When the Senior Dev Raises an Eyebrow at Your AI-Built App
You've just shipped something. It works. Users are clicking around, data is saving, the thing does the thing. You're quietly chuffed.
Then you show it to a developer colleague — someone with ten or fifteen years of experience — and they go a bit quiet. Not impressed-quiet. Concerned-quiet. They start asking questions like "where does this validate the input?" and "what happens when two people submit at the same time?" and you realise you don't fully know the answers. They're not being cruel. But you can feel the gap opening up.
So who's right? Are you ahead of the curve, or are you standing on a trapdoor?
The honest answer is: both things are true at once. And understanding why that's possible is probably the most useful thing you can do before you build another feature.
What you're actually experiencing
The AI coding experience gap is real, but it runs in both directions. Non-developers and junior builders often get better results from LLMs on first contact than experienced engineers do. That sounds paradoxical. It isn't.
When you ask an AI to "build me a form that saves customer enquiries to a database," you get a working form. It looks right, it behaves right in your tests, and you move on. A senior engineer asking the same question is simultaneously evaluating a dozen things you haven't yet learned to see: input sanitisation, SQL injection risk, error states, what happens under load, whether the database schema will make sense in six months. They're not just looking at whether it works. They're looking at whether it holds.
This is why LLM results feel inconsistent depending on who's using them. It's not that the AI is producing different outputs. It's that different people are evaluating those outputs against completely different checklists — one visible, one invisible.
The vibe coding reality check that nobody really says out loud: the AI is very good at producing code that looks finished. It is often not very good, without specific prompting, at producing code that is finished in the way a production system needs to be.