Viral 'Claude Code Harness' article argues the harness, not the model, makes the agent
blaizedsouza · x · 2026-11-20
A circulating blog post claims to break down the runtime architecture behind Claude Code, arguing that two agents on the same model can differ more than two model generations do — the model is the engine, the harness is the car. It proposes a 4-tier model hierarchy (Opus 5.5 in high-effort plan mode drafting an immutable blueprint, Sonnet 5.5 executing, plus tiers for file scouting and self-review) and argues that a single generalist frontier model handling planning, scouting, patching and self-review causes silent context rot and token exhaustion. Note: the model naming and 'official Anthropic systems paper' framing are unverified and likely third-party speculation.
More from coding & agent
- Open-source repo offers 500+ AI-generated low-poly 3D landmarks for maps under CC0 — tomjohndesign · 2026-11-20
- SwirlAI: nine places to plug decision models into agentic RAG, with runnable code — blaizedsouza · 2026-11-20
- Agent safety rule: human-approve any irreversible, externally visible action — and show the exact action, not 'proceed?' — Beneficial_Use2116 · 2026-11-20
- Prefill beats decode: SGLang wins agent runs on DGX Spark despite slower tok/s — AIFlow_ML · 2026-11-20
- Running ComfyUI fully via Claude in a container — no manual UI anymore — xWranglerWranglerx · 2026-11-20
- AI writes all code to make a real robotic arm play chess, open-sourced — FinanceYF5 · 2026-11-20