nvidia CEO ‘disappointed’ after China bans its AI chips
China has reportedly told ByteDance, Alibaba, and others not to buy Nvidia’s AI chips — just weeks after a U.S. deal allowed limited sales. Regulators also launched an antitrust probe into Nvidia’s Mellanox acquisition.
CEO Jensen Huang called the move “disappointing” but tied to bigger U.S.-China tensions, adding that Nvidia can only “be in service of a market if the country wants us to be.”
Why it matters: Nvidia has told analysts to exclude China from forecasts, but losing access to one of the world’s biggest AI markets is a blow. The company is leaning on other regions, announcing £11B of UK AI investment this week.
The bottom line: Nvidia’s China business is still hostage to geopolitics.
Most Americans don’t want AI everywhere
A new Pew survey shows Americans are open to AI in some areas but want clear limits in others. They’re fine with it forecasting weather (74%) or detecting financial crime (70%), but overwhelmingly reject it in relationships, religion, and politics.
The numbers:
66% say AI shouldn’t judge romantic compatibility.
60% don’t want AI involved in governing.
Over half worry AI will erode creativity and relationships.
The big picture: Only 25% now say AI’s benefits are “high,” down from earlier years. Yet most admit they struggle to tell human work from AI output, even as they call it “extremely important” to know.
Why it matters: As AI tools creep into therapy, companionship, and creative work, Americans are signaling they’re not comfortable handing over values-based decisions. Regulators are paying attention too — the FTC is investigating risks to kids.
The bottom line: Efficiency is in. Judgment is out.
Tech pours billions into UK AI as Trump visits
U.S. tech giants are dropping major AI investments in the UK during President Trump’s state visit — moves the White House is framing as proof of U.S. tech dominance.
The deals:
Microsoft: $30B through 2028, including $15B for AI infrastructure and the UK’s largest planned supercomputer.
Nvidia, OpenAI, Nscale: “Stargate UK” data center project, plus bets on quantum computing and robotics training.
CoreWeave: $2B with Nvidia and DataVita to deploy advanced chips in Scotland.
Why it matters: Washington gets its victory lap, while Big Tech locks in long-term control of one of the world’s key AI markets.
The bottom line: Politics and business align — Silicon Valley gets the UK, and Trump gets the headlines.