AI Futures says 2027 again. The freelance market says not so fast
Under 4%. That is how much probability the AI Futures Project's most aggressive model, less than three months ago, gave to AI speeding up progress by 1.7x by November 2026. On Tuesday the group behind AI 2027 posted an update titled "Looks Like the Intelligence Explosion Will Happen in 2027 After All," and 1.7x is exactly the number they now read out of METR's assessment of Anthropic. All the authors now forecast a Superhuman AI Researcher, an AI good enough that a lab would rather rely on it than on its best human researchers, "likely" in 2027.
The numbers: their simple research-taste Monte Carlo gives a median of July 2027, with 65% of draws by the end of 2027 and 84% by the end of 2028. Conditioned on 1.7x, their full model's median is February 2027. They put a 35 to 65% chance on AI that can automate all white-collar work arriving before 2028, and say 1.7x puts us about five months ahead of the AI 2027 scenario. Two honest notes from the post itself: they rushed it out and say there may be modelling errors, and the 80% attribution behind 1.7x rests on evidence Anthropic redacted from METR's report, which the authors criticize directly. They also call for an industry-wide pause or "pacing," so read this as forecasters who are also advocates.
Now the ground truth from the same day. Scale Labs and CAIS updated the Remote Labor Index, which measures how often an agent completes a real freelance job well enough for a paying client: Claude Fable 5.2 at 41.3%, Opus 5.5 at 21.25%, GPT-6 Astra at 20.83%, per The Decoder. A year ago the leader, Manus, managed 2.5%. Anthropic had cited the 41.3% at launch; this is the benchmark's own leaderboard confirming it.
And then OpenAI and Upwork's study of what actually happened to freelance demand. Across 63 work categories, 43 followed the "common in ChatGPT, lost share on Upwork" pattern and 20 went the other way. OpenAI's own words: ChatGPT use alone does not explain the changes, and the findings "do not point to a simple substitution story." AI-related postings got 45% more diverse, from roughly 15 equally common tasks to about 21, with similar hourly rates but bigger fixed-price budgets and more hours. It's a comparison of shares, not a causal experiment.
So the forecasters say a superhuman researcher by 2027, an agent now finishes four in ten real gigs, and the gig market is reshuffling rather than collapsing. Those aren't contradictions. Capability and adoption run on different clocks, and the gap between them is where most people will actually live for the next two years.
Source: https://blog.aifutures.org/p/q35-timelines-update-looks-like-the , https://openai.com/index/ai-effect-on-freelance-work , https://the-decoder.com/anthropics-fable-5-2-completes-41-percent-of-real-freelance-jobs-up-from-2-5-percent-a-year-ago/
← Back to all articles
The numbers: their simple research-taste Monte Carlo gives a median of July 2027, with 65% of draws by the end of 2027 and 84% by the end of 2028. Conditioned on 1.7x, their full model's median is February 2027. They put a 35 to 65% chance on AI that can automate all white-collar work arriving before 2028, and say 1.7x puts us about five months ahead of the AI 2027 scenario. Two honest notes from the post itself: they rushed it out and say there may be modelling errors, and the 80% attribution behind 1.7x rests on evidence Anthropic redacted from METR's report, which the authors criticize directly. They also call for an industry-wide pause or "pacing," so read this as forecasters who are also advocates.
Now the ground truth from the same day. Scale Labs and CAIS updated the Remote Labor Index, which measures how often an agent completes a real freelance job well enough for a paying client: Claude Fable 5.2 at 41.3%, Opus 5.5 at 21.25%, GPT-6 Astra at 20.83%, per The Decoder. A year ago the leader, Manus, managed 2.5%. Anthropic had cited the 41.3% at launch; this is the benchmark's own leaderboard confirming it.
And then OpenAI and Upwork's study of what actually happened to freelance demand. Across 63 work categories, 43 followed the "common in ChatGPT, lost share on Upwork" pattern and 20 went the other way. OpenAI's own words: ChatGPT use alone does not explain the changes, and the findings "do not point to a simple substitution story." AI-related postings got 45% more diverse, from roughly 15 equally common tasks to about 21, with similar hourly rates but bigger fixed-price budgets and more hours. It's a comparison of shares, not a causal experiment.
So the forecasters say a superhuman researcher by 2027, an agent now finishes four in ten real gigs, and the gig market is reshuffling rather than collapsing. Those aren't contradictions. Capability and adoption run on different clocks, and the gap between them is where most people will actually live for the next two years.
Source: https://blog.aifutures.org/p/q35-timelines-update-looks-like-the , https://openai.com/index/ai-effect-on-freelance-work , https://the-decoder.com/anthropics-fable-5-2-completes-41-percent-of-real-freelance-jobs-up-from-2-5-percent-a-year-ago/
Comments