It passed all the tests it could and then broke a piece of functionality that didn t seem to have any test coverage but certainly had manual QA. It was clean code. That was the issue, clean code that was also confident is hard to be suspicious of when compared to code that doesn t look good at all.
Now I have agents document all the things they cannot test before I allow them near anything that s core. Slow process, but no more surprises.
There's a specific moment that happens with almost every project: the agent produces code, it passes your tests, everything's green, and there's a real pull to just move on to the next thing. Totally understandable, the whole appeal of this workflow is speed.
But there's a gap between "this works" and "I understand why this works," and that gap is where the real risk hides. Not because the code is wrong, most of the time it isn't, but because the next time something breaks nearby, you're debugging code you never actually read closely the first time.
during the BC (before ChatGPT) era, this gap barely existed. You wrote it, so you understood it, mostly by necessity. Now understanding is a separate step you have to choose to take, not something that comes free with writing the code yourself.
One habit that closes that gap fast: before merging anything non-trivial, ask the agent to explain its own implementation back to you, line by line, in plain language. Not as a formality, actually read the explanation. Two things tend to happen. Either it confirms your understanding and takes thirty seconds, or it surfaces an assumption you didn't know was baked in, which is usually the more valuable outcome.
The tools keep getting better at the first week. Describe an app, get an app. Impressive and the demos aren't lying.
What I keep hearing about (and lived myself) is month two. You ask for one change and five files move. Styling drifts. The AI forgets decisions from last week because the only place your app's structure exists is the chat history and the chat history runs out. Every builder community has the same complaints in almost the same words.
So, honest questions for people actually shipping with these tools:
I ve noticed that my workflow has changed completely over the last year. I rarely start a new project with a blank file anymore. Instead, I pick a template, reuse snippets, or let an AI helper suggest the structure and then I just vibe my way through the build.
It s faster, but sometimes I miss the old blank screen energy, when every line felt handcrafted.
I ve run into this with longer agent runs and I m curious how others handle it.
The original issue might contain the scope, constraints and evidence required. But when another model or session picks up the work later, those conditions often get reduced to a summary.
