I'm a software architect. I stopped asking what GenAI is and started using what it can do.
The interesting discussion has moved on. I'm moving with it.
Where I stand
I accept GenAI as a form of intelligence.
Whether it's "really" intelligent is beside the point. I look at what these systems deliver, and I work with that.
I define capabilities independently of humans.
Tying terms like creativity, understanding, and intelligence firmly to humans — "creativity is what humans do" — narrows the field and makes new forms invisible before anyone bothers to look. I measure such capabilities by their properties and effects, not by their origin. That takes nothing away from humans. It widens the view.
I don't wait for AGI.
Waiting for a term nobody can define isn't a strategy. It's an excuse. I work with what's on the table today, and that's more than enough.
Pandora's box is open.
GenAI is in the world, and nobody is putting it back. The question isn't whether, but how. I anticipate and shape, instead of waiting for "all this to settle down". It won't.
I don't apply GenAI blindly. I understand it.
Using GenAI sensibly is an engineering discipline. You don't read a specification — you investigate a behavior. So I work empirically and want to know why something works, where it breaks, and when you shouldn't trust it. That protects against cargo cult on one side and fear on the other.
I bring craft, not just enthusiasm.
A language model doesn't replace architecture, clarity about the problem, or discipline in building. That's exactly what I bring. The tools are new. The standard for good work isn't.
I use GenAI for people, not against them.
I don't measure it by what it promises, but by what it does when it costs something. AI that surveils people, demeans them, or floods them with disinformation — or is set loose without human accountability — isn't what I mean. No matter whose logo is on it.
Real control beats the ritual of control.
A mandatory review that ends in review fatigue is safety theater, a checkmark without substance. Where human inspection no longer truly catches things, I build safety into the process: into tests, into architecture, into workflows that catch errors. The human isn't the gate. The human is the one who builds the system so it holds.
The duty to label AI will undo itself.
Today, a marker is supposed to warn: "a machine was here". But once almost everything is AI-touched, the marker means nothing. A warning on everything is no warning. What matters moves from origin to care: not whether a machine was involved, but how it was checked.
In short
Understanding over awe.
Building over debating.
Capability over label.
Action over promise.
Today over someday.
What I want to talk about
If you want to explain to me over dinner why "AI doesn't really understand anything", that's an interesting conversation. But it isn't mine.
If you'd rather show me what you built with it last week: any time.