Skip to content
Trending storyDeveloping

Meta's GitSwarm lets AI agents build on a shared Git repo, including failed attempts

1 article1 sourcesince Oct 11Last article 2h ago ·

Overview

AISummary of 1 article

Meta researchers present GitSwarm, a system in which many identical AI agents work on a shared Git repository without a central coordinator.

Each agent reads earlier commits, including failed attempts, and records which ones it used. On 50 ProgramBench program-rebuilding tasks, GitSwarm scores a 79.4% mean, compared with 65.1% for a single Codex agent told to keep working, at similar compute, according to the paper as reported by Rohan Paul on X.

Written by AI from the articles below · updated Oct 11, 6:41 AM ET

Check the sources:

Latest developmentOct 11, 6:32 AM ET
Meta's GitSwarm lets AI agents build on shared Git history, failures included

Article timeline

The articles in this story. Times are ET.

Oct 11
  1. Rohan PaulX
    Meta's GitSwarm lets AI agents build on shared Git history, failures included

    AIMeta researchers present GitSwarm, a system where many identical agents work on a shared Git repo without a central coordinator. Each agent reads earlier commits, including failed attempts, and records which ones it used. On 50 ProgramBench tasks, GitSwarm reaches a 79.4% mean score, versus 65.1% for a single Codex agent at similar compute.

    Image from @rohanpaul_ai's post

Heat trend

Current comparable heat
9
Peak comparable heat
10Oct 11, 7:00 AM ET
Change in past 24 hours
–

The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.