AWS engineer: AI developers need caution and a dislike of busywork

Yanacek, who works on agentic AI and safety, said coding itself is no longer the main bottleneck. Engineers previously spent days waiting for manual tests, code approvals and design changes, often requiring reviews from teammates. In the workflow he described, agents generate code, run tests, fix problems and prepare updates for merging and deployment. A loop can take hours, but Yanacek can work on other projects or add product features while it runs. Human input is focused on design discussions and complex reviews.

That approach depends on carefully set limits. Yanacek said he spends much of his time deciding which tools agents can use independently, which actions require approval and how those boundaries should be enforced. He relies on an “auto-trust” file that lists permitted tools and commands alongside actions requiring human sign-off.

“I always think about what is in the auto-trust file, because that's how you get productivity: by having the agent be able to do things without you having to micromanage every single command that it runs,” he said. His aim is to have agents make appropriate decisions without constant review, backed by clear rules for what they should and should not do.

“Most of what I think about is AI safety,” Yanacek said. “And because I'm doing that, I sleep pretty well.” The goal, he said, is to do enough work upfront to make agents reliable enough to operate unsupervised and take over tedious tasks.