The Age of Agency

Words carry history.

We use them casually, often without realizing that buried inside them are centuries of philosophy, politics and human experience.

Agency is one of those words.

It sounds distinctly modern. Silicon Valley modern, even. One could imagine it emerging from a whiteboard somewhere in San Francisco, next to the words agents, AGI and autonomy.

It did not.

The word traces back through agent to the Latin agens, from the verb agere.

To act.
To drive forward.
To set in motion.
To make something happen.

That distinction matters.

Because we have spent the last few years talking obsessively about Artificial Intelligence, when the deeper transformation may not ultimately be about intelligence at all.

It may be about action.

We thought machines were learning to think.

They are now learning to act.

And those are two very different revolutions.

Intelligence without action is only observation

For most of the digital age, software waited for us.

We opened the application.
We searched for the information.
We filled in the form.
We copied the data.
We sent the email.
We created the meeting.
We approved the transaction.

Even the first generation of generative AI followed essentially the same model.

You asked a question.
It answered.
You asked for an email.
It wrote one.
You asked for a strategy.
It suggested one.

Then you took the answer and did the work.

The machine could think.

The human still had to act.

That boundary is beginning to disappear.

The important shift is surprisingly easy to express.

Yesterday, we asked:

What should I do?

Tomorrow, we increasingly say:

Make this happen.

There is an entire economic transformation hidden between those two sentences.

The first asks for intelligence.

The second delegates an intention.

From tools to actors

Imagine telling an AI:

Organize my trip to Tokyo.

A chatbot might produce an itinerary.

An agent must understand your calendar, your budget, your travel preferences, your meetings, your airline accounts and your constraints.

It may need to compare flights, coordinate schedules, reserve hotels, draft messages, identify conflicts and come back to you only when a decision exceeds its authority.

Now take the same principle into business.

Instead of:

Show me which opportunities in the CRM have stalled.

We say:

Identify the opportunities that are at risk, understand why they have stalled, prepare the appropriate follow-up, execute what you are authorized to execute and escalate the rest.

Instead of:

Tell me which project milestones are late.

We say:

Detect the delays, identify the downstream consequences, gather the supporting information, propose corrective actions and assign the next steps to the appropriate owners.

The software is no longer simply displaying the world.

It is beginning to participate in it.

This is the transition from software as a tool to software as an actor.

And this is why I believe we are entering what I call: the Age of Agency.

Technology has always been the externalization of human capability

The history of technology can be read as a history of externalization.

We externalized physical strength into machines.

A crane can lift what hundreds of humans cannot.

We externalized speed into vehicles.
We externalized calculation into computers.
We externalized memory into writing, databases and eventually the internet.
We externalized navigation into GPS.

Generative AI began externalizing parts of cognition.

Writing.
Translation.
Programming.
Analysis.
Reasoning.

Agents add another layer.

They externalize parts of execution.

And execution is fundamentally different.

Because an idea does not change the world.

Its consequences do.

Knowing that a customer should be contacted creates no value.

Contacting them might.

Knowing that a supplier is late changes nothing.

Acting on the delay might.

Knowing that a project is drifting is information.

Correcting its trajectory is agency.

For decades, technology reduced the cost of information.

AI is reducing the cost of cognition.

Agents will increasingly reduce the cost of execution.

And when the cost of something collapses, value moves elsewhere.

When intelligence becomes abundant, judgment becomes scarce

The internet made information abundant.

As a result, merely possessing information became less valuable.

Generative AI is now doing something similar to intelligence.

Writing a reasonable document is becoming cheap.
Generating software is becoming cheaper.
Producing an analysis, translation, presentation, image or first strategic hypothesis is becoming cheaper.

So where does value go?

Toward what remains scarce.

Judgment.
Trust.
Reputation.
Relationships.
Access.
Proprietary context.
Responsibility.

And increasingly: the ability to determine what should happen next.

If execution becomes abundant, direction becomes scarce.

This may become one of the defining economic principles of the next decade.

