Scores: Scale Labs leaderboard, read November 11, 2026. Posting dates as reported by CellCog. Each point is a new record. Full numbers lower down.
By Scott Covert · November 11, 2026
Since ChatGPT landed in late 2022, economists have read the same payroll data and come out with different answers.
In May 2025, Anthropic CEO Dario Amodei told Axios that AI could wipe out half of entry-level white-collar jobs and push unemployment to 10–20% within one to five years.
The overall unemployment numbers barely moved.
Where the research has started to line up is narrower: the youngest workers.
Stanford's Digital Economy Lab, working from ADP payroll records, finds that in the jobs AI can do most easily, the number of 22-to-25-year-olds working is 19% lower than it would be if those jobs had kept pace with jobs AI touches less. Older, experienced workers in the same jobs show no such drop.
The Dallas Fed found the drop comes from fewer people being hired in, not from layoffs. Even if every one of those missing jobs became unemployment, the overall unemployment rate would rise by only about one-tenth of one percentage point.
Even that signal is tangled. Oxford Economics found AI named in nearly 55,000 US job cuts in the first 11 months of 2025. That's only about 4 or 5 out of every 100 cuts. Oxford also said some firms dress up cuts from past overhiring as a technology story.
No study I've found cleanly separates the AI part from the older pressures:
Those squeeze junior hiring with or without AI.
Most AI job studies score how many of a job's tasks a model could touch. The Remote Labor Index, built by Scale AI and the Center for AI Safety, grades finished work instead.
It gives AI agents 240 real projects taken from paid freelance work. Expert reviewers grade each result the way a client would: accept it or send it back.
At launch in October 2025, the best agent got 2 or 3 out of every 100 projects accepted (2.5%). By March 2026 the record was still only about 4 out of 100.
Then it moved. Fable 5 finished about 16 out of 100 in July, and Fable 5.2 finished 41 out of 100 in November.
For years tech leaders said this curve was coming, and until this summer the measured numbers barely moved. Now they have.
41% of these projects is not 41% of anyone's job.
For three years the job-loss argument was “AI takes your jobs” predictions against payroll data, and the payroll data kept winning.
The entry-level squeeze is the first real crack, and it's still mixed up with ordinary corporate cost-cutting.
The Remote Labor Index is different. It grades finished work a client would pay for, and it went from almost nothing to four in ten in about a year.
Whether it keeps climbing at that pace is open. If it does, I'd expect the payroll numbers to follow, and the entry-level squeeze to turn out to be the start, not the whole story.
The next few leaderboard entries are the ones to watch.
| Posted | Agent | Score |
|---|---|---|
| Oct 27, 2025 | Manus 1.0 | 2.50% |
| Jan 9, 2026 | Claude Opus 4.5 (thinking) | 3.75% |
| Mar 5, 2026 | Claude Opus 4.6 (CoWork) | 4.17% |
| Jul 1, 2026 | Fable 5 | 15.83% |
| Sep 17, 2026 | GPT-6 Astra | 20.83% |
| Oct 7, 2026 | Opus 5.5 | 21.25% |
| Nov 9, 2026 | Fable 5.2 | 41.30% |
The leaderboard shows scores only. Posting dates are as reported by CellCog.
For the wider data on AI and employment, see AI vs Jobs: What the Data Actually Shows.