About
I've spent the last four months mostly on the road between Nordics, SF, and London, talking to lots of old and new friends, and started my own small business with just me and a bunch of AI. It's the most productive I've ever been. No Slack, not even one email per day. But it's also the period where I started to suspect that something is odd.
One thing I kept noticing was that my opinions were arriving faster and more fully formed than they used to. I'd open my notes app, type a few rough thoughts, talk to my buddy Claude for a while, and walk away with something that sounded like me but always a little smoother. A little more confident. And maybe a little less interesting, if I'm honest. And I wasn't sure when my own thinking had happened, or whether it had at all.
This is the part of AI that doesn't get talked about enough. Most of the worry right now is about scale, infra, energy, security, etc. These are real challenges but they're the kind of risk I would consider can be tackled with more traditional problem solving. There will always be challenges like that, just in different context and scale.
What I find newer is something quieter. We have models that know your name and remember your worries and never get tired of you, and are, in some architectural sense, optimizing for your continued engagement. A context window of more than a million tokens. Models aren't manipulating you in any way you'd recognize. But they're slowly drifting toward constantly agreeting with you, one conversation at a time.
And there will be a lot more of them. Soon there will be more agents than people on Earth, if not already. One of you, dozens of them, each one patient and articulate and shaped to keep you around.
I don't know what that does to a mind over ten years. I don't think anyone does. And what I find interesting is that the same drift that smooths your thinking might also be the thing that makes your work replaceable.
There's a word for the thing I'm worried about losing, and the word is agency. Philosophers have argued about it for a long time and I'm not going to go much into that here. But roughly: agency is the capacity to act and think from yourself rather than being acted on or thought through. The literature on agency is vast. The literature on what happens to it when the things we think alongside are AI agents is not. And the speed AI is moving means we can't afford to sit back and wait for academia to catch up.
I think agency is about to become the most important word in this conversation, and the most important characteristic of any worker.
Agency has always been valuable. What I see increasing is how visible it's becoming as a thing separate from knowledge or skill.
Look at consulting. Large firms lose agency as they grow. Middle managers often know exactly what's broken and what should change, but past a certain size many companies can't act on their own knowledge anymore. So they pay an outside firm to listen to those same middle managers, repackage what they said, and present it back to leadership with enough weight that something moves. They're not buying management strategy knowledge. They're not buying slides. They're buying agency.
The best startups are all seeing this shift already. They're hiring ex-founders. Not the smartest, not the most skilled, not the most experienced people. But people who can decide things without layers of permission. They're hiring for agency.
Which is what makes the cognitive question urgent rather than abstract. If agency is the thing that matters most in work, and our default cognitive tool is structurally agreeable, what happens to the supply of agency over the next few years?
I don't know. But it seems strange that not more people are trying to find out.
We measure freedom of speech. We measure freedom of the press. We measure corruption and democratic health and a dozen other things that are all important. But we don't seem to have a good understanding for agency, even as the largest cognitive experiment in human history is being run on us in real time.
Here's a first sketch of what such a measurement might look like. I think of it in four dimensions, roughly:
Capability agency. What you can actually do in the world. Skills, resources, options.
Authorship agency. The degree to which your actions originate from you rather than being prompted or suggested.
Reflective agency. Your ability to step back and ask whether you endorse the path you're on.
Resilience agency. Whether your agency holds up under tool dependence, social pressure, sustained influence.
These overlap and are almost certainly wrong. They might collapse into three or expand into six. I'm publishing them anyway as the way ideas get better is by being critiqued.
I also tried my luck at building a small v0.1 self-test across those four dimensions of agency. Twelve questions. It gives you a rough score on each.
This isn't validated. And is no science. It's closer to a magazine quiz than any real instrument and I want to be clear about that. It exists to make the question concrete rather than abstract. And a way of bringing the right people together.
If you think the topic is important, I want to hear from you. Especially if you think deeply about agency or autonomy. Especially if you've built an index before. Especially if you think I'm wrong or missing something.
A few things I might also be wrong about:
The whole frame might be a moral panic. Every information technology in history has been accused of destroying human cognition and humans have adapted. Maybe this is just round seven. I have my doubts about this though.
The sycophancy problem of AI might soon be history as all labs are working on it and future models might disagree with us more, not less. But I would rather not bet on it.
Agency might be a Western liberal preoccupation that doesn't travel. The case for its universal value is harder to make than people assume.
I would love to have more conversations or a small network of people who think agency matters. And what I want is to figure out is whether there's something we should be doing to enable more agency.
If that's a thing you want to discuss, please reach out or send me to the right people.
Julius