AI agents will not do a huge fraction of human knowledge work in 2025, 2026, or 2027, except in narrow use cases.

Did the forecast come true?

Outcome: Unresolved

We checked, and the evidence so far doesn’t settle it.

Judged on a plain reading. They named no measure, so we used the reading a typical informed reader would take. It is spelled out under “How we judged it”.

Evidence up to 2026-10-28. Checked 2026-10-31.

Why

Marcus forecast that AI agents would not do a huge fraction of human knowledge work in 2025, 2026 or 2027, except in narrow use cases. The window ends on 31 December 2027, after the evaluation date. Under the contract, the forecast cannot be supported before then. A preprint posted on 30 October 2025 reported that the highest-performing agent reached an automation rate of 2.5% on its Remote Labor Index, a benchmark of real-world projects (e1). That is a benchmark result. It does not measure the share of knowledge work agents actually did across the economy. The packet has no such economy-wide measure. So the early-contradiction rule is not met. That rule needs credible measurement showing agents doing a large share of knowledge work broadly.

Caveats

• The packet has no economy-wide measure of the share of knowledge work done by agents.
• Benchmark results are not the same as real-world deployment data.
• 'Huge fraction' and 'narrow use cases' have no defined thresholds.
• A McKinsey survey on agent adoption could not be loaded and was not used.

Evidence (1 item)

• Context only: arXiv (Scale AI / Center for AI Safety), 2025-10-30

  AI agents perform near the floor on RLI, with the highest-performing agent achieving an automation rate of 2.5%.

  Why it matters: In late 2025, agents automated very little end-to-end remote work. That fits the forecast so far, but the window runs to the end of 2027.

  Where

  Abstract

  Publisher

  arXiv (Scale AI / Center for AI Safety)

  Kind of source

  Preprint, not peer reviewed

  About events on

  2025-10-30

  Published

  2025-10-30