Agency in an AI World

There's a certain irony unfolding in the software community: as AI systems become more capable, the human value in the system is shifting towards the ability to want things clearly and acting on that want deliberately. This is to say, having agency inside the system itself.

What Agency as a skill means

It isn't just motivation and work ethic. It's a particular set of capacities: identifying what matters, setting a direction without being told what to do, dealing with ambiguity and still being able to act, owning the outcome, and course correction (again, without being prompted to do so).

If most people were honest with themselves, then they'd soon find that this "muscle" has largely atrophied. Most of what feels like agency today is actually just well-executed compliance - following a structure someone else has built, in a role someone else has defined, toward goals that someone else has set.

The reason this hasn't been exposed as a deficit is that institutions have long done the heavy lifting. Your employer defines the problem. Your manager sets the priority. Your process tells you the next step. Competence, under this regime, means executing well within a given frame.

Displacement via Automation

The commonly talked about fear is that AI will take jobs. The more precise truth is that AI dissolves the scaffolding that made it easy to work WITHOUT agency. What are you to do when the model can draft that email? When it can write that code 1000x faster than you can? When it can analyze far more data than you're capable of understanding? What's left is the judgement layer: deciding what email to write, what the code should be accomplishing, what the data analysis is actually for.

This is a point where people hit a wall. Execution is comfortable because it's legible. You know when you're done, and you know if you did it right. The judgement is uncomfortable because it's open-ended. It requires you to have a point of view, to be able to defend it, and ultimately, live with the outcome should you be proven wrong.

The divide

As far as I can tell, we're likely heading towards a labor market with a sharper-than-ever split between people who can direct AI systems and those who are waiting to be told what to do by those systems. However, those waiting to be told will find out that the entity capable of telling them is now a model and that it doesn't need them to do the task anyways.

Those who will thrive in this new paradigm aren't necessarily the most technically savvy. They are the ones who can walk into something undefined, develop a clear sense of what success should look like, and then marshal whatever tools are available. Pushing until something real happens. That's a personality and practice more than it is a credential or technical ability.

Why Agency is underdeveloped in most people

School. That's it.

Okay, for real though. School is a 16 year long training program in the opposite of agency. You're told what to learn, when to learn it, how you'll be evaluated, and what success looks like. The same could be said for early careers too. By the time someone has reached the level where genuine initiative is expected, they've been so thoroughly beaten into compliance that initiative feels like overstepping.

The problem, then, is structural, and AI is going to make it visible in a way that's truly disruptive. Not because robots are coming for the jobs, but because the veil of comfort around structured execution is going to be lifted. A lot of people haven't built the muscle memory to function without it.

What developing Agency looks like

It's mostly practicing under discomfort. Trying projects without clear success criteria. Making the decisions yourself without complete information and not waiting for permission to do so. Building things nobody asked for to see if they matter. Getting comfortable with the anxiety of not knowing if you're doing the right thing and acting on it anyway.

The people building this now, sometimes by accident, are the indie devs, hackers, early startup employees, researches who create their own agendas. Anyone who has ever had to generate their own momentum without institutional backing. Their advantage in an AI saturated economy isn't that they know more. Rather, that they're already used to operating without scripts or permissions.

In a world where AI can do almost anything you can clearly specify, the most valuable thing you can do is clearly specify what matters — and then care enough to make it happen. That's Agency. It has always been a skill. Now it's becoming the skill.