What actually works, what fails in production, and what I am still figuring out - across AI engineering, autonomous systems, sports analytics, and the path from research to shipping.
Agentic AI is not magic autonomy. In production, it is a controlled loop where a model can plan, use tools, observe results, and decide what to do next inside strict operational boundaries.
Pro sports analytics runs on oceans of data. A college season gives you twelve games, a roster that dissolves every year, and no time to wait for statistical significance. Building AI that works here is a different discipline, and a more interesting one.
Engineering has spent two decades building digital twins of jet engines and factory lines. Now sports wants one of next week’s opponent: a simulated rival you can rehearse against before kickoff. Some of that transfers. The most important parts don’t, because a play-caller, unlike a turbine, watches film of you too.