The New Leverage

If Anthropic is really on track to hit $100B ARR this year, we should pause on that for a second.

Not because “AI is growing fast.” We know that.

Because that would mean Anthropic grew roughly 100x from January 2025.

At scale.

I am not sure any company in history has grown like that before. Not this quickly. Not with this much revenue. Not while selling into the most demanding customers in the world.

The obvious explanation is demand for AI.

The more interesting explanation is leverage.

The four old leverages

Every major economic era has been shaped by a new form of leverage.

First, humans learned to coordinate labour. Villages, armies, firms, institutions. One person could only do so much. Many people working together could build cities, run factories, ship goods, and coordinate at a scale no individual could touch.

Then machines gave us physical force. Steam engines, looms, railways, electricity, assembly lines. The industrial revolution was not just about better tools. It was about taking human muscle out of the bottleneck.

Then capital became leverage. Accumulated resources could be deployed at scale. Factories, ships, distribution networks, media, real estate, infrastructure. Capital let people compress time by buying capacity they did not personally have.

Then software changed the game again. Software gave us infinite replication. Write the code once and distribute it to millions of people at almost zero marginal cost. That is why software companies could scale in ways old industrial companies could not.

Humans. Machines. Capital. Software.

Each one changed what a small group of people could do.

Now we are getting the next layer.

AI agents are the new leverage

We have had AI for a long time.

Search ranking is AI. Fraud detection is AI. Recommendation engines are AI. Spam filters are AI. Curacel has used machine learning in claims workflows for years. None of this started with ChatGPT.

But most of that AI was narrow, passive, or embedded inside existing software. It classified. Predicted. Ranked. Recommended. Summarised. It helped the system make a better decision, but it did not go away and do work in the world.

Agentic AI is different.

An agent can take a goal, use tools, make attempts, evaluate the result, retry, and keep working. It can write code, test it, read logs, draft emails, update a CRM, search documents, analyse a market, monitor a workflow, or coordinate with another agent.

That makes it feel less like software you operate and more like digital labour you direct.

That is the shift.

Not AI as a feature.

AI as a workforce layer.

Anthropic is sitting at the intersection

This is why Anthropic’s growth is so interesting.

They are not just selling a better model. They are building infrastructure for the new leverage.

And they are also using that leverage themselves.

That second part matters.

The fastest growing AI companies are not just selling AI. They are early examples of what AI-native companies look like when the operating model changes.

Fewer people per dollar of revenue.

More output per team.

Shorter cycles from idea to product.

More experiments running in parallel.

Support, sales, engineering, research, operations, and marketing all compressed by agents doing work that used to require headcount.

This is the part people underweight.

The AI boom is not only about replacing tasks. It is about changing the shape of the firm.

In the machine age, the winning companies learned how to organize around factories.

In the software age, the winning companies learned how to organize around code.

In the agent age, the winning companies will learn how to organize around fleets of AI workers.

The correction: agents are not humans

There is an important trap here.

The lazy version of this argument is “AI agents replace humans.”

I do not think that is the right frame.

A useful objection is that AI can optimize, produce, classify, draft, and iterate. But it cannot decide the deepest objective function. It does not know what matters. It has no body, no childhood, no losses, no memories, no relationships, no skin in the game.

That sounds philosophical. It is actually practical.

Most valuable work has two human ends.

At the beginning, someone has to decide what is worth doing.

At the end, someone has to judge whether the output is good.

The machine can handle a huge amount of the middle. That middle is enormous. It is where a lot of cost, delay, coordination, and busywork lives.

But the machine does not care. It does not know why this customer matters, why this tradeoff is acceptable, why this product should exist, why this risk is worth taking, or why this work would be embarrassing even if the metrics say it is fine.

Humans still choose the objective.

Humans still provide judgment.

Humans still decide what matters.

So the point is not that agents replace humans.

The point is that agents multiply human agency.

More work, not less

There is another trap in the replacement argument.

When a tool solves problems faster, the problem space does not shrink. It expands.

The internet did not reduce the amount of work in the world. It created cybersecurity, search, social media, cloud infrastructure, creator businesses, remote work, attention markets, and a thousand other categories.

Software did not make companies simpler. It made it possible to build more complex companies.

AI agents will do the same.

The first-order effect is productivity. People will get more done with fewer resources.

The second-order effect is new ambition. Things that were too expensive, too slow, or too annoying suddenly become possible. Once they become possible, people start caring about them. Then those new possibilities create new problems, new companies, new jobs, new expectations, and new forms of competition.

Anyone using agents seriously already feels this.

The backlog does not disappear. It mutates.

You automate one workflow and immediately see five more that should exist. You ship one thing faster and your standards move. You give one person a fleet of agents and they do not work less. They start attempting things they would never have attempted before.

That is what leverage does.

The new operating model

The companies that understand this early will not just add a chatbot to the product and call it transformation.

They will redesign how work gets done.

They will ask:

What should humans still decide?

What can agents attempt independently?

Where do we need human judgment before something goes live?

Which workflows should run continuously instead of waiting for a meeting?

Which teams become smaller because agents handle the middle?

Which teams become more ambitious because agents remove the old bottleneck?

This is a different operating model.

A human no longer has to be the person doing every step. The human becomes the person setting direction, defining taste, judging quality, handling exceptions, and deciding what matters.

That is a higher-leverage role.

It is also a more demanding one.

Bad objectives will scale faster. Weak judgment will scale faster. Sloppy taste will scale faster. If your company does not know what good looks like, agents will not save you. They will just produce more mediocre work at higher speed.

Leverage always cuts both ways.

The point

Anthropic’s growth is not just a story about one company.

It is a signal that a new leverage layer is being installed into the economy at absurd speed.

Humans gave us coordination.

Machines gave us force.

Capital gave us scale.

Software gave us replication.

AI agents give us goal-directed digital labour.

That is the new leverage.

The winners will not be the people who ask whether AI will replace humans.

The winners will be the people who learn how to direct fleets of agents while keeping humans responsible for taste, judgment, and what matters.