June 20, 2026
Ctrl-Alt-Delete the robot’s homework
When I reject AI code even if it works
AI can write the code, but the internet says humans still have to clean up the mess
TLDR: A developer says he often throws away AI-written code even when it runs, because fast results can still create confusing, bloated messes. Commenters mostly backed him, arguing that “it works” is a low bar — though some pushed the colder reality that cheap and finished often wins.
A programmer confessed to a very 2025 problem: the robot helper finishes the job fast, but then you have to stare at a giant wall of changes and decide whether any of it actually makes sense. His rule is brutal but simple: if he can’t explain what the artificial intelligence did in plain English, if the change looks way bigger than the problem, or if it makes the system harder to understand, he scraps it and starts over. That instantly lit up the comment section like a reality show reunion.
The biggest crowd reaction was basically: working isn’t the same as good. One commenter said if you swapped “AI code” for “coworker code,” nobody would argue — rejecting code that technically works is just normal quality control. Another brought pure penny-pincher energy with the zinger that “adequate often means done and cheap,” which is the kind of line that starts office fights before lunch. Others piled on with war stories about asking AI to make small fixes and getting back a sprawling maze of extra layers and complexity, only to be met with the world’s most passive-aggressive machine apology: “You’re right to push back.”
There was humor too. One skeptical commenter cut straight to the bone with, “Even if it works? How do you verify that it works?” Ouch. And the strongest vibe from the thread was clear: sure, let AI play in low-stakes side projects, but if it touches the money-making core of the business, humans better know exactly what that code is doing. The bots may be fast, but the comments say trust is still very, very manual.
Key Points
- •The article says AI-assisted development shifts the bottleneck from implementation speed to reviewing large volumes of generated code.
- •The author reports experiencing cognitive overload when reviewing AI-generated diffs they did not fully reason through themselves.
- •The author says they often discard AI-generated changes and restart after taking more time to understand the problem.
- •The article lists specific reasons to reject AI code, including poor explainability, oversized diffs, premature abstractions, and reduced system clarity.
- •The author argues that human review should remain mandatory because code that works locally and passes CI can still be an inadequate engineering solution.