2025: Agency gained and lost

If you’ve had even a passing glance at tech journalism over the past few months, you know the top buzzword for AI in 2025 is agentic. Agentic AI (according to many think pieces and press releases at least) is going to revolutionize industry. It’s going to streamline complex pipelines and maximize efficiency. It will make corporations smarter, better, faster, and for cheaper. What makes agentic AI so powerful? Agentic AI can take action. These models—equipped with language modules, external databases, and ‘chain of thought’ reasoning—are designed to set goals, make plans, and execute actions. As a result they can have real impact in the world.

But what about us?

2025 is not just the year of agentic AI but also the start of a second Trump administration. In the short time since the Nazi-salute-containing inauguration day, the administration has enacted a slew of executive orders impacting everything from prescription drug prices to offshore drilling. ICE agents are conducting raids in previously safe spaces like churches and hospitals and sweeping up US citizens in the process. Government employees are actively encouraged to snitch on their peers. Scientists are being silenced and their funding requests halted or even canceled. Combined with the legacy of the first Trump administration, we are currently facing an upheaval in our access to everything from healthcare to voting to clean energy.

Each of these efforts constricts the space around every one of us, making our possible set of actions smaller and smaller. Thinking of expressing an opinion on diversity at work? Be careful, your co-worked may report you. Want to attend a Spanish mass in Chicago? They’re canceled, for fear of raids. Need to look up a fact about climate change or Jan. 6 investigations? The government websites are gone.

The result (and intention) of these rules and actions are to create a learned helplessness in us—to reduce our own sense of agency. And in fact we see signs of it working. Voter turnout dipped by about 5 million people in 2024 compared to 2020. The Women’s March and other protests that mobilized hundreds of thousands after the first Trump inauguration dwindled to a much smaller trickle this year.

Even the simple act of social media activism has been constrained. The owners of the largest information sharing channels in the world sat as esteemed guests at the inauguration, and their actions within their companies suggest the invitation was a welcome one. X (formerly twitter) won’t be of much use unless the actions you want to take are to the right. Threads has always been clear that it is not interested in political content, preferring to support the brand-rot banter of social media interns at large corporations instead. And should you manage to actually get a useful and actionable post out, say about how women can access Plan C, onto Instagram or Facebook, it will be taken down before many people can see.

So it seems the same people who want to push AI agents on us in their tech also want desperately to remove our own agency in real life. Wouldn’t Instagram be a nicer place if all the pesky, politically-motivated humans just sat back and observed the (fake) actions of AI-generated social media accounts? Perhaps the writers of Wall-E had the most accurate take on the future: an Earth so covered in trash it is abandoned, but nobody really cares because they’re given a constant stream of personalized entertainment without having to lift a finger, or even make a decision. Leave the agency to the robots.

The truth is, it can actually be a lot more fun to be an observer than an agent. I know, because I spent much of my life not feeling like an agent at all. Perhaps because I grew up in a cornfield-filled suburb where nobody really did anything that mattered, I never expected I would either. And so I simply observed. I observed the physical world and human behavior and patterns of thought. And this turned me into a scientist. But even doing science can feel like not really doing anything. A lot of academic scientists joke about getting a “real job” some day.

But through a combination of events personal, professional, and political, my perspective began to change. When my dad died shortly before I finished my PhD I was confronted with the uncomfortable possibility that I was now the adult in the room. When the pandemic struck and global chaos abounded I realized that people in power are—terrifyingly—just like me. And when I published a book and eventually heard from young people that this book impacted them, sometimes to the point of changing their careers, I realized that I was already acting in the world. And so I decided to take on the identity of an agent more strongly.

And this is how I know that living as an agent is both essential and life-giving, and also an unbearable burden. With agency comes responsibility. If you have the power to change the world then why aren’t you? ‘Can’ quickly becomes ‘ought’. Even after deciding to shift my career towards climate change (an agentic move if I ever made one), I still wonder if I’m doing enough. Even when I became directly involved in political campaigning (including overcoming my millennial phone-phobia to make calls for the Harris campaign), there was always the sense that I could have, and should have, done more. If modern AI agents had to experience true agency, with the moral weight that is intrinsically tied to it, their chains of thought would snap under the pressure.

Counterintuitively, this moral weight can feel even more acute the more constrained your action space is. I don’t have billions of dollars to invest in clean energy and so instead I agonize over every decision to take a flight or use a disposable cup. There is a stifling sadness in these small actions. When those with power take action free from moral constraints, you can feel like an ant moving grains of sand in the shadow of a bulldozer. You still can act. In fact you must act, but will your tiny actions matter?

I think a lot about the phrase “May you live in interesting times”, wrongly claimed to be a Chinese curse. If it is a curse, it is only a curse for the agent. To an observer, interesting times are entertainment: think of all the gripping movies that have been written about past times of turmoil and will be written about the current one. To the agent, though, who is actively trying to write the story as they live in it, interesting times are a minefield of potential wrong decisions and wasted efforts. At the same they also present the very real possibility of impact.

When we think of the current moment we have to remember why so many people in power want to constrict our agency. It is obviously because our actions do matter. Uncertain times present many possible futures. We, individually and collectively, do have the power to set the course. If our efforts were like a child playing with a toy cellphone, pressing buttons that lead to no outcomes, there would be no reason to take them away. The game of finding your route to greatest impact has certainly gotten harder, with more hurdles and dead-ends added along the way. But all that means is that we need to be cleverer, more committed, and take more action. In short, we need 2025 to be the year not of agentic AI but of agentic humans, operating with the full moral urgency that comes with that term and that the moment demands.