Developers start using always running AI loops to improve code

In short: Some developers are starting to use “AI loops,” where multiple AI helpers keep working in the background continuously, especially on software code.

What's going on

A growing idea in AI is the “loop,” which means giving AI agents (AI helpers that can take steps on their own) permission to keep running without a clear end point. Instead of asking one question and getting one answer, a loop keeps checking its own work, making changes, and checking again.

Boris Cherny, who leads Claude Code, discussed this at Meta’s @Scale conference. He described a shift from people writing code by hand, to AI agents writing code, and now to agents prompting other agents to write and improve code. In his example, one agent keeps looking for ways to improve the overall structure of a code base, and another looks for repeated parts that can be combined. They can submit “pull requests” (suggested code changes, like editing suggestions in a shared document), and because the code keeps changing, the agents never stop.

The loop concept is not brand new in computing. Traditional software uses loops too, like a recipe step that repeats until a condition is met. The newer twist is that an AI agent may decide when to stop, rather than following a fixed rule.

What to watch

Always running loops can be expensive. They use a lot of “tokens,” which are small chunks of text that AI systems count and charge for, similar to a taxi meter that keeps running. The practical question is whether companies can get enough value from nonstop background improvements to justify the ongoing cost and the added need for oversight.

Source: TechCrunch AI