Simon Willison finds his own boundary between vibe coding and agentic engineering dissolving as Claude Code reliably ships routine production code he no longer reviews line by line.
Key Takeaways
Willison defines agentic engineering as AI-assisted coding by professionals who apply 25 years of experience to security, maintainability, and operations – not just output speed.
He now treats Claude Code like a trusted internal team: ships its output without reading every line, audits only when bugs surface – and calls this uncomfortable.
The normalization-of-deviance risk is explicit: each correct unreviewed output raises the odds of misplaced trust at a genuinely risky moment.
Evaluating repo quality has broken down – a half-hour vibe-coded repo with 100 commits, a README, and full test coverage looks identical to a carefully built one.
Bottlenecks have shifted upstream: design processes built around expensive engineering cycles may need to be riskier and faster when build cost collapses.
Hacker News Comment Review
Commenters pushed back on Willison’s JSON-endpoint confidence: a correct Claude Code output still requires answered decisions on naming, auth, error contracts, and rate limiting – and those decisions are where real engineering lives.
Consensus formed around errors becoming more subtle, not fewer: compile failures are obvious, wrong edge-case behavior or silent security holes are not, so trust calibration is the actual unsolved problem.
Several commenters noted LLMs act as amplifiers, not equalizers – good engineers accelerate, undisciplined engineers ship faster garbage, and the review pipeline is the only discriminator.
Notable Comments
@zarzavat: argues AI errors have grown more subtle, not disappeared – compiling and working code can still be wrong in edge cases or carry silent security debt.
@ilikebits: reframes the LOC-per-day point usefully – “LOC is useful here not because it’s a metric for output but because it’s a metric for understandability.”
@peterbell_nyc: maps the vibe-vs-agentic split to pipeline rigor: agentic engineering means caring about correctness, resilience, scalability, and maintainability with structured review gates.
