If someone pitches you on an AI system, make sure you ask them how it learns. You don’t need all the technical details. Just ask how it handles new or updated information over time.

This will let you know whether they’ve grappled with the core challenge. Do they have a genuine solution, or are they hand waving past one of the hardest problems in AI? Are they being honest about the tradeoffs?

It will also reveal their understanding of deployment and maintenance realities. How a system learns impacts everything from compute costs, to model accuracy, to security, and ultimately user trust.

Because no matter who built it, every model is out of date the moment you deploy it.