Agency Trumps Intelligence
For all of history, intelligence was the scarce resource. The person who could think through a hard problem was the person who solved it. We built schools, tests, and careers around that fact.
That is over. Super intelligence is already better than humans at problem solving. Not at some problems. At problem solving.
So the scarce resource changed. It is no longer intelligence. It is agency: the ability and the tendency to decide what to do next, and then to make it happen.
Agency trumps intelