Stop conflating "harness": labs build the pilot, companies build the mech
robleclerc · x · 2026-10-28
Rob LeClanc argues the term "harness" conflates two different things:
- The model-side harness: the loop that turns a model into an agent — tools, memory, autonomy. Claude Code and Codex are harnesses, built by labs with generic capabilities and shipped with the model.
- The company-side harness: everything you build so the agent does real work in your business — systems, data, rules, workflows, objectives, skills, and tools.
His analogy: as agents get more autonomous and interchangeable, treat them like pilots inside mechs. Labs train and supply the pilot; companies build the mech — sensors, systems, data access, rules, mission control. Pilots get swapped as better ones ship, and every company will need to build its own mech.
Related event: Rob LeClerc: Agent Harness Splits Into Model Side and Company Side(2 posts)→
More from coding & agent
- Multimodal GraphRAG models missing and contradictory evidence for storm damage assessment — JeremyCMorgan · 2026-10-28
- Vibe coding in real life: 20 hours of personal-project work done in 2 — gleech · 2026-10-28
- LangChain open-sources Content Studio, a Slack design agent, and shares its harness design — BraceSproul · 2026-10-28
- "If AI can do it, humans shouldn't": team lets AI agent review PRs — Yuchenj_UW · 2026-10-28
- Sequoia, Citi and OpenAI launch No Limit: build a poker bot in 60 minutes, $100K grand prize — polynoamial · 2026-10-28
- Picteus: open-source local app searches your AI image library by prompt, model, LoRA or seed — shivam_dewan · 2026-10-28