The last 20% of an AI-built app is almost never 20% of the work — but it is a knowable amount of work, and the first useful step is finding out exactly what it is. That means an honest inventory of what's genuinely finished versus what only looks finished.
AI is extraordinarily good at the first 80%: the screens, the models, the obvious flows, the parts with a clear shape. It is much weaker at the remainder, because the remainder is mostly integration with things that are specific to you — your payment account, your domain, your data, your users' actual behaviour.
That's also the part with the least available context. There is no generic right answer for how your refund flow should work, so nothing can be generated confidently. It has to be decided, then built.
Recognisable? Most projects at this stage have four or five of these outstanding.
I go through the app and record what's finished, what's stubbed and what's a placeholder. You get a real number instead of a feeling.
New work matches the conventions already in the codebase, so it stays coherent rather than becoming a second app inside the first.
Done means running on your infrastructure, reachable by your users, with the environment configured correctly.
Usually not. Rewriting is the most expensive option and it’s rarely the one the code actually calls for. I’d rather find the specific things that are broken and fix those.
Yes, and it’s one of the most common reasons people get in touch. "It works locally" is almost always an environment, configuration or build problem rather than a code problem, and those are findable.
You send me what you have and describe what’s going wrong. I go through the application and the code, then come back with what I found, what I’d fix first, and what I’d leave alone. You get that in plain language, not a report you need a developer to interpret.
The service
Most of it exists. The last stretch is the part where the AI stopped being useful — the integrations, the edge cases, the settings screen nobody wants to build, the deployment that never quite worked.
Nine differences between your machine and production, in the order they break: env vars, build output, ports, database, CORS, paths, time, memory and secrets.
Read
9 min read
The last 20% of an AI-built app is integrations, edge cases and deployment. Take an honest inventory, wire real payments and auth, and get it live.
Read
9 min read
Sometimes it needs an engineer.
Send the messy version. That’s usually the useful version.