I do not like the word automation very much.
It describes what the machine does.
It says almost nothing about what happens to the person.
The industrial framing
Automation comes from a world of repeatable physical processes.
A machine does the weld. A conveyor moves the object. A control system maintains the temperature.
The human labor removed is visible.
So when we talk about AI, we inherited the same question: which human tasks disappear?
That is an important question. It is not the only one.
The other side is agency
Agency is the ability to cause an outcome.
A person has agency when they can decide to build something and actually get enough of the world to move for it to exist.
Historically, the gap between intent and outcome has been enormous.
You can want to make a film and not have a production budget.
You can want to start a company and not know how to code.
You can understand a local market perfectly and not be able to afford a marketing department.
You can know exactly what research question matters and not have the analytical resources to answer it.
AI reduces pieces of that gap.
This is why productivity statistics feel incomplete
Suppose AI lets an analyst finish a report in two hours instead of four.
That is productivity.
Suppose it lets a person who could not previously perform the analysis produce a useful first version and understand enough to ask better questions.
That is capability expansion.
The second effect is harder to measure because the counterfactual is not "this took longer."
The counterfactual is "this did not happen."
We can already see users describing that effect
Microsoft's 2026 Work Trend Index reports that 58 percent of surveyed AI users said they were producing work they could not have produced a year earlier, rising to 80 percent among its most advanced group of users.
MICROSOFT WORK TREND INDEX · 2026 · FROM EXPERTISE TO AGENCY
I think that is one of the more important numbers in AI.
Not because self-reported survey data is perfect.
Because it asks a better question.
Not "did you save time?"
"Could you do something you could not do before?"
There is a design implication here
If the goal is agency, good AI products should not make people passive.
They should make intent easier to express, consequences easier to understand and execution easier to direct.
The person should become more capable of owning the outcome, not less capable of understanding it.
That means the best system is not always the one that hides everything.
Sometimes the user needs to see the reasoning, approve the action, change the standard or understand why the machine is uncertain.
Agency requires control.
The word changes what you build
If you optimize for automation, you ask:
How much work can the machine remove?
If you optimize for agency, you ask:
What can this person now accomplish?
I prefer the second question.
It leaves room for automation where automation helps. But it keeps the purpose attached to the human being on the other side of the software.
The future I want is not one where people become unnecessary to their own work.
It is one where more people can make meaningful things happen.
