Why top coding models without dedicated s-tier harnesses go untested: the harness-model pairing effect
illscience · x · 2026-11-09
A developer explains why he only seriously uses coding models with s-tier harnesses (Codex, Claude Code, Cursor, plus OpenCode): without quality-of-life basics like remote and browser control, it's hard to see what a model can really do.
Key points:
- As top capability shifts to things like agent teams, the best harness+model pairs will diverge further from naive sum-of-parts combinations.
- He's heard the best models are post-trained alongside their own harnesses, which preserve more info (e.g. full CoT) than vanilla harnesses — partly to resist distillation, mainly to capture vertical-integration economics.
- "Getting the most out of models" at the harness/agent layer is an emerging discipline; the right pair feels like living in the future.
More from coding & agent
- Design in the AI era: study great interfaces before reinventing them, says designer — bbssppllvv · 2026-11-09
- Prompt Master: open-source Claude skill that nails prompts on the first try across 18+ tools — nikola_mr64990 · 2026-11-09
- After 18 months full-time, founder publishes agentic brain architecture and 10 guides — Cold-Cranberry4280 · 2026-11-09
- Grok Bot turns YouTube channels and podcasts into a weekly magazine with EPUB — elonmusk · 2026-11-09
- Harrison Chase's North Star: agents that improve via model, harness, and context updates — hwchase17 · 2026-11-09
- Open-source video-to-game skill turns one gameplay clip into a playable Three.js game for ~$6 — OdinLovis · 2026-11-09