Better Offline AI Isn't Making Engineers 10x As Productive with Colton Voege
58 snips
Sep 19, 2025 Colton Voege, a software engineer known for his insights on web application development, discusses the real impact of AI on engineering productivity. He argues that while LLMs can handle boilerplate code, they struggle with systems thinking and maintaining code consistency. Colton highlights the hype surrounding AI, pointing out that it shifts rather than replaces engineering work, and warns of risks like security vulnerabilities in AI-generated code. His candid take challenges the notion that AI can massively boost productivity in software development.
AI Snips
Chapters
Transcript
Episode notes
Code Generation ≠ Software Engineering
- Generating code is easy for LLMs, but producing maintainable software is the hard part.
- Software engineering requires system thinking beyond writing lines of code.
Boilerplate By AI Builds Tech Debt
- Tech debt arises when code choices make future changes harder and LLMs can worsen that by spitting out boilerplate.
- Relying on generated code without designing for reuse increases long-term maintenance costs.
Models Mirror Web Code History
- LLMs default to internet patterns, so they reproduce both good and bad coding styles.
- Languages with messy historical baggage, like JavaScript, expose models to lots of bad examples.