V4.1's agent-team decisions look reasonable on single-file projects, but subagents remain half-baked
teortaxesTex · x · 2026-09-10
The author tested V4.1 on a single-file project with an agent team and found its decision-making process pretty reasonable. He notes that subagents are a completely separate mechanic and still fairly half-baked.
More from coding & agent
- Hy4 preview tested: playable 3D survival game from a single prompt in WorkBuddy — mhdfaran · 2026-09-10
- Astra xhigh's token efficiency may stem from its subagent instructions — Al_Grigor · 2026-09-10
- DeepSeek V4.1 Flash vs GPT-6 Astra: same coding task, wildly different code structures — kevinkern · 2026-09-10
- OmniParse: open-source tool turns 20+ file formats into LLM-ready structured data — tom_doerr · 2026-09-10
- Meta reportedly preps Shared Agents for Muse, timed a week before OpenAI's Managed Agents at DevDay — testingcatalog · 2026-09-10
- Hidden Shared Agents section found in Meta's Muse app with working sub-agent creation flow — testingcatalog · 2026-09-10