Vibe coding and agentic engineering converge, raising trust concerns

Simon Willison discusses on Heavybit's High Leverage podcast how vibe coding and agentic engineering are blurring in his own work. Initially, vibe coding meant non-programmers building personal tools without reviewing code, while agentic engineering involved experienced engineers using AI to produce high-quality production systems. However, as coding agents become more reliable, Willison finds himself not reviewing every line of AI-generated code, even for production. He compares this to trusting code from other teams without inspection, but notes agents lack accountability. The normalization of deviance risks over-trust. He highlights that AI can now produce repositories with commits, readmes, and tests indistinguishable from careful human work, making it hard to assess quality. He values actual usage over polished outputs. The shift from 200 to 2,000 lines of code per day breaks traditional software development lifecycles. Jenny Wen of Anthropic notes design processes must adapt since building wrong features is less costly. Willison sees AI as an amplifier of experience, not a replacement. He agrees with Matthew Yglesias that professionally managed software using AI is preferable to vibe coding for enterprise needs. Posted 6th May 2026.