AI - Generated Code
Posted by Internal_stability@reddit | Python | View on Reddit | 13 comments
Hi guys do you trust AI generated code in production? How do you check it? Let me know where it breaks for you and where it works perfectly
ultrathink-art@reddit
Depends on the task scope. Short isolated functions with clear inputs/outputs — yes, I trust it. Long chained tasks where the model has to track state from 50 decisions ago — I review much more carefully. The failure mode isn't usually wrong logic, it's accumulated shortcuts that each look reasonable in isolation.
dave-the-scientist@reddit
Nobody should trust anybody's code in production without reviewing it.
sajjad_sui@reddit
real
Prudent_Psychology59@reddit
well said
Particular-Plan1951@reddit
Your_mag@reddit
I think this question usually comes up when someone is newer to coding or experimenting with “vibe coding.” In real production environments, experienced developers don’t just deploy AI-generated code without reviewing and understanding it first.
AI is great for speeding up prototyping or exploring ideas quickly, but when it comes to building something reliable and maintainable, you still need proper expertise, testing, and code review. Otherwise, you’re just taking unnecessary risks.
AI
modern-dev@reddit
I always try to understand the code, once I understand what is going on, I also learn something new
and also feel safe about my code.
AssociateEmotional11@reddit
I don’t trust and I am trying so bad to change the way people use AI for coding by developing an cli tool that help everyone AI generate that make the code safeter and cleaner It is named pyneat in GitHub https://github.com/khanhnam-nathan/Pyneat It is still beta so if you are interested please leave a feedback for me , thank you!
BranchLatter4294@reddit
Same as with any other code. You review it and you run unit tests.
BigTomBombadil@reddit
Review, test, validate. Like any other code change.
This is a simpler task if you’re not committing thousands of lines of AI code at a time or trying to release a whole vibe-coded project.
Interesting-Frame190@reddit
Do we - yes. How does it work - it doesn't, its in the backlog to fix
marlinspike@reddit
You sound like you don’t use tests. Thats a non-optional requirement of going to prod. As is real CU/CD.
I haven’t written code in a few months - Claude Code does it for me. It’s not vibe, it’s instructions, plans, validation, execution, tests, staging, tests in staging.
If you’re just telling a model to go code, yeah, better today than 6 months ago, but you’re not coding for even a good personal project.
tmclaugh@reddit
How do you check a coworker’s code? Another team’s code? Or even your own? Same rules apply.