« Loop engineering »: new discipline or cron job rebranding?

The term is on the rise, the Pragmatic Engineer investigation concludes to a mix of triggers, cron jobs, and LLM slop. A real underlying question remains: who possesses the robustness of an observe-decide-act loop powered by LLM?

The fact

The Pragmatic Engineer (Gergely Orosz) publishes on 14/07/26 a survey on the term « loop engineering », which dates back to the vocabulary of AI teams. His conclusion, as summarized at the top of the article: under the word, we find a mix of triggers, cron jobs, LLM orchestration and « slop » - and a part of hype.

Our reading

The word is not new. What it condenses is a bit more: when a system repeats in a loop a cycle observe → decide → act powered by an LLM, who bears the responsibility for the robustness of the loop? Not the model, which has no state between turns. Not the classic devops, which does not know the fine non-determinism. A new niche is emerging: instrumenting the runs, containing the drifts, managing the idempotent retries, measuring the cost per turn, planning the rollback when a decision engages an external system. These subjects are already addressed by the agent SRE - the vocabulary is only trying to name them, which is a sign of adoption rather than a sign of fashion.

To watch

The real signal will not be the proliferation of tools with the word « loop » on their landing page. It will be the appearance of public postmortems that explicitly talk about « loop failure » - drift, retry-storm, costly loop - rather than « hallucination ». On that day, the discipline will have taken hold.