Built in a week, still growing

A skunkworks bet on AI search that turned into a real, shipped tool

Hudl — AI-Powered Command Bar

During Hudl's annual one-week skunkworks event, my coworker Gabby wanted to build an AI-powered search bar just to learn how to build with AI. I already suspected we'd need real filter-search for our upcoming Volleyball Consolidation work, so we built toward that instead of a demo. A week later we convinced our PM to ship the V1 straight to production. I've spent the eight months since watching real search queries come in to learn what people actually want from it.

Origin: Hudl's annual skunkworks week — one week to build anything.
Early bet: Filter-search we knew Volleyball Consolidation would need soon anyway.
Timeline: 1 week to build V1
8 months of ongoing research since
Team: Me (design), 7 engineers, 2 QA

Where it started

Every year, Hudl runs a skunkworks week: everyone on the product team gets a week to build whatever they want, no roadmap required. This one landed right at the start of AI tools becoming genuinely useful — just before most of us started reaching for them at work every day.

My coworker Gabby wanted to build an AI-powered search bar, mostly to learn how AI tools actually get built. I had a more specific reason to say yes: I already knew, from our upcoming Volleyball Consolidation effort, that we were going to need a search bar that could search and filter. Skunkworks gave us a week to find out if AI could do that well before we had to commit to it on a real roadmap.

Shipping out of a hackathon

We didn't just want a good demo for the skunkworks showcase — we wanted it live.

In one week, with seven engineers and two QA on the project, we built a working V1 of the command bar. That's an unusually large team for a skunkworks project, but it meant we could build something real instead of a throwaway prototype.

Because we already knew this was solving a problem we'd have to solve anyway, we made the case to our PM to skip the usual "skunkworks project quietly dies after the showcase" fate and ship V1 straight to production. We got the yes, and shipped it.

Eight months of listening before building more

Instead of jumping straight into a V2 roadmap, I spent about eight months on this as a side-of-desk project — watching the search prompts people actually typed into the command bar. Rather than guess what people wanted from an AI search tool, I let real usage tell me: what people searched for, where they got stuck, what phrasing they expected to work and didn't.

Prompt patterns and specific findings from this research are going into V2, below.

Where this is headed

This project is still active — V1 shipped, V2 is being scoped from eight months of real usage data, and V3 is already on the radar.

Built in one skunkworks week and shipped to production — AI-powered search with early support for filtering, aimed at the need we knew Volleyball Consolidation would create.

In progress — shaped by eight months of watching real search prompts come in. More detail coming here soon.

Not yet scoped. Check back for what's next.