Anthropic engineer Addy Osmani, who works on Claude Code, argued in an essay on X that coding agents are wearing down developers’ critical thinking, writing that nothing in his own workday pushes him to think hard now that agents handle so much of it. Osmani separates healthy “cognitive offloading,” where a developer hands off work but still owns the result, from “cognitive surrender,” a term two Wharton researchers coined for trusting AI output without question. He says handing off low-stakes tasks entirely is fine, but engineers who can’t explain an answer, even with their agent’s help, no longer own it. Osmani suggests forming a hypothesis before prompting and asking agents what they can and can’t check, warning that skills people stop using fade faster than they expect.
