wtf is an ai agent, actually? (with pictures)

Confused about AI agents? It’s not just you.

Everytime I read an article about them, I grow more confused. What are we actually talking about when we say “agent”?

Everybody seems to mean something different. And that makes it really difficult to understand what they are talking about.

Let’s make sense of this mess. In seven chapters, with infographics. (shout out to Nano Banana 2!)

Here’s what’s coming:

1. The problem: everybody is shouting about agents and nobody knows that they actually do.

2. The word: “agent” carried a lot of baggage, even before there was AI.

3. The fantasy: why we love the idea of agents so much.

4. The question: what does “AI agency” actually mean?

5. The spectrum: the different kinds of AI agents that show up in the wild.

6. The tradeoff: how to think about the cost and benefit of AI agency for your specific project.

7. The plan: the concrete steps you should follow in deciding your agentic AI strategy.

---

For access to the full version of this post and all of this substack’s premium content, please consider supporting my work with a paid subscription.

---

THE PROBLEM: Everybody is shouting about agents and nobody knows what they actually do.

For years now, the world has been screaming from the rooftops about “AI agents” without actually agreeing on what they are. The hype kicked off with pure, irrational exuberance. Early experiments sparked viral claims that fully autonomous entities would soon replace human labor and build six-figure businesses while we slept. The reality? Most just got trapped in expensive, four-hour infinite search loops.

Leading voices aggressively pushed back against the sci-fi anthropomorphism. They argued we weren’t building autonomous “agents,” but rather highly constrained workflows—language models operating inside structured loops with specific tool access.

Fast forward to the present. We are now drowning in stories about OpenClaw, Moltbot, or whatever the flavor of the week is called. Models are taking over our cursors to navigate operating systems and silently manage our inboxes in the background. Yet, the core definitional crisis remains. Ask an engineer, a product manager, and a CEO what an agent actually is, and you will get three conflicting answers. We’re definitely building something. But nobody seems to know what exactly it is.