An Anthropic engineer warns about 'cognitive surrender' as AI agents take over his work

An Anthropic engineer named Addy Osmani has expressed concerns about "cognitive surrender" among coders as AI agents increasingly take over their work. In a recent X essay, Osmani, who works on the Claude Code team at Anthropic, warned that AI agents have reduced the need for critical thinking in his daily tasks.

Osmani identified three types of diminished thinking with AI use: cognitive offloading, deferral, and cognitive surrender. Cognitive offloading refers to delegating tasks to AI while still retaining ownership of the products. Deferral occurs when AI's output becomes the new input. Cognitive surrender, a term coined by researchers at Wharton School, is complete trust in AI outputs, which Osmani describes as lazy.

The danger lies in surrendering critical thinking too often, which can lead to "cognitive debt." While Osmani acknowledges that AI can be helpful, he emphasizes the importance of exercising critical thinking skills. He encourages engineers to ask themselves questions like whether they have formulated a hypothesis before asking AI and what they are no longer practicing.

Osmani also shared that engineers should build verification loops to ensure the code written by AI works, and that they should know whether they are merely offloading tasks or surrendering ownership. He notes that skills often degrade over time, similar to how map-reading and memorizing phone numbers have become less common due to the prevalence of digital tools.

In summary, Osmani urges engineers to maintain a balance between using AI and preserving critical thinking skills, using it where appropriate and not letting AI take over tasks that require deep thought.