Kyle Mistele says coding agents need control loops, not giant prompts
AI Engineer · youtube · 2026-07-26
Kyle Mistele argues coding agents need control loops, not bigger prompts
In this talk, HumanLayer’s Kyle Mistele says the real fix for agents that produce unreadable 40,000-line pull requests is to design a control loop: sense the current state, choose a small next action, apply it, then measure again.
He maps that idea to code migration work:
- Sensor: Grep or structural search finds the smallest unmigrated piece.
- Controller: picks the next unit of work.
- Actuator: an agent makes a small change against hand-written golden patterns.
- Verifier: deterministic CI checks each step before the loop continues.
The workflow tracks PRs in version control, avoids stacking new work on top of unfinished changes, and keeps migrating code incrementally even while the team is offline.
More from coding & agent
- OpenAI ships GPT-6.1 Terra; Brockman says its computer use now beats his own speed — gdb · 2026-10-27
- Kolab Benchmark Uses Pro Screen Recordings to Show AI Agents Overestimate Collaboration Quality — minnesotanlp · 2026-10-27
- Dev argues Claude burns far more tokens than GPT models for the same coding work — haider1 · 2026-10-27
- Stop building agents that can't change models: stable contracts, external state, real evals — rseroter · 2026-10-27
- Cheap models can cost more in agent harnesses than frontier ones, researcher warns — vasuman · 2026-10-27
- Kanzo: A 5k-Line Rust Kanban Binary With MCP for Claude Code and Codex — allmudi · 2026-10-27