Five principles for working with Loops

An engineer at Linear, the issue-tracking company, widened an AI agent's authority in stages before letting it assign cleanup work to colleagues.

Andrew Moedinger, an engineer building the Loops feature at Linear, describes how Linear's own teams use it. Loops lets a team describe recurring work in plain language and have Linear's AI agent run it on a schedule or when an event happens. Linear's security team uses loops to sort vulnerability reports, and its customer success team uses one to route requests to whoever is on call.

• To remove about 80 cases of outdated code, the first version of the loop only identified candidates.
• The next version filed five issues at a time to a triage inbox and explained whom it would assign, without assigning anyone.
• Only then did the loop assign issues and hand them to a coding agent, with the assigned engineer reviewing the resulting pull request.
• Any team member can open a loop, see what each run did and change it.

Moedinger presents delegation as the engineering work that remains: deciding what to hand off and giving the AI agent context and limits. This is a vendor's account of its own product.

Derived by Working Surface from the article; more in the Playbook. Source line: You probably won’t get everything you want from a loop with a one-shot prompt.

14 October 2026: as AI agents raise the volume of code changes, teams are moving routine checks into automated tests and installable rules, and keeping people for intent and contested decisions.