AI 'Loops' Emerge as Next Evolution Beyond Agents
Original: The AI world is getting ‘loopy’
Why This Matters
Loops represent evolution in AI autonomy, enabling continuous background optimization without human intervention as model capabilities improve.
Claude Code creator Boris Cherny highlighted agentic loops at Meta's @Scale conference on June 22, 2026, describing them as a significant step forward where agents continuously prompt other agents to write code, with loops running indefinitely to improve systems.
At Meta's @Scale conference, Boris Cherny, creator of Claude Code, addressed the emerging importance of agentic loops in AI development. When asked if loops represent hype or reality, Cherny responded affirmatively, stating: '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.' He characterized this shift as equally significant as the transition from manual coding to agent-based coding. Cherny described his own implementation: one agent continuously searches for architecture improvements while another identifies duplicated abstractions that can be consolidated. These agents submit pull requests autonomously, running perpetually as code evolves. The concept builds on recursive loops from classical computing but applies non-deterministic logic where AI agents determine stopping conditions rather than preset parameters. Popular implementations include the Ralph Loop, which summarizes completed work and assesses goal achievement. The approach aligns with test-time compute strategies, where additional computational resources enable models to solve complex problems. For hill-climbing tasks like codebase optimization, continuous incremental improvements can proceed indefinitely toward performance thresholds.