Mollick: success with long-running AI agents means supervising the loop

Ethan Mollick explains that using agents like Codex successfully hinges on knowing when and how to intervene—injecting instructions versus forking conversations—and that while humans can't stay fine-grained in the loop on long tasks, they can supervise the loop itself, urging labs to improve interfaces for this.

2026-09-10 ~ 2026-09-10 · 4 related posts

• Ethan Mollick: you're probably not steering your long-running coding agents enough — emollick · 2026-09-10
• Ethan Mollick on steering long-running agents: when to instruct, queue or fork — emollick · 2026-09-10
• Mollick: you can't be in the loop for long-running agents, but you can oversee it — emollick · 2026-09-10
• Ethan Mollick: You Can't Stay in the Loop, But You Can Oversee It — Labs Should Build for That — emollick · 2026-09-10