There was a first shift: from hand-written code to AI agents writing code on instruction. That was already a huge turning point.
Now there's a second shift, and Boris Cherny — creator of Claude Code — described it at Meta's @Scale conference with one precise word: loopy.
The AI world is becoming "loopy."
What This Means
An AI loop is a system in which one or more AI agents work in the background in a continuous, autonomous way — they don't wait for instructions, they don't stop when they finish a task, and they don't need manual supervision at every step.
Cherny described a concrete example: imagine two agents working in parallel on a codebase.
- One agent continuously looks for architecture improvement opportunities in the code
- Another looks for duplicated abstractions that can be unified
Both open pull requests just like any other developer on the team would. And because the code keeps changing constantly, they never finish — they keep finding new opportunities, adapting to changes, improving.
It's not an assistant answering a question. It's a collaborator that's always working, in the background, without waiting for you to ask.
The Progression Cherny Describes
Cherny positioned loops as the third step in an evolution:
- Hand-written code — a developer writes every line
- Agents on instruction — a developer asks the AI to do something, the AI does it
- Agents in continuous loops — agents work autonomously, the team supervises and approves
The jump from point 2 to point 3 is, according to Cherny, comparable in importance to the jump from point 1 to point 2. It's not an incremental optimization — it's a paradigm shift in how development teams work.
The Problem: Costs
AI loops have an inherent problem: costs scale with no ceiling.
An AI chatbot answers a question — it consumes a defined number of tokens, then stops. An AI loop works continuously — there's no natural point where it stops. If you leave a system of looping agents running 24 hours a day, 7 days a week, token costs can become significant very quickly.
This isn't an unsolvable technical problem — it's managed with budget limits, scope definition, and cost monitoring. But it's a factor anyone implementing AI loops needs to consider from the start.
What This Means for Non-Developers
The AI loop story sounds technical, but it has broader implications.
The same model — agents working autonomously and continuously in the background — will apply to many other contexts: SEO monitoring, content management, second-level customer support, data analysis, compliance. In all these cases, the shift from "AI on demand" to "AI running continuously in the background" will profoundly change how organizations are structured and which human roles remain central.
Supervision, goal-setting, and strategic judgment become even more important when execution is automated. Not less human — differently human.
"Loopy" is an informal word for something highly structural. The fact that it's used by the person who built one of the most advanced AI tools for developers suggests the industry is still at the stage where the precise words haven't been invented yet. The concept, though, is already in production.