Artificial Intelligence is not intelligent

TTL 62

Artificial Intelligence is not intelligent. It’s relentless.

Sorry to spoil it, but the name sets the wrong expectation. These systems are not smart. What they have is something I don’t: infinite energy, discipline, persistence. The most interesting thing AI has produced is not a clever answer, but a new kind of digital entity, working without us.

Let me explain how we got here.

We grew accustomed to bots and crawlers browsing our websites. Silent, predictable, rule-bound. Already back then, there was no intelligence involved — just relentless execution of simple rules. But about three and a half years ago, something new started: we began chatting with our computers. Before, we would write code and they would do things. Suddenly, we could talk, and they would respond. It felt like a leap. And in some ways it was — but the fundamental nature hadn’t changed. We ask questions, they answer. If we stay silent, they stay silent. So we expanded the conversation — giving them context, then tools, then access to our email, calendars, documents.

(Not everybody is that reckless, but I am.)

Each step made them more capable, but not more autonomous. Every single action still required me to trigger it.

The problem was me.

I can only have so many conversations at once. I had become the bottleneck — the one piece of the system that couldn’t scale. Not because the technology wasn’t capable enough, but because it was still waiting for me to move first.

The fix was a change of nature, not of degree: instead of giving instructions, give a goal. The system has context. It has tools. Let it attempt a step, check if it succeeded, loop back if it hasn’t, and try again. In a loop like that, it can achieve complex goals without requiring our permanent supervision.

That’s what an agent is: a prompt given to a large language model, with a set of tools, a goal, and the capacity to check its progress and try again. It might sound complicated. It really isn’t — building one requires no coding, no programming, only talking to your system.

And this is where the relentlessness finally becomes the point.
Agents are not smart. They don’t understand what they’re doing in any meaningful sense. But they will attempt that step, check the result, and loop back ten thousand times without frustration, without distraction, without needing a coffee break. Goals that would be extremely tedious for me cost them nothing. They don’t experience tedium. What looks like capability is really just persistence at a scale no human can match.

Hopefully, we’ll bring them what they don’t have: the intelligence.

If someone forwarded this to you, you can subscribe.

I also publish on paolo.blog and monochrome.blog.