Post #2568345

2026-05-09 12:44 UTC

And that brings me to the second point: if the focus is on productivity and volume, then it's not on quality. That is, things like reliability, usability, security, privacy, efficiency, etc. More code and overwhelmed code reviewers just means more undetected bugs and flaws. We're already seeing this with the Claude Code leak or GitHub's recent instability. Major software services are having failures, right and left. This gets worse if coding agents are used irresponsibly. They generate software, but don't do engineering. If you let them run the show they'll make something that looks right but which isn't "up to code," so to speak, and will fail in ways people don't expect. For instance, Claude Code itself is riddled with critical security flaws, some of which are shocking to see with supposedly "professional" software deployed at this scale. It's negligent, honestly. (3/4) #ai #llm #programming

Replies (1)

@ngaylinn@tech.lgbt 2026-05-09 12:44

Then there's the culture problem. Most obviously, if you replace junior engineers with LLMs, then there is no pipeline for senior engineers. Worse, the seniors who now spend all day doing LLM code reviews may be losing their skills, getting bored, and not paying attention. Certainly, it seems many of them hate this change to their job, and are leaving because of it or actively sabotaging their company's AI efforts. If my quality concerns are valid, then there's also the slow erosion of software engineering best practices, and the loss of institutional knowledge that keeps that alive. I'm honestly afraid that Silicon Valley will forget how to make good software well, and will just resort to continually hacking up garbage code and garbage products to look "innovative" while becoming less and less reliable, safe, and efficient. I worry that this is already well underway! (4/4) #ai #llm #programming

Open ##2568346