On Friday, Claude Code creator Boris Cherny took the stage at Meta’s @Scale conference in San Francisco and fielded a question that cut straight to the heart of the current AI development cycle: “Are loops the next hype cycle, or are they for real?” Cherny’s answer was unequivocal. “Yes, they’re for real,” he said. “Two years ago, we wrote source code by hand. We started to transition so agents write the code. And now we’re transitioning to the point where agents are prompting agents that then write the code. As big as the step from source code to agents was, loops are just as important and as big a step.”
The concept, detailed by Cherny later in his talk (around the 32-minute mark of the published video), describes a fundamental shift in how developers interact with AI. Rather than managing individual agents with clear goals and discrete check-ins, the loop model authorizes a persistent swarm of agents to work continuously in the background. Cherny offered a concrete example from his own workflow: one agent is continually looking for ways to improve the code architecture, while another hunts for duplicated abstractions that can be unified. They submit pull requests like any other coder, and because the code is constantly changing, they never stop running.
Also read: AI chipmaker Groq raises $650M, pivots after Nvidia's $20B licensing deal
The recursive roots of a new AI pattern
The first thing to recognize is that this isn’t entirely new. Recursive loops — functions that call themselves to repeat an action, along with a condition that stops the loop — are a mainstay of introductory computer science courses. The agentic loops Cherny describes follow a non-deterministic logic: a sub-agent, rather than a clear condition, chooses when to stop the loop. But the same basic approach is at work. As soon as programmers started using AI to complete tasks, some version of a recursive loop, with AI overseeing AI, was bound to emerge.
Unlike classic computing, however, these agentic loops can be maddeningly simple. One of the most popular tricks is the “Ralph Loop” (named for Ralph Wiggum), which basically sums up all the work the model has done and asks if it has accomplished its goal. It’s a crude but effective way of dealing with AI models that get lost as they run for too long — essentially bouncing the model back and forth until the task is complete.
Also read: Nobel laureate John Jumper leaves DeepMind for rival Anthropic
The economics of never-ending compute
Another way to think of loops is as part of the general push for more test-time compute. As OpenAI researcher Noam Brown observed earlier this month, contemporary models can solve nearly any problem if you throw enough compute at them. That means one way to ensure a problem gets solved is to just keep throwing compute at it until it’s finished. This is particularly true for “hill-climbing” problems like improving a code base, where the model can just keep making incremental improvements until it reaches a given threshold — or, as in Cherny’s example, for as long as there is compute to spend.
If that sounds expensive, it should. Like agentic AI before it, AI loops burn through tokens a lot faster than simple Q&A chatbots. Because the point is to keep the loop running all the time, there is no ceiling to how much you can spend. That’s fine for Anthropic, which is ultimately in the token-selling business, but for everyone else, it may be a pricey way to work. Still, depending on the problem the loop is trying to solve, and with the right setup for oversight of token spend and model drift, the benefits could be staggering enough to outweigh the costs.
Frequently Asked Questions
What exactly is an AI loop in the context of agentic coding?
An AI loop is a continuous workflow where multiple AI agents work in the background indefinitely. For example, one agent might constantly look for ways to improve code architecture, while another identifies and unifies duplicated code, submitting pull requests like a human developer.
Who is Boris Cherny and why is his opinion on this significant?
Boris Cherny is the creator of Claude Code, a prominent AI coding tool from Anthropic. His endorsement of loops carries weight because he is a leading practitioner in the field of agentic AI, and his work directly shapes how developers interact with AI.
What is the ‘Ralph Loop’ mentioned in the article?
The Ralph Loop, named after the character Ralph Wiggum, is a simple but effective technique where the AI agent repeatedly summarizes its work and asks if it has accomplished its goal. It is a way to prevent the model from drifting off task during long-running operations.
How do AI loops relate to the concept of ‘test-time compute’?
AI loops are a practical application of test-time compute, a concept highlighted by OpenAI researcher Noam Brown. The idea is that by throwing more computational power at a problem, a model can solve nearly any task. Loops operationalize this by continuously applying compute until a problem is solved.
What are the main risks or downsides of using AI loops?
The primary risk is cost. Because loops are designed to run continuously, they can burn through tokens and compute resources very quickly, with no natural ceiling. This makes them potentially expensive for all but the most well-funded organizations.
Be the first to comment