Bloated agent harnesses hurt you: minimal setup yields polite, zero-mistake runs
jasonkneen · x · 2026-11-08
Jason Kneen argues you should strip your agent harness down to minimal tools, no skills, no polluted memory/history. Building slopeisley multiplayer with his lazar minimal harness and GPT Sol (just a handful of base skills), the model responded unusually politely ('May I...') and made zero mistakes in a day, with built-in verification handling errors. His takeaway: over-bloated harnesses hinder rather than help.
More from coding & agent
- Solo founder runs his whole business on a team of 6 Grok Bots — PrajwalTomar_ · 2026-11-08
- GergelyOrosz: Codex blocked random innocent accounts beyond the AI slop list — unironictechbro · 2026-11-08
- Bank dev tests 1.7B local model offline: all traffic stayed on localhost, emails in 2.5s — Ok_Resolve4741 · 2026-11-08
- Dev ships interruptible history podcast generator, 80% coded by Claude Code — Goldenchild123 · 2026-11-08
- Dev builds custom workspace bridging ChatGPT and Codex with Planner/Codex modes — Specific-Animal6570 · 2026-11-08
- How Ian Nuttall would grow Typefully with SEO: 4 data-driven moves — iannuttall · 2026-11-08