I've seen this hype cycle before. This one is different, mostly.

I started building enterprise software when "digital transformation" still meant getting a company off paper. Since then I've lived through a long parade of things that were each going to change everything: service-oriented architecture, big data, the move to mobile, blockchain, a dozen flavors of cloud. Some of them did reshape how we work. Several were mostly noise wearing the costume of a revolution. After enough cycles, you stop reacting to the volume and start watching for a particular signature.

So when people ask whether I think AI is overhyped, my honest answer is: yes, and also it's the most consequential shift I've worked through. Both things are true, and holding them at once is the whole skill.

How I tell a real shift from a loud one

The test I've come to trust has nothing to do with how impressive a demo is. It's whether the technology changes the unit economics of something people already pay for. A real shift doesn't just do an old thing slightly better. It makes a previously expensive thing cheap, and that price change unlocks behavior that wasn't viable before.

Big data passed that test in places and failed it in others. Blockchain, for most of the uses it was sold for, never did; the demos were striking and the economics never moved. AI passes it almost uncomfortably well. Tasks that used to require a skilled human hour now take a skilled human minute, or no human at all. When the cost of something falls by an order of magnitude, the second-order effects are where the real change lives, and those are always harder to see than the demo.

A real shift doesn't do the old thing better. It makes an expensive thing cheap, and the cheapness changes what people do.

What the scar tissue is good for

There's a version of experience that just makes you cynical, and I try to avoid it. But there's another version that's genuinely useful right now. Having watched several hype cycles up close, I've developed a decent nose for the gap between what a technology can do in a slide and what it can do in production, on a Tuesday, when the data is messy and the users are busy and nobody read the manual.

That gap is where most AI projects are quietly dying right now. Not because the models can't do the work, but because the work of getting a model into someone's actual workflow, reliable and trusted and compliant and fast enough to matter, is unglamorous and hard, and it's exactly the part the hype skips. Twenty-two years of shipping software to enterprises turns out to be good preparation for that part specifically.

Where I'm placing my bets

The companies I've chosen to build and advise share a pattern, and it isn't an accident. None of them are trying to invent the underlying intelligence. They're taking capability that now exists and doing the difficult, specific work of making it deliver in a real domain: hiring, commerce, banking, capital markets. Verified hiring instead of CV theater. Commerce that acts on intent in the moment. AI that runs inside a factory or a lending workflow without demanding a research team. Market intelligence that turns a hundred-page filing into a signal you can use before the market moves.

That's the bet underneath all of them: the value in this cycle won't accrue mainly to the people who build the most powerful models. It'll accrue to the people who do the patient work of fitting those models to the contours of a real problem. The model is becoming a commodity. The judgment about where and how to apply it is not.

I could be wrong about the details; anyone claiming certainty about how this plays out is selling something. But the shape of it feels right, and at this stage of a career the interesting question isn't whether to participate. It's where to spend the years you have left building. I'd rather spend them on the unglamorous middle, turning genuine capability into outcomes that hold up on a Tuesday. That's the part the hype always skips, and it's the part that has always mattered most.