Uncle Bob argues AI won’t replace developers who can review and judge code quality, using Claude’s brittle permission implementation as evidence.
Key Takeaways
Claude writes code at speed but replicates average developer habits, producing coupling and brittleness unless explicitly prompted to refactor.
The permission/action implementation Uncle Bob observed worked until edge cases broke it, a classic silent-failure pattern.
Effective AI coding requires domain knowledge to verify output, just as disputing Hawking radiation requires physics expertise to validate the result.
“Everyone thinks I speak English so I can get Claude to write me a program” is the core misconception Uncle Bob targets.
The workflow he describes: read AI output, assess quality against experience, prompt for refactoring, repeat.
Hacker News Comment Review
Commenters split on the framing: some read the title as cancellation clickbait, others treated it as a real inflection point for craft-focused developers.
One commenter pushed back on readability as the key metric, arguing structural complexity and coupling are fundamental limits for LLMs and humans alike, not a human-specific problem.
Skepticism toward Uncle Bob personally ran high, with commenters suggesting Clean Architecture dogma makes his credibility on this topic complicated.
Notable Comments
@nine_k: spaghetti code is as bad for an LLM as for a human because coupling is a structural limit, not a readability preference.
