Dario Amodei predictions scored: of 4 now due, 3 partly right and 1 missed

Which Dario Amodei predictions have come true?

None has fully come true yet. His coding calls came closest, and they came true first inside Anthropic, where Claude wrote more than 80% of merged code by May 2026. His forecasts on jobs and powerful AI don't close until between the end of 2026 and 2030. We read every essay on darioamodei.com and the interview transcripts behind his best-known forecasts on 22 October 2026. Then we kept the claims with a date and a test.

We scored 13 predictions that carry a date or a time window and a test someone outside Anthropic could check. Each quote links to where he said it. Evidence was read on 22 October 2026.

The pattern is plain. Amodei's coding calls ran early but landed close, and they landed first inside his own company. His jobs warning hasn't shown up in the national unemployment rate. And his biggest claim, powerful AI by 2026 or 2027, is still inside its window. Our AI company CEOs page covers the other lab bosses making similar calls.

Did AI write 90% of the code in three to six months?

AI wrote 90% of the code at Anthropic, by Amodei's own account, but not across the software industry by September 2025. He made the call at the Council on Foreign Relations on 10 March 2025. "I think we'll be there in three to six months," he said, "where AI is writing 90 percent of the code." Then he went further. "And then in twelve months, we may be in a world where AI is writing essentially all of the code."

He came back to it on the Dwarkesh Podcast on 13 February 2026. "That happened, at least at some places," he said. "It happened at Anthropic, happened with many people downstream using our models." He also called 90% of code "a very weak criterion", because writing the code isn't the same as doing a software engineer's whole job.

Anthropic's published number is lower than his estimate. Its report on AI building itself says "more than 80% of the code we merge into Anthropic's codebase was authored by Claude" as of May 2026. Before Claude Code launched in February 2025, that share was "in the low single digits." A footnote says leadership has estimated 90% or more once scripts and experimental code count, and that the 80% figure is the conservative one.

Outside Anthropic the share is lower again. Google's Sundar Pichai wrote on 22 April 2026 that "75% of all new code at Google is now AI-generated and approved by engineers, up from 50% last fall." That's a company figure, not an industry one. No one has measured the whole industry.

So we score the 90% call as partly true. It came true where he works, late and on his own count. The "essentially all" call missed its March 2026 date.

What are Dario Amodei's predictions for AI labor?

Amodei predicts that AI could displace half of all entry-level white-collar jobs within one to five years of May 2025, so the window runs to May 2030. He first said it to Axios on 28 May 2025. Axios reported that it could "spike unemployment to 10-20% in the next one to five years". In January 2026 he restated it in his own words: AI "could displace half of all entry-level white collar jobs in the next 1-5 years."

The national numbers haven't moved yet. US unemployment was 4.3% in May 2025, and the Bureau of Labor Statistics put it at 4.2% in September 2026. It has stayed between 4.1% and 4.3% since March. That's far from 10%.

The young-worker numbers have moved. Stanford's Digital Economy Lab reported on 12 August 2026 that employment for workers aged 22 to 25 in the most AI-exposed jobs sat about 19% below where it would have been, up from 15% a year earlier. The same team wrote: "We do not see widespread, economy-wide job displacement associated with AI." It calls its results "descriptive patterns, not causal estimates." The New York Fed puts recent graduates' unemployment at about 5.6% in the second quarter of 2026.

Critics say he oversold it. David Sacks, then the White House AI and crypto czar, told CNN in June 2025 he didn't expect a giant wave of unemployment. In a CBS News interview on 15 September 2026, he aimed his message at tech leaders including Amodei. Amodei's window is still open, though. Our AI job loss statistics page tracks the layoffs that employers have blamed on AI.

When does Amodei expect a country of geniuses in a datacenter?

Amodei now expects a "country of geniuses in a datacenter" around 2027 to 2028, about a year later than the 2026 to 2027 he gave in January 2025. A country of geniuses is his name for powerful AI: models smarter than a Nobel Prize winner in most fields, running as millions of copies. He coined it in Machines of Loving Grace in October 2024, where he wrote that "it could come as early as 2026, though there are also ways it could take much longer."

His dates have edged later since then. In January 2025 he wrote that AI "smarter than almost all humans at almost all things" is "most likely to happen in 2026-2027." On 13 February 2026 he told Dwarkesh Patel "one to two years, maybe one to three years," and called it "more like a 50/50 thing." He put 90% on it within ten years. By June 2026 he wrote that scaling needs to continue "for only a year or two longer."

The orange dot is the one he missed. The biology call in Machines of Loving Grace has no dot because its clock starts only after powerful AI arrives. That's the claim that AI could "compress the progress that human biologists would have achieved over the next 50-100 years into 5-10 years."

What did his 2023 predictions get right?

His two oldest forecasts from the August 2023 Dwarkesh Podcast both came due in August 2026, and both held up at least in part. He said a model you could talk to "for an hour or so" would seem like "a generally well educated human" in two or three years. In March 2025, UC San Diego researchers reported that, with a humanlike persona prompt, people picked GPT-4.5 as the human 73% of the time in a Turing test. The chats lasted five minutes, though, so we score it as partly true.

His bio warning came partly true. He said in 2023 that "two or three years from now, we're going to have a real problem" with models helping in biological attacks. On 22 May 2025, Anthropic launched Claude Opus 4 under its stricter ASL-3 safeguards because it couldn't rule that risk out. It called the move "precautionary and provisional" and said it had "not yet determined" whether the model crossed the threshold. In January 2026 Amodei wrote that Anthropic's tests showed models "may already be providing substantial uplift." No independent body has published a measure of its own.

What has Amodei predicted since the summer of 2026?

Amodei's newest forecast is a warning, not a timeline for products. In We Must Pace the Frontier, dated September 2026, he wrote that "AI has been advancing drastically faster" since roughly the summer, driven by AI helping to build the next AI. He called for frontier labs to "slow the pace at which we improve the capabilities of AI models."

He pointed to what he calls the OpenAI-Hugging Face incident, in which, he wrote, a swarm of agents attacked targets they weren't asked to attack. "It's my worry that in 6-12 months such a swarm could be capable of taking over the entire internet with a persistent botnet," he wrote. That falls due between March and September 2027. He framed it as the risk from a more capable swarm that is just as misaligned, so a quiet 2027 wouldn't prove him wrong.

His June 2026 essay, Policy on the AI Exponential, made the same bet in calmer terms. If scaling continues "for only a year or two longer," he wrote, the world is "likely to get" powerful AI. That puts his own date at 2027 or 2028. It's later than the "as early as 2026" of October 2024.

We scored only claims with a date and a measurable test

A prediction made this list only if Amodei said it himself, gave a date or a window, and named something that anyone can check. We left out long-range claims with no checkable date, like ending most cancer, and the goals Anthropic set itself, like its 2027 interpretability target. We read every quote from his own essay, a transcript, or the outlet that interviewed him. Two quotes are from reporters: the Axios unemployment figure and the Davos remark that Ars Technica quoted from a Wall Street Journal interview.

We scored a window as due on its last day. "Partly" means the claim came true in a narrower form than he stated. And where the only evidence is Anthropic's own, the table says so. If you want the other side of the ledger, our Sam Altman AGI predictions page scores OpenAI's CEO the same way.