The Hardest Room

A while back, The Wall Street Journal ran a piece arguing that the coolest job in tech might actually be in a bank. I read it and I've been turning it over ever since — partly because I have that job, and partly because I think it gets the why wrong. Banks didn't get cool. They got forced.

Why finance went first

Money at stake. Data that can't leak. Regulators watching. And professionals who will not tolerate a tool that wastes thirty seconds of their day. That's the room. Nobody wanders into it for fun.

Finance went first not because it was ready — nobody was ready. It went first because it had the sharpest reason to move: the money, the data stakes, and the regulatory pressure to force the issue. The order of AI adoption across industries was never about which one was "ready." It was about which one couldn't afford to wait.

The problem was never the model

Here's what the hard room actually teaches you: the models are the easy part. They're extraordinary, and they get better on a schedule I don't control. The hard part is the last mile — a skeptical expert, a live deadline, and a new way of working that has to beat the way that already works. Technology without adoption is just an expense. And adoption without sustained use is just a pilot with better PR.

That's not a finance problem. It's a human one. I've watched more than 900 professionals cross that line — the moment someone goes from skeptical to "wait, this changes everything." That moment is the whole job. Everything else is infrastructure for it.

The playbook travels

So the question the article should have asked isn't why the cool AI work landed in banks. It's which domain is next. Science, healthcare, law, the public sector — each is approaching the moment finance hit a few years ago, when waiting stopped being the safe option. And what they'll need at that moment isn't a better model. It's the playbook the hard rooms already wrote: meet people inside their real workflows, remove the friction before you ask for the behavior change, make governance the steering instead of the brakes.

I watched that playbook produce 98% voluntary adoption and 83% sustained active use in one of the most regulated, skeptical environments there is. The number that matters isn't the 98 — it's the word "voluntary." Nothing that produced it was finance-specific. That's the point.

It cuts both ways

There's a blind spot in all of this worth naming. The adoption problem looks completely different depending on which side of the table you sit on. From inside a lab, adoption is a metric that moves or doesn't. From the customer's chair, it's a Monday morning: a skeptical expert, a live deadline, and a demo that has to survive contact with real work. Most of the people building these models have never sat in that chair — and it shows in what gets built. The distance between a great demo and a changed habit is the least understood distance in AI.

I've spent the last few years in that chair — the one being sold to, the one accountable for whether any of it lands. Odd vantage point: close enough to the frontier to see what's coming, close enough to the users to see what actually sticks. Deploying AI and building it are, in the end, the same problem wearing different badges.

So maybe the coolest AI job is in a bank. Or maybe the industry was never the point. The piece buried its own best insight: the most interesting place in AI isn't where the models get built — it's where they meet the people who have to trust them. That's the hard room. It moves. The work stays the same.