The demo trap
Outcome
You can tell a demo from a tool, and you know exactly where the real work begins.
Concept
The first time your thing works, it feels like the finish line. It isn't. It's about the 80 percent mark. A demo is something that worked on the one example you happened to try. A tool is something that works on the examples you haven't tried yet. The whole gap between them — getting from "works once" to "works almost every time" — is where the actual building happens. Most people stop at the demo, because the demo feels like success, and then they wonder why their AI projects never quite become things they can rely on. The good feeling of the working demo is the trap.
Weak approach vs. strong approach
The ask
I had Claude Code build a tool that pulls the to-dos out of today's class announcement. It worked. Done.
What comes back
It worked on today's announcement — but that's the only input you tried. Run it on last week's and it drops the task whose deadline was written "by the end of the week," and it leaves two tasks fused in a single sentence. The demo hid both.
Try it
Take something you built that works. Run it on five more real inputs you have not tried yet. Write down every place it is wrong, vague, or misses something. Do not fix anything yet. The point is to watch a single success hide four problems.
Takeaway
A demo works on the example you tried. A tool works on the ones you didn't. Finding that gap is step one — everything after this is closing it.