The first generation of operational AI learned to answer. Questions in plain language, answers with the source attached, no invention. It was a leap — and it was the easy part.
The second generation learns to act: it opens the ticket, adjusts the setpoint, escalates the incident — always with human approval, always with a record. We're there already, and that's where most of the hard work lives.
The third is the one that really matters, and it's where we're looking now: anticipating. Not answering the question you asked, nor running the task you ordered — but seeing the problem before it exists and putting it in front of you with the action already proposed. The chiller that's going to fail in two weeks, read in the signature of the data before the physical alarm. The demand peak that's coming, and the warning to shed load before it hits the bill. The consumption anomaly that hasn't cost anything yet, caught while it's still cheap to fix.
The temptation here is to promise full autonomy: the AI that "handles everything on its own". We don't go there, and it's not modesty — it's engineering. In operations and compliance, the cost of a confident error is too high to take the human out of the equation. Our principle doesn't change with the generation: the AI anticipates and proposes; the person decides; the system records. Anticipation without approval isn't intelligence — it's automated risk.
And there's an honest technical difference between "predicting" and "anticipating usefully". Predicting a number is easy and almost always useless. Anticipating in a way you can act on means knowing the context — which equipment, in which building, with what history, with what impact if nothing is done — and reaching the right person with the decision already half made. It's the difference between a chart that says "this will rise" and a warning that says "this will fail on Friday, want me to handle it?".
The next step for operational AI isn't talking better. It's seeing earlier. Answering was then; acting is now; anticipating is what we're building — always with the person having the last word, because they're the one who signs underneath.