SOFTWARE ENGINEERING
When the prototype works, the engineering begins
AI-assisted code still needs requirements, tests and someone who takes responsibility.
A working demo answers only one question
A quick prototype can show that an idea is possible. It says much less about security, data handling, failure recovery or maintainability. The speed of generating code should not be mistaken for evidence that those questions have been answered.
Recover the intent
When I review an AI-built application, I start with what it is supposed to do. I compare that intent with the actual routes, data structures and behaviour. The useful work is often to find the gap between a convincing screen and the system behind it.
Leave a maintainable result
The next step may be a small fix, a set of tests, a simpler architecture or a decision not to continue with part of the prototype. A documented decision can be as useful as more code. The objective is work that the next person can understand and continue.