AI And The Horrible Roadmap
"Why should I pay you guys for a roadmap or blueprint project if I can just ask Claude to make me one?"
Glad you asked. Because I've seen what happens when a smart but non-technical person asks an AI chatbot for a feature roadmap on a technical project.
We told one founder we'd want to build that roadmap before rushing into implementation. Wanting to save money, or just to "speed things up," they shared their own 37-page Word document of "user stories." The usual AI tells were all there, the devious em-dash and the general verbosity included. But the real problems ran deeper than the language:
User stories should focus on the user: "As a customer, I want to add items to my shopping cart." This founder's were all "As the system, I want to connect to this API." We'll shelve the debate on whether a system wants anything. The real issue is that the backlog was a list of technical tasks. Create this API connection, ingest that data, compute this measure. None of it says what the user gets out of it.
Stories were sliced horizontally instead of vertically. All the low-level plumbing first, then the data processing, then somewhere around page 25 we'd finally reach anything user-facing. That's risky, because value only shows up at the very end. On any data-intensive project, leaving the end-to-end connection until last guarantees nasty surprises: this data format won't play nice with that analysis tool; oh, and given what we want to show the user, we should have built that pipeline stage back at the start.
The problem isn't so much that the AI created a bad roadmap; it's that the founder couldn't tell.
