When using AI to generate code, I made an observation: it always seems to contain 10% bullshit.

This is observation is the sole reason why I don’t vibe code. Instead, whenever an agent has implemented my prompt, I skim the code trying to sniff out the 10% bullshit that’s always there. If I can’t find it, I get suspicious: did I miss the bullshit? Will I step in it later?!

The reason I’m so careful, is that bullshit scales. Like many people noted online, AI output heavily depends on the input: your prompt, code etc. determine the quality of the response. As a consequence, if you don’t find the 10% bullshit, it will gradually degenerate your codebase and AI coding effectiveness:

Interestingly, a similar phenomenon can be observed when letting ChatGPT re-generate an image 100 times, whilst telling it to not change anything:

Conclusion: if you don’t want this to happen to your codebase, take some time and sniff out the bullshit.