i'm no developer. can't write production code from scratch.
but I've been deep in AI-assisted coding for two weeks. Wanted to see if i could ship something real.
So last weekend i built three tools. fully functional. live by Sunday night.
A LinkedIn AI Audit that analyses your profile and post history, then spits out a prioritised 30-day action plan as a PDF.
An AI Scout that checks whether ChatGPT, Gemini, Perplexity, and Claude actually recommend you when someone searches for a coach in your niche.
Weekly Leaderboards across 30 coaching niches and geographies, updated every Monday from live AI responses.
None of that existed on Friday.
Here's what it taught me.
the bottleneck was never the code. it was knowing what i wanted precisely enough to describe it. every time something broke or came out wrong, it was because my spec was fuzzy, not because the AI failed. Humbling realisation when you've been shipping product for years.
second thing: taste is still entirely on you. The tools generate exactly what you ask for. whether that's something people actually want to use - that's a product judgment call no model can make.
vibecoding didn't remove the hard part. It exposed it. The constraint was never syntax. it was clarity.
AI will build anything you describe. the real gap is whether what you described is actually worth building