The Best Way to Show It Is to Show It
Someone asked me recently how to get an AI tool built inside their organization.
I would just build my own. I would build the prototype and show them exactly what that is, and then I would bring the prototype to the team. I think the best way to show it is to show it.
A description gets you a different tool in every head
People can't evaluate a system from a description. A spec or a deck asks every reader to imagine the tool, and every reader imagines a different one. Then the meeting is spent sorting out which version you meant.
A rough working version skips that. Someone clicks it, it does the thing or it doesn't, and the feedback is about what is on the screen. I made a smaller version of this point in Not Everything Needs to Be a Deck, where a short doc replaces the slides. Here the prototype replaces the request. You stop asking for the tool and hand over the first version of it.
Step one is a list
Step one is to list the tools you already use and the real data pulls you already do. Then build an MVP with one or two of them. One or two is enough for someone to watch it work, and that is all the first version has to do.
Know the limits before you promise anything. It can't create APIs that don't exist. Sometimes it's actually faster to just go into the platform and pull it, and it's fine to say so. When there is no API, you can very easily build skills that control your browser, so a missing API doesn't have to end the idea.
It also helps to be plain about what the model is doing. It takes stuff it already knows and makes the connections. So the prototype should connect the tools and data you already have, and you should say that out loud when you show it.
The harder part comes after
The issue is what happens once it works. How do you take the one thing that's built over in this corner and scale it out?
People build useful things in their own corner all the time, and those things stay there. Most of their colleagues just aren't familiar with the tools yet, and they have never watched one of them run. A message saying "I built a thing" doesn't change that.
So I would handle the spreading the same way as the asking. Put it in front of people, live, and let them try it. In Build Something Nobody Asked For I wrote about a platform that got demoed at an AI club I help run, and we tested it right there in the session.
If there is a tool you want built, write the list this week and build the smallest version that runs. And if you have gotten something out of one corner and across a whole team, I'd like to trade notes on how you did it.