Agent swarms run whole projects

What the forecast means

Milestone stage: deployment

partially-demonstrated

Long projects can be decomposed and partly executed by agents; accepted end-to-end work under the stated human-involvement limit is not established.

Completion criteria: At least three commissioned projects, including software and research work, are accepted by independent clients; disclose all agent retries, human code/content repairs, active labor, elapsed time and follow-up defects. The month refers to estimated expert effort, not idle wall time.

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

Built on

• AI agent swarmsRecorded · 2026enables
• AI writes most new codeRecorded · 2026enables

Led to

• AI research run by agent swarmsForecast · ~2028influences
• First company run by agent swarmForecast · ~2030enables
• Software that builds itself from goalsForecast · ~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.

• Time horizons 2026-09-08

  The page was last updated September 8 and is no longer actively updated. Its horizons measure expert-time difficulty at selected success rates on a software/ML/cyber task suite; estimates beyond roughly sixteen hours are unreliable. Missing new systems are a coverage gap.

Record reviewed 2026-10-18