Boris Cherny Backs AI Loops as Next Step Beyond Agents at Meta's @Scale 2 Years On
Updated
Updated · TechCrunch · Jun 22
Boris Cherny Backs AI Loops as Next Step Beyond Agents at Meta's @Scale 2 Years On
1 articles · Updated · TechCrunch · Jun 22
Summary
At Meta’s @Scale, Claude Code creator Boris Cherny said AI loops are “for real” and as important a leap as the shift from hand-written code to coding agents.
Cherny described always-on systems in which one agent keeps improving code architecture while another hunts duplicated abstractions, with both continuously submitting pull requests as the codebase changes.
Those loops push agentic AI beyond one-off prompts by letting agents prompt other agents and keep working until a sub-agent decides the task is done, a non-deterministic version of recursive iteration.
The approach also fits the broader push for more test-time compute: models can keep making incremental gains on tasks like code improvement, but the loops consume far more tokens and can run without a natural spending ceiling.
That cost and oversight burden could limit adoption outside AI vendors, though Cherny’s endorsement suggests continuous multi-agent workflows are moving closer to real production use.
With AI's token costs spiraling, companies are curbing spending. Can the 'agentic engineering' revolution survive its own expensive price tag?
AI agents promise 200% productivity but write dangerously insecure code. Are we building the future or automating a global security crisis?
Loop Engineering in 2026: How AI Loops Are Redefining Software, Jobs, and Governance
Overview
In June 2026, industry leaders like Boris Cherny of Anthropic announced the rise of Loop Engineering, marking a major shift in AI-powered software development. Unlike the earlier era of Prompt Engineering, where users controlled large language models like ChatGPT through single prompts, Loop Engineering introduces automated, recurring AI loops. These loops continuously guide and refine AI agents, moving beyond treating models as black boxes. This new approach transforms how software is built and maintained, enabling more scalable and efficient systems that require less manual intervention and open new possibilities for automation and improvement.