Anthropic engineer raises cognitive surrender on November 4, warning precise AI output weakens independent thinking
Addy Osmani of Anthropic's Claude Code team argues that as AI output grows more correct, people skip verification, risking cognitive debt and orchestration tax.
On November 4 Addy Osmani, an engineer on Anthropic's Claude Code team and former Google technical executive, published a long post on X about maintaining critical thinking when AI agents do the work. He argues that as agents get more capable, people may gradually lose the ability to think independently, verify results and understand complex systems, eventually treating AI judgement as their own, a phenomenon he calls cognitive surrender.
Osmani worked at Google for more than 14 years on Chrome developer experience and Google Cloud AI, and now works on Claude Code at Anthropic. He said he is a firm supporter of agentic workflows and routinely runs several agents in parallel, but has noticed a concerning shift: as models improve, obvious errors in output become rarer and humans find it harder to stay alert. Previously an answer required gathering information, analysis and judgement, but that process can now be compressed into seconds. AI output is typically well structured and fluently expressed, and correct most of the time, which makes it easier for users to skip independent thinking and verification.
Osmani further argues that users should form an initial hypothesis before handing a task to an agent, then use AI to fill in information while retaining key judgement. He believes the absence of that habit accumulates into cognitive debt and an orchestration tax, where users must spend extra effort re-learning problem structure that should have remained theirs.