An Anthropic engineer warns about 'cognitive surrender'

Addy Osmani, an engineer on Anthropic’s Claude Code team, argues that AI coding agents can quietly reduce the critical thinking developers exercise. Because agents now handle much of his daily coding work, he says people must deliberately make room for thinking rather than assume their jobs will require it. He distinguishes three ways AI can affect judgment: “cognitive offloading,” delegating work while staying responsible for the result; “deferral,” treating an AI’s answer as the starting point for further decisions; and “cognitive surrender,” accepting its output without scrutiny. Osmani sees the last as the greatest risk, since repeated uncritical reliance can build “cognitive debt” and leave users unable to explain or own their work.

His warning is not a call to avoid AI or reason intensely about every task. Delegation is appropriate when the stakes are low, provided users recognize when they are handing over judgment. For important work, he recommends asking whether they formed a hypothesis before consulting an agent and which skills they may no longer be practicing. Developers should also make verification habitual: test generated code and ask the agent to identify what it has and has not checked. Osmani’s broader point is that abilities erode when unused, just as navigation and memorizing phone numbers have faded for many people. Engineers should keep exercising the skills they want to retain while using AI deliberately.