Glossary
What is agentic loop?
An agentic loop is the observe-decide-act-verify cycle an AI agent runs repeatedly without a human issuing each step: it reads state, picks an action, executes it, checks the result, then loops again. What triggers the next cycle varies: a heartbeat timer, a cron schedule, a webhook/hook, or a goal condition that's not yet met.
Why it matters
A loop without a stop condition doesn't stop: it burns tokens, drifts from the goal, or ships changes nobody checked. The four trigger types set how often the agent runs (heartbeat: fixed interval regardless of state; cron: scheduled; hook: event-driven, e.g. on push or on file change; goal: runs until a condition is met), but none of them guarantee the output is correct. That's what the verify step is for, and it's the step teams skip first when they're moving fast.
How Patchrooms uses it
Patchrooms sits at the verify step. You embed one script tag on the preview the loop just produced; a reviewer clicks the broken element, types or records a note, and that becomes a structured report (DOM selector, screenshot, URL, console errors). "Copy agent feedback" turns it into a Markdown prompt block, or the agent reads it directly via MCP (list_reports, get_report). Either way, the loop's next iteration acts on a real verification signal instead of guessing.
Try the live demo