Artificial Intelligence · Projected milestone

AI research run by agent swarms

Estimated year~2028

75% by 2030 · including earlier years

The year is an editorial timing estimate. The percentage applies to occurrence by the stated deadline. These estimates are not calibrated odds.

Swarms of AI researchers run most experiments at a frontier AI lab, with people setting goals and checking results.

Explore this milestone on the interactive map

What the forecast means

Milestone stage: deployment

partially-demonstrated

Agents contribute substantive research workflow steps today; majority end-to-end execution is not independently established.

Completion criteria: For a disclosed three-month period, agents execute more than half of a representative experiment portfolio, including design, implementation, running and analysis; independently audit sampling and report human corrections. This does not require fully autonomous lab governance.

This page describes a forecast, rather than an achieved result. Read the methodology and assumptions.

Built on

AI research partnersRecorded · 2025influences
Agent swarms run whole projectsForecast · ~2027influences
Gigawatt AI campusesForecast · ~2027enables

Led to

Automated alignment researchersForecast · ~2029influences
Million-agent swarmsForecast · ~2030enables
Controlled recursive self-improvementForecast · ~2032enables
Superhuman AI scientistsForecast · ~2032enables

Sources & evidence

Research acceleration: The view inside OpenAI 2026-09-06
OpenAI reports an automated research intern, increasing agent contribution and substantial human intervention on longer tasks; it targets an automated researcher in March 2028. Internal measurements are preliminary, and compute and controls still constrain work.
Measurements for understanding the pace of AI development inside frontier labs 2026-09-17
Anthropic reports AI leading 26% of R&D in August, no fully autonomous measured R&D subset, and about 30,000 concurrent internal agents. It reports complete action-monitor coverage, which is not complete detection. The measurements use internal/model judgments.

Record reviewed 2026-10-18