Automating the Wrong Thing

Most AI-in-ops projects aim at the wrong target. Everyone wants to automate the big, impressive thing, the one that gets a slide at the all-hands. In my experience, that's exactly the thing you shouldn't start with.

The flashy automations tend to be full of judgment, edge cases, and context, which is exactly where AI is least reliable and most expensive to get wrong. You spend months, it half-works, and everyone quietly stops using it.

The payoff is almost always in the boring stuff nobody wants to demo. The waiting. The status-chasing. The small breakages that sit in a queue for two days. The grunt work that eats an afternoon. None of it is impressive, and the unglamorous friction is exactly where AI quietly earns its keep, the last place anyone thinks to look.

So the question isn't "what's the most impressive thing AI could do here." It's "where is my team losing time to friction nobody would miss." Automate the plumbing, not the judgment. The plumbing is reliable, cheap to check, and safe when it's wrong. Judgment is none of those.

The teams getting real ROI from AI aren't automating the exciting things. They're quietly deleting the friction everyone else just tolerates.