Anthropic Just Admitted AI Will Displace Jobs — Then Created a $85K Fellowship to Help You Adapt.
Dario Amodei's 15,000-word essay said AI may cause permanent job displacement. The press release that followed the same day created 1,000 paid roles for people who can help organizations adopt it. The skills those roles need are learnable now.
By Forge Team
The most practical thing Dario Amodei said on June 10 wasn't buried in his 15,000-word essay on AI regulation. It was in the press release that went out the same afternoon: Anthropic is hiring 1,000 people at $85K a year — no degree required — to help organizations adopt AI. The CEO who built Claude just spent nine figures creating a new job title.
What he said, and what he funded
On June 10, Amodei published "Policy on the AI Exponential" — calling for FAA-style government oversight of frontier AI, the authority to block unsafe releases, and universal basic income funded by taxing AI companies if job displacement proves permanent. He called widespread job loss "undesirable and dangerous" while acknowledging it may be "an intrinsic property of the technology." (Amodei, "Policy on the AI Exponential," June 10, 2026.)
The same day, Anthropic committed $350M to the problem it's helping create: $200M for an Economic Futures Research Fund studying AI's labor-market effects, and $150M for Claude Corps — a fellowship placing 1,000 early-career professionals inside 400+ nonprofits at $85K a year to help those organizations adopt AI. No degree required. Applications open now. First cohort starts October 2026. (TechTimes/Fortune, June 10–11, 2026.)
What this tells you to do differently
The Claude Corps isn't hiring coders. The job is AI-fluent generalist: someone who can identify a specific organizational problem, decide whether AI is the right tool, and build a workflow a non-technical team can run. Those three steps are learnable skills, not credentials.
The same week, Demis Hassabis, CEO of Google DeepMind, pushed back on the other half of the debate. Companies using AI as justification for layoffs are showing "a clear lack of imagination" — a 3-4x productivity gain should fund growth, not headcount cuts. (Metaintro, June 8, 2026.) A Hacker News piece titled "CEOs who think AI replaces employees are just bad CEOs" hit 834 points the next day.
The two positions sound like opposites. They're not. Both require the same thing from you: the ability to actually use AI on real work.
Leila: the program coordinator who didn't need the fellowship
Leila is a program coordinator at a 15-person nonprofit providing legal aid to asylum seekers in the UK. She spends 90 minutes every Friday compiling a weekly intake report — cases opened by category, documentation outstanding, anything flagged for upcoming court dates.
She tried AI for one section last month and stopped when the output didn't match the format her director expects. The problem wasn't her prompt. She hadn't framed the task before writing one: what goes in (the intake spreadsheet), what format comes out (the five-section template her director uses), and what she always checks herself before it goes anywhere (any case with a court date in the next 30 days). That's a task frame. It takes ten minutes to write once. It's also the first thing a Claude Corps fellow would build on day one.
Write the task frame for one repeating piece of work — input, output format, and what you verify yourself before it leaves your desk.
James: the ops director with a board presentation
James is operations director at a 180-person property management company. His board wants an "AI strategy" next month and his instinct is to frame it as a headcount question: could AI reduce admin costs?
The more useful question is different. He has seven admin staff each doing three things: entering data into their property management system, generating monthly owner reports, and answering routine tenant queries by email. For each task, one question matters: does this have clear enough inputs and output format that AI can handle the first draft — and what happens if it's wrong?
Cost of error matters. Routine tenant queries (low stakes, clear format): worth trialing. Monthly owner reports (medium stakes, known format, reviewed before sending): worth trialing. Data entry into a live property system (immediate downstream consequences if wrong): not yet. That sort takes one meeting. It's also six months of a real rollout rather than a deck full of percentages.
Sort a set of tasks by how much human judgment each one needs before AI can handle it — and notice where you keep second-guessing yourself.
The job description you already qualify for
The professional Anthropic is paying $85K to find can assess a task, decide what AI should handle, and design a workflow that runs reliably. That's not a new kind of credential. It's practice.
Work through the first decision in any AI task — whether to use it at all, and what would make that call obvious sooner.
Like this post?
Get the next one in your inbox. Practical AI skills, no filler.