Chess agent lesson: model plus harness together make the agent, not the model alone
andreisavu · x · 2026-11-11
Andrei Savu shares the core lesson from his LittlePly chess agent project: the model and the harness together are the agent.
- Learning to suggest a good move is only one part of the problem
- Deciding what to explore, managing the clock, and recovering from failures matter just as much
- The agent is strong enough to teach you tricks without feeling unbeatable
Related event: LittlePly: Open-Source Chess Agent Combines 87M Model with Search Harness(4 posts)→
More from coding & agent
- The $200 AI music video was 'just a simple scaffold' — here's the method behind it — Afinetheorem · 2026-11-11
- Kimi K3.1 Pro + Cerebras speedrun: one prompt completes an online purchase in 11.5 seconds — Kimi_Moonshot · 2026-11-11
- Jeremy Howard: Claude's clean-room compiler copied LLVM's known mistakes — Machine Learning Street Talk · 2026-11-11
- Indie dev runs product with four AI 'employees' — kanban shows per-task costs from $0.02 — pramodk73 · 2026-11-11
- Running a local Qwen 27B coding agent: no CI failure callbacks is the pain point — BraceletGrolf · 2026-11-11
- FINCOSMOS: open-world finance agent benchmark where none of 150 sessions passes — TheFinAI · 2026-11-11