“Loop engineering” spread after Boris Cherny (Claude Code’s creator) and Peter Steinberger mentioned it on X. Andrew Ng amplified it, calling it a “hot buzzphrase.” The Pragmatic Engineer’s investigation found the practice is less about prompting and more about building automated cycles around an LLM.
The concrete pieces the Pragmatic Engineer identified inside these loops:
- Triggers — events that kick off an AI agent’s work without human prompting
- Cron jobs — scheduled runs that keep agents iterating on a task
- Observation cycles — agents check their own output, decide what’s next, and repeat until a goal is met
- AI slop — the waste and noise these loops generate when poorly designed
The core shift: instead of hunting for one magic prompt, you build a reliable loop around the model. The agent acts, observes the result, revises, and repeats. The human engineers the loop, not the prompt.
NOTE
The Pragmatic Engineer piece openly questions whether loop engineering is a durable practice or a “here today, gone tomorrow” trend. The term has no formal definition — it’s a label builders on X started using, not something from a paper or framework.
The takeaway: loop engineering is real in the sense that teams are building automated agent cycles with triggers and cron jobs. Whether the label survives is a separate question from whether the practice does. The full breakdown is worth reading before adopting the terminology in your own team’s docs.