The loudest pitch for AI coding tools goes something like this: your developers will write 10x more code, ship 10x faster, and maybe you won't need as many of them. A recent analysis from developer Matt Olson makes a compelling counter-argument: the most valuable thing AI does for software teams isn't writing more code. It's catching mistakes in the code you already have.
Olson's piece draws a parallel to past tech hype cycles like Agile and object-oriented programming. The pattern is predictable: massive overpromising, followed by a correction, followed by the genuinely useful parts surviving. He thinks AI-assisted development is heading down the same path, and his breakdown of what's real versus what's noise is worth reading closely.
The "Everyone's a Coder Now" Problem
One persistent claim is that AI tools will turn non-developers into builders, eliminating the need for specialized engineering talent. Olson pushes back hard on this. What actually happens, he argues, is that "everyone is expected to be an engineer at the expense of their other competencies." Your product manager shouldn't be debugging Python. The fact that a chatbot can generate a script doesn't mean your marketing lead should be shipping production code.
This tracks with what I've seen. AI coding tools like Cursor, Claude Code, and GitHub Copilot are genuinely powerful, but they work best when operated by someone who already understands software architecture, error handling, and system design. The tools amplify existing skill. They don't replace it.
Counting Lines of Code Is Still a Bad Metric
Olson takes aim at organizations measuring AI's impact by code output volume. "Counting lines of code has long been recognized as a bad metric," he writes, and AI makes the problem worse by making code almost free to produce. When you can generate 500 lines in seconds, the bottleneck shifts entirely to judgment: deciding what to build, how to structure it, and what to leave out.
This is the gap a lot of AI productivity claims fall into. The headline number ("40% more code!") sounds impressive until you realize that more code often means more surface area for bugs, more maintenance burden, and more complexity that someone has to understand later.
Where AI Actually Delivers
The strongest part of Olson's argument is his case for two specific use cases where AI genuinely pulls its weight.
Rapid prototyping for validation. Instead of spending weeks building something to test a hypothesis, AI lets teams spin up rough prototypes in hours. The key distinction: these prototypes exist to validate decisions, not to ship. You build fast, learn whether the idea works, then build the real thing properly if it does.
Code review and quality assurance. This is Olson's sleeper pick, and I agree with him. AI is "extraordinarily good at reviewing code, finding bugs, and writing tests." Most of the conversation around AI coding focuses on generation, but the review side is where the reliability gains actually live. An AI reviewer doesn't get tired at 4 PM on a Friday. It doesn't skip edge cases because the PR is 800 lines long.
Olson's core prediction: organizations will eventually realize that "AI's ability to raise the quality and reliability bar is just as, if not more, valuable as its ability to allow them to ship faster." That's a bet I'd take. The teams getting the most from tools like Cursor and Claude Code right now aren't the ones generating the most code. They're the ones using AI to catch problems before those problems reach production.