CopilotKit/AG-UI, generative UI, human-in-the-loop; harness engineering for production coding agents.
The model is a stateless next-token oracle; the harness is the control plane that turns one good thought into a hundred reliable actions — and it's where production agents actually die.
A coding agent is not a smart model — it is a tight loop where read/edit/grep/bash tools meet a sandbox, a staleness check, and a test runner that feeds failures back until the diff is green.
Turn an agent from a background batch job into a streaming, interruptible collaborator that renders its own UI — by mastering the AG-UI event protocol, CoAgents state sync, and human-in-the-loop approvals.
Turn the audit trail your harness already records into a training signal — sample agent rollouts, let an LLM judge rank them, and GRPO bends a 14B model past a frontier model on your one narrow task.