What gets easier when machines can finally watch?

That's the question I've been turning over for years. The honest version is I'm not done answering it. What I am sure of is that sport is the first place I want to find out, because the data is unusually rich and the people who'd use it feel the gap most acutely.

99% of the world's footage sits unused. Youth games on hard drives, college tape no analyst will ever touch, international leagues with no tools at all. The reason isn't that the cameras are bad. It's that turning video into intelligence has historically required proprietary hardware, dedicated analysts, and software priced for the top 1%. The other 99% just lost.

The bet

At AirPLAi we're betting that frontier vision models flip this. Any camera, any sport, any level. Two products power it:

• PLAi Ground, a self-service portal. Upload film, get tagging, searchable stats, and analytics back.
• PLAi Chat, an agentic assistant. Natural language video search. “Show me every pindown three from the last 50 games” returns the clips and the numbers behind them.

The wedge

We're not cheaper Hudl. We're what Hudl would look like if it were founded in 2026 with LLMs and modern CV from day one. The flanking move is youth, emerging sports, and AI-native interfaces. We win where the incumbents can't economically reach.

The asymmetric data angle is referee intelligence. Officiating tendencies, challenge success probability, the patterns that pro teams currently pay consultants for. Nobody else is building that.

The deeper thing

Behind the product there's a thesis I keep coming back to. Software that runs companies, not just tools that help them. AirPLAi is being built by five humans and a fleet of agents. The agents handle the connective tissue. Ingestion, labeling, reporting, follow-ups, internal automations. The five of us handle the irreducible work. Customer taste, founder relationships, judgment calls.

The operating system is part of the product. Maybe most of it, on the days I'm being honest.

Why now

CV in sports is still mostly Wardley-genesis. You build proprietary CV where it differentiates, you rent commodity where it doesn't. The window where being early matters is shorter than people think. The window where being thoughtful matters is longer than people think. Both are open.

What I don't know yet is which parts of this stack end up mattering five years out. The model? The pipeline? The agent layer on top? The specific events we've taught it to care about? I keep changing my mind. That feels like a sign we're in the right neighborhood.

I'll keep working it out on thoughts. If any of this is on your mind too, I want to compare notes.