The "I don't know, Claude wrote this" pandemic

As AI coding assistants become standard, engineers increasingly merge code they cannot explain, exemplified by an engineer answering an architecture question with 'I don't know, Claude wrote this.' Google's Addy Osmani calls this 'cognitive surrender' — distinct from 'cognitive offloading' — where the AI's output becomes the human's output without any perceived need to verify it. A viral Reddit post and this Hacker News thread show how widespread the issue has become, with the author emphasizing that the engineer who opens the PR remains responsible for the code, regardless of which tool suggested it.

The article is sponsored by CodeRabbit CLI, positioned as an external reviewer for AI-generated code. Unlike asking the same agent to review its own work — compared to 'a student grading their own exam' — CodeRabbit uses a different agent with 40+ static analyzers and no attachment to the code it reviews. It flags problematic parts and is used by teams at Nvidia, Indeed, and over 15,000 other organizations.

The described workflow that leads to surrender starts with a planning session using Claude and a 'superpowers' brainstorming skill, where the AI explores the codebase and proposes a plan after clarifying questions. After correcting the plan and executing, the engineer skims the resulting code and opens a PR without fully understanding it. For larger ambiguous tasks that in a pre-AI world would be split into smaller pieces, the rush leads to even less scrutiny, skipping the part where the engineer actually understands what's going on.

When a reviewer asks fundamental questions the engineer can't answer, the only acceptable next step is to try to understand the code — not to blindly forward comments to Claude. The author warns that this is what AI reviewing its own code looks like: no human oversight, and no real accountability.