Pawel Huryn · 10 min read · AI & Product
Agent Loops for PMs: 20+ You Can Run This Week
Read the original on productcompass.pm →
An AI agent loop only stops when you define what 'done' looks like—including acceptance criteria and sign-off. The piece clarifies the difference between routines, workflows, and loops, and shows how to build a loop that runs until a condition is met using a /goal pattern, complete with a copy-paste template and a practical PM-focused example.
Key Takeaways
- Define acceptance criteria, a success metric, and who signs off to avoid endless retries.
- Distinguish between routines, workflows, and loops; a loop repeats and checks its own work until a condition holds.
- Use the /goal pattern to run until a defined condition is met, with built-in stop logic and optional cadence.
- Most loops don’t require a fixed schedule; the stop condition and guardrails are what matter.
- Include a concrete template and example (e.g., competitor-watch loop) to implement loops quickly in PM workflows.
Excerpt
A loop does not fail because the agent keeps working. It fails because nobody told it what “done” means. That’s the part most loop guides skip, and it isn’t a coding detail. Defining “done” is writing acceptance criteria, picking a success metric, and deciding who signs off. A loop is only as good as the thing that tells it to stop. It also explains why the loop debate sounds so confused. Nvidia's Jensen Huang: "Nobody writes prompts anymore. The new job is to write and handle loops." Boris Cherny, who built Claude Code: "My job is to write loops." Then…