Build your own agent harness: omarsar argues vendor harnesses will quietly constrain users
dair_ai · x · 2026-10-31
- Hugging Face's omarsar argues in a long post that developers should build their own agent/eval harnesses rather than depend on vendor-provided ones.
- Core argument: harnesses belong to companies, which can modify them at will; as models grow more capable, harnesses will keep adding constraints — some so subtle even experts won't catch them — a terrible position for users.
- Guardrails, capabilities, behaviors, patches and constraints are integral to how you produce value with agent harnesses, so they shouldn't be offloaded; as long as models need harnesses, you need full control over them.
- He notes many full-stack AI companies are already building their own harnesses. The post quotes steipete: a vendor changing what you can or can't do at will is another reason to run your own.
More from coding & agent
- Chasing the Karpathy Peak: mapping AI productivity's climb and the Slop Cliff past it — sirbayes · 2026-10-31
- Lauren Tan's agent playbook: set up rules and verification once, then scale to 200+ agents — anselm · 2026-10-31
- Letting Claude organize an iPhone Home Screen by color via MCP — BTCLSD · 2026-10-31
- Dev on single-agent orchestrators: 'like I suddenly became a multithreaded processor' — nathanborror · 2026-10-31
- Pydantic AI agents get durable execution on Temporal with one capability — samuelcolvin · 2026-10-31
- ASC CLI now covers Xcode commands, fastlane migration, and Linux builds via xtool — rudrank · 2026-10-31