What I actually think about “agentic AI” — from someone who uses it and refuses to stop paying attention.

I’ve watched a word take over the room this year. Agentic. It’s on every deck, every launch, every breathless thread. And I’ve noticed my own reaction split cleanly in two directions. Both of them tired.

One half wants to roll my eyes — I’ve been in the actual work for five years, in digital tools for twenty-five before that, and I know the sound of an industry falling in love with a new word. The other half knows better than to dismiss it, because underneath the hype there is, this time, something real. And I’ve learned that the moment I most want to look away from something is usually the moment I need to understand it.

So I sat with it. Here’s what I found, and why I think it matters more for the rest of us than the marketing lets on.

What the word is actually pointing at

Strip away the noise and the idea isn’t that complicated.

Most AI you’ve used answers you. You ask, it responds. One question, one answer, done. However good it is, it doesn’t do anything — it makes something and then waits for you to come back.

An agentic system chases a goal instead of answering a question. It reasons about what needs to happen, takes an action, looks at what came back, decides the next move — over and over, until it’s done or hits a wall. The magic word everyone’s chasing isn’t smart. It’s loop. The machine gets to act, see the result, and adjust.

If you’ve followed anything I’ve written about my own practice, that shape might feel familiar. My work doesn’t live in a single prompt. It lives in the iteration — the layering, the back-and-forth, the thing I make reacting to the thing I made a second ago. The workflow is the medium. Agentic AI is that same loop turned outward and handed to software: not one output, but a cycle of acting and adjusting toward something.

I don’t say that to make it sound magical. I say it because once you see the loop, you stop being impressed by the word and start being able to think about the thing.

Why the hype is the tell

Here’s what the launch posts skip.

When something genuinely new arrives, it usually shows up quietly and gets a name later. When something gets a name first — a name that suddenly appears on every product at once — that’s a signal the label is running ahead of the substance. Not always. But often enough that I treat a wall of identical buzzwords as a reason to look closer, not a reason to relax.

That’s exactly what’s happening here. Real agentic systems exist and are doing real work. There’s also a lot of ordinary software wearing the word like a costume. The hype isn’t proof the technology is fake. It’s proof that the gap between what’s promised and what’s delivered got wide enough for people to hide inside it.

Refusing to understand the technology doesn’t protect you from that. It just makes you easier to sell to.

What it actually changes — for people, not systems

This is where I stop being neutral. Because this is the part I care about.

A loop that acts on its own moves decisions away from people and into a process that runs without a hand on it at every step. On the right problem — something repetitive, high-volume, checkable — that’s genuinely good. It takes the grinding, soul-flattening parts of work off a human’s plate. I’m not romantic about drudgery. If a loop can do the reconciling and the routing and the sorting, let it.

But a loop doesn’t know when it’s wrong. It can run its whole cycle, hand back something that looks clean and finished, and be completely mistaken — no flag, no hesitation, none of the friction a person feels when something’s off. The polish is not the proof. And when you point one of these systems at a decision that lands on an actual human life — a claim, a benefit, an application, a flag on someone’s record — the question of who is accountable when the loop is confidently wrong stops being technical and starts being moral.

That’s the conversation I want us to be having. Not “is agentic AI real” — it is. But: where are we putting it, who does it affect, and who answers for it when it fails. Those are not engineering questions. They’re questions about power and responsibility that I’ve been circling in everything I make, and the technology only makes them more urgent.

Why I bother understanding any of this

I could make my art and never learn what’s under the hood. Plenty of people do — the boosters who don’t want to know the limits, the critics who won’t learn the mechanics because knowing might complicate the anger.

I don’t get to do either. Understanding how the thing actually works is the advocacy. You cannot fight for a humane version of a technology you refuse to comprehend — you’ll aim at the wrong target, get talked out of your own position by anyone with a slide deck, and mistake the marketing for the machine. The people who will shape how agentic AI lands on the rest of us are counting, a little, on the rest of us not bothering to look.

So I look. I use these tools and I stay awake to what they are. That’s the only honest place to stand right now: close enough to use it, clear enough to name what it can’t do, and unwilling to hand the whole conversation to the people selling it.

The loop is real. So are we. I’d like us to stay in the room.

─────────────────────────────

I make art with AI and think out loud about its future — for humans — here and in the studio. If this resonated, there’s more of it in the journal.