Code Isn't Dying — It's Becoming More Precise

Steve Krouse, founder of Val Town, published a counterargument to the 'AI will replace coding' narrative that's been gaining steam since the latest wave of agentic coding tools. His thesis, titled 'Precision,' lands at an important moment: we're drowning in demos of AI writing entire apps from a prompt, and the takes about developers becoming obsolete are getting louder.

Krouse's argument centers on a deceptively simple observation: code is a precision tool, and precision doesn't become less valuable when approximation gets cheaper. Natural language is inherently ambiguous. Code is not. When you need something to work exactly right — not 'pretty close' or 'good enough for a demo' — you need the exactness that only a formal language provides.

This resonates because it names the gap that every developer who's used Cursor, Copilot, or Claude Code has felt: AI is extraordinary at generating plausible code, but 'plausible' and 'correct' are different things. The last 10% of getting software to behave precisely as intended is where the actual engineering lives. AI makes the first 90% faster. It doesn't eliminate the last 10% — and that last 10% is why we write code instead of memos.

The piece hit 320 points on Hacker News, which tells you something about the developer mood right now. There's a growing fatigue with the 'coding is dead' narrative, not because developers are in denial, but because they're the ones actually shipping with these tools daily and can see both the power and the limits. The discourse has been dominated by people making predictions; Krouse is describing what practitioners already know.

What makes this more than just cope is the practical framing. Nobody serious is arguing AI isn't transforming development. The question is whether the transformation looks like 'developers become unnecessary' or 'developers become more productive and focus on harder problems.' Krouse is firmly in the latter camp, and the evidence from actual production usage supports him. Companies adopting AI coding tools are hiring more developers, not fewer. GitHub reported Copilot users complete tasks 55% faster — they're not laying off half their engineers.

The uncomfortable corollary Krouse implies: if you're a developer whose entire job is writing boilerplate that AI can generate perfectly, then yes, your role is changing. But that's not because code is dying — it's because the floor for what counts as skilled development is rising. Precision, architecture, system design, understanding failure modes — these are the things AI can't approximate its way through.

For senior developers, this is the piece to send to your non-technical leadership the next time someone forwards a 'no-code AI' demo and asks why the team needs engineers.