Prototype While the Idea Is Still Cheap
AI makes it stupidly easy to spin up a small idea and see if it is actually worth building.
Most small ideas used to die in the same place. Not because they were bad. Because getting from “this might work” to something you could click took a weekend, and the idea had not earned a weekend yet.
That is the bit AI is actually good at. It does not make you a better engineer overnight. It just makes the first version cheap.
Get it on the screen
A prototype is not a tiny product. It is a way to answer a simple question: do I still care about this now that I can see it?
You can throw a messy brief at an assistant. A data model, one screen, a boring CRUD app. And you can have something running before you have talked yourself out of it. You are not asking it to invent the thing. You are asking it to put the thing on the table.
If you cannot reject it, it was not a prototype. It was a commitment you made too early.
This is for small stuff
I am not talking about generating a platform. I mean the ideas that always felt a bit too small to start:
- A decision log you might actually update
- A status page that tells the truth
- A tiny tool for something you do every week
- A spike to check a boundary before you build a real service around it
Those do not need a plan. They need a first path and enough UI that you can tell if the idea has any weight. AI is great at the first path. It is pretty bad at knowing when to stop.
Do not ship the chat
The rough first draft is fine. The problem is treating a fluent first draft like it is finished.
If the spike survives, the next job is the same as it always was. One database. One way to deploy. One source of truth. A short list of things the product is allowed to do. The assistant can help write that. It cannot own it.
Quick test: if you cannot explain the running system without scrolling back through the chat, you do not have a prototype. You have a demo you cannot maintain.
Use it, then throw most of it away
The loop that used to eat a Saturday can now take an hour:
- Say the constraint in one sentence.
- Get a working sketch in a stack you already know.
- Use it. Break it on purpose.
- Keep what survived.
- Bin the rest.
That last step is the whole point. Cheap prototypes only help if you are happy to delete them. The generated code is not the prize. The prize is finding out the idea is wrong while you still have the energy to start again.
You do not need more software. You need to find out sooner whether the software is worth writing.
AI just gets a small system in front of you before the excitement runs out. Keep it, cut it, or start again on the boring path. That part is still on you.