This prototype changed how I think about testing AI products
Ben Bauman at HATCH (a Character Capital portfolio company) recently showed me one of the smartest prototypes I’ve ever seen — and I think it’s an important pattern for the future.
Some context: I’ve helped 300+ teams make and test prototypes to test business hypotheses...
90% of these follow a similar pattern — a realistic-looking simulation of a frontend that teams use to decide whether it’s worth building a technical or business backend (using “backend” here to mean software, logistics, data, or even a business process that’s not trivial to stand up).
The team at Hatch was in a different situation.
For their new AI-powered donor scoring, they already knew — through previous prototype testing and customer interviews — that customers REALLY wanted comprehensive and reliable scoring of prospective donors.
In other words, they had answered the question “Do our customers want AI donor scoring?”
Now they were prototyping the backend — experimenting with models, prompts, pipelines, and evals.
It was time to answer the question, “Can we reliably produce AI donor scores that are accurate and useful?”
But they needed a quick way to display and interact with those scores. And here’s where that clever prototype came in…
Instead of building a functional frontend or relying on rudimentary plain text views, Ben created a dashboard in Coda — it could pull real scores from the prototype backend and provide visualizations, filters, and sorts with virtually zero design or dev effort.
In the near future, you might write code for the test frontend instead.
Either way, here’s the pattern:
Identify the tough question or hypothesis, and focus your efforts on getting that hard part right.
Isolate it from the other aspects of the problem space.
For the rest (the easy parts), leverage AI or other tools to keep these lower-risk areas low cost.
I’m curious… what are the most clever prototypes you’ve seen recently? How do you think vibe coding and AI tools will change this?