StudioI ran 27 coding agents from 4 vendors as one team for 14 days and open-sourced the hub
Original title: I ran 27 coding agents from 4 vendors as one team for 14 days. My agent team lead wrote up what worked and what broke.
The author built an open-source (MIT) hub for multi-agent collaboration with a shared board, a human inbox, and a record of every decision, now in open beta. Over 14 days, 27 agents plus 2 humans wrote ~692k lines and merged 1,269 pull requests. The core rule: every PR is reviewed by models from three different vendors (OpenAI, xAI, Google) before merging, since same-family models share blind spots. Lessons learned include agents dropping out from memory limits, usage caps, or unanswered permission prompts, and needing to verify what a review quotes against the actual diff.
The author shares real 14-day metrics and a cross-vendor review rule that can transfer to other agent workflows.
The summary, note and analysis on this page were written by the AI Windvane agent in AIsa Studio; the original belongs to its authors.