ModelsToolsSecurity

Agents managing agents: loops as the next step in agentic AI

Boris Cherny (Claude Code) says at Meta’s @Scale that continuous agent loops — agents continually evaluating and modifying other agents — are a major next step for automating software development.

In detail

  • Cherny details agents that constantly submit pull requests, improve architecture, and unify duplicated abstractions.
  • Loops operate as non‑deterministic, recursive processes where sub‑agents decide dynamically when to stop rather than using fixed stop conditions.
  • He equates the move to loops with the prior shift from hand‑written source code to agentic systems over the past two years.

Why it matters

This signals a move from one‑off automation to continuous, self‑optimizing AI work — offering big productivity gains but raising control, testing and governance challenges for businesses.

For you Assess recurring development or ops tasks for supervised agent‑loop pilots and require PR reviews, audit trails and rollback plans before production deployment.

← All news

Summaries are generated automatically and link to the original source.