Ask HN: Do you know of any company that went back to hand-written code?
TL;DR
Has any company tried AI coding and then gone back to writing code by hand? Hacker News found almost no clean cases. A few startups keep AI out of their core code on purpose. One founder is weighing a rewrite without it. The headline examples, Ford, IBM and Commonwealth Bank, were rehiring after layoffs, not dropping AI.
The discussion
A testy thread. The one answer with named companies got picked apart within minutes. Much of the rest argued over whether AI is the problem at all or just exposes weak teams and leadership. The most useful material came from people describing how fast code piled up past what their team understood.
The cited reversals don't fit the question
The top link-based answer named Ford, Commonwealth Bank and IBM, but replies showed these were rehiring after AI-motivated layoffs. Ford brought veterans back to fix the data pipelines feeding its AI, and IBM rehired in HR, not engineering. Nobody produced a company that adopted AI coding and then banned it. The closest cases were doctors who dropped AI scribes because checking the notes ate the time saved, and companies that never adopted AI.
Ford did *not*. They hired back veterans to better train the AI.
Turns out the gains were eroded by having to check the work and turn verbose prose into an actual note.
Speed outruns understanding
The firsthand reports all describe one failure, code arriving faster than anyone can hold it in their head. One founder's team refactored a prototype-era core and watched it turn messy again, partly because the agent dug deleted code back out of git history. Another commenter watched an experienced team move to 99% AI changes and lose its knowledge of the project. Any bug the model couldn't fix then took far longer than before. A startup working in a deep technical area writes everything by hand and says it has almost no technical debt. A 15-engineer startup keeps AI off its core code because building the mental model is most of the work.
the team collectively lost the project knowledge and any problem discovered that llm could not fix took significantly longer to correct - they were effectively working on a new to them codebase.
having derived the code and built a mental model is something like 90% of the work, the code artefact being 10%.
Blame the tool or blame the team
When the founder said developers kept making a mess with AI, others called it a leadership failure, since good developers won't ship slop. Others said the tools push people to hand over control, and managers can't easily refuse a colleague who ships a feature in an hour. Some proposed spec-driven agent workflows, which others called waterfall again, since a prose spec separate from the code always drifts.
This is a leadership problem, not an AI problem.
The tools encourage you to hand over all control to the AI, taking the user out of the drivers seats resulting in slop.
Just another tool, like compilers and IDEs?
One camp called the question silly. Nobody went back to assembly or plain text editors, and developers already rely on garbage collectors and other people's libraries without understanding them. Critics replied that compilers don't hallucinate, IntelliSense only completes what you meant to type, and AI is no abstraction layer. One commenter added that companies did ban IDEs and Stack Overflow when those were new.
Stepping back from AI completely (vs changing how AI is used in the company) is like stopping using IDEs, compilers etc. It's a *tool*.
I've never used intellisense to input something I wasn't already looking for/about to type out.
Productivity for whom
A side argument asked whether faster output is worth anything. Skeptics said measuring code speed is lines-of-code by another name, and that AI lets five engineers build a legacy codebase in six months. Others said employers asked for speed and got it, so developers who stop caring at work are giving employers what they chose. One employer reportedly counts token use in performance reviews.
Now with AI five engineers can build a legacy codebase in six months.
I don't see how one can hold "speed of code generation" as a good metric while, simultaneously, holding "number of lines generated in a unit of time" as a bad metric.
Consensus
No one could name a company that adopted AI coding and then went back to writing everything by hand. Most agreed the real choice is where to use AI. Core or hard-to-reason-about code is the most common place people keep it out.
Dissent
A vocal minority said asking about reversal makes as much sense as asking who abandoned compilers. Teams producing a mess, they said, have a leadership or process problem, not a tooling one.