Ask HN: Do you know of any company that went back to hand-written code?
I'm curious about the current phase of the LLM hype cycle and whether reluctance to rely on LLMs is economically viable for an average developer. Many companies have moved from 'use LLMs for anything and everything' policies to limiting budgets and scopes of approved usage. I wonder if some companies have decided that generating more code faster doesn't meaningfully contribute to their bottom line and have foregone LLMs in favor of human authoring. I'm not trying to point out merits or demerits, just want to know if any companies have actually gone back to the old ways.
I don't know of any company that has gone back to hand-written code entirely, but I know of teams that have significantly reduced their LLM usage after realizing the code quality issues and the cost of reviewing generated code. They found that for complex, legacy codebases, the overhead of verifying AI-generated code outweighed the speed benefits.
We actually reverted to hand-written code for our core authentication module. The LLM-generated code had subtle security flaws that our security review caught, but it took more time to fix than writing from scratch. For critical security-sensitive components, we now prefer human authoring.
It's not about going back to hand-written code, but about finding the right balance. We still use LLMs for boilerplate and tests, but for business logic and architecture, we rely on senior engineers. The key is to use LLMs as an assistant, not a replacement.
I think the question is premature. The LLM hype cycle is still in the 'peak of inflated expectations' phase. Many companies are still experimenting, and it's too early to see a clear trend of reverting. But I've heard anecdotal stories of startups that went all-in on AI-generated code and then had to rewrite everything when they hit scale.