The important question will gradually shift from:

What do you know?

to:

What can you cause to happen?

The next digital divide

The next digital divide will not simply separate people who use AI from those who do not.

Most people will eventually use AI.

The more interesting divide will be between people who use AI as a tool and people who build systems of agency around it.

Consider two entrepreneurs using exactly the same underlying model.

The first asks:

Write me a sales email.

The second connects intelligence to context, company knowledge, CRM data, communication systems, calendars and carefully defined permissions.

The system knows the commercial objective.
It watches the pipeline.
It identifies stalled opportunities.
It prepares the right next action.
It executes actions within its mandate.
It measures what happens.

And it asks the human for a decision when judgment, risk or responsibility requires one.

Same intelligence.

Radically different agency.

The competitive advantage does not reside only in the model.

It resides in the architecture surrounding the model.

The context.
The data.
The tools.
The permissions.
The workflows.
The relationships.
The institutional knowledge.
The rules.
The human judgment.

The model may increasingly become rented intelligence.

The operating system around it is where differentiation begins.

The dangerous misunderstanding: autonomy

There is, however, a trap.

Agency is often confused with autonomy.

They are not the same thing.

Autonomy asks:

How long can the machine operate without a human?

Agency asks:

How effectively can the system turn an intention into a useful outcome?

These are different objectives.

A highly autonomous system that produces the wrong result faster is not progress.

Nor should the ultimate goal of enterprise AI be removing humans from every decision.

The objective should be removing unnecessary friction while preserving responsibility where responsibility matters.

An AI can detect.
It can explain.
It can recommend.
It can draft.
It can coordinate.
It can execute clearly bounded actions.

But when a decision materially affects money, people, contracts, safety, reputation or strategy, someone still needs to own the consequence.

This is not a limitation of AI.

It is a principle of governance.

The strongest systems will therefore not be those with maximum autonomy. They will be those with optimal delegation.

The sovereignty paradox

Every technology that extends human freedom can also produce a new dependency.

The car liberated us from the limits of walking and made civilization dependent on transport infrastructure.

The smartphone liberated us from the office and made millions of people incapable of being separated from a screen.

AI may liberate us from thousands of repetitive cognitive tasks.

It could also make it extremely easy to delegate parts of ourselves.

Our memory.
Our organization.
Our reading.
Our choices.
Our communication.

Eventually, even our decisions.

The philosophical question therefore becomes unavoidable.

What happens when a machine knows enough about us to consistently make good decisions on our behalf?

It can optimize our calendar.
Our spending.
Our diet.
Our travel.
Our work.
Our relationships.

Perhaps one day it will know precisely what we normally choose before we choose it.

At that point, convenience and sovereignty begin to collide.

There will be people who use agents to extend their will.

And there will be people who allow agents to replace their will.

Both may appear more productive.

Only one remains fully sovereign.

The rarest resource may become intention

There is a final paradox.

As execution becomes easier, knowing what to execute becomes more important.

If you have one employee, finding the next task is easy.

If you suddenly have the equivalent execution capacity of one hundred agents, the bottleneck changes.

It becomes you.

Your ability to prioritize.
To define objectives.
To set boundaries.
To distinguish motion from progress.
To decide what is worth doing and what should never be done at all.

For most of history, humans could blame scarcity.

Not enough information.
Not enough people.
Not enough capital.
Not enough time.
Not enough expertise.

Artificial intelligence will not eliminate these constraints.

But it may reduce several of them dramatically.

And that leaves us confronting a much older problem.

What do we actually want?

When intelligence becomes abundant, intention matters more.

When execution becomes abundant, direction matters more.

When machines become capable of acting, human responsibility matters more.

That, to me, is the real meaning of the Age of Agency.

Not the age in which machines finally become autonomous.

The age in which humans acquire unprecedented leverage over the world around them, and therefore fewer excuses for failing to decide what they want to do with it.

The machine may soon be able to do almost anything we ask.

The decisive question will remain ours: what should we ask it to do?