2026 Is the Year of Agency, but Not the One You Think
Introduction
When people talk about “agency” in ML right now, they usually mean AI agents.
LLMs that plan, call tools, run workflows, do things end to end.
That’s fine.
But that’s not the agency I’m thinking about.
What I think will matter much more in 2026 is personal agency — the very human ability to decide “I’m going to do this” without waiting for permission, consensus, or perfect clarity.
Most people underestimate how much of that they already have and they don’t exercise that muscle.
Agency is not dramatic
Here’s a boring example from work.
I was on vacation.
A training run needed to be started. If I waited until I was back, I’d lose a few days of data.
Data points of the real world inform every subsequent decision and the faster I have them the better it is.
So I logged in, submitted the code to start the run, and logged out.
No heroics. No grind culture. Just a small intervention that meant I came back to results instead of nothing.
I can get off early the next working day to get back some time.
no big deal!
This happens all the time at work.
The gap between “this will take three extra days” and “this is already running” is often just one person deciding to act instead of waiting.
Trust your gut
Another work-related pattern I’ve seen repeatedly:
Some of the highest-leverage wins come from things that nobody is doing, not because they’re useless, but because they don’t fit cleanly into existing OKRs.
If something feels valuable to you, but:
it’s not explicitly on the roadmap
nobody has time to own it
it’s slightly outside your scope
that’s often a signal, not a warning.
