Coding is dead. Long live the endless loop
anthropic's boris cherny heralds 'loops' as the next big shift in ai development—swarms of agents that improve code indefinitely.
claude code creator boris cherny hailed 'loops'—where agents prompt other agents to write code—as the industry's next major step, akin to the initial shift to agentic ai. cherny detailed using these non-deterministic loops to have agents constantly seek architectural improvements and unified abstractions, submitting pull requests indefinitely without human intervention. the approach relies on extensive test-time compute to let models incrementally improve codebases, a technique that burns tokens rapidly but offers compounding returns. while critics note this mimics classic recursion, the distinction lies in autonomous agents deciding when to stop rather than fixed conditions. the trend underscores a move toward background autonomy, though high token costs remain a barrier for companies not selling inference.