AgenC now defaults to one agent after multi-agent systems fell 39%–70% behind
tetsuoai · x · 2026-07-23
AgenC defaults to one agent after multi-agent benchmarks fell 39%–70% behind
An X thread explains why AgenC 0.9.0 now defaults to one agent, isolates parallel writers, and treats integration as an evidence problem instead of a prompt instruction.
The cited evidence comes from two benchmark results:
- CAID’s PaperBench Code-Dev isolation ablation:
- shared-workspace multi-agent: 55.5
- single agent: 57.2
- worktree-isolated multi-agent: 63.3
- Scaling Agent Systems on sequential PlanCraft:
- every evaluated multi-agent architecture finished 39.1%–70.0% below the single-agent mean
The thread’s takeaway is that parallel agents sharing one checkout can interfere with each other even if prompts warn them not to. AgenC therefore gives every parallel writer a spawn-unique Git worktree and only returns an integration reference when the output is committed and clean.
Related event: AgenC Defaults to Single Agent After Multi-Agent Setbacks(4 posts)→
More from coding & agent
- 'My intelligence stack': developer splits AI coding work across Opus 5.5, Fable and Astra with human review gates — sh_reya · 2026-10-27
- Baseten webinar: evaluate open models for coding agents by benchmarking on your own merged PRs — baseten · 2026-10-27
- Thorsten Ball on working with agents: run more evals, automate everything, use the best models — omarsar0 · 2026-10-27
- DeepMind paper: LLMs can't sample randomly on their own — pass random numbers from code instead — rohanpaul_ai · 2026-10-27
- BELL: Open-Source Stack Combines Browserbase, Eve, iMessage and Stripe Link for Personal Agents — jeff_weinstein · 2026-10-27
- Legora runs 2,000 parallel agent evals on Modal Sandboxes for legal AI — blelbach · 2026-10-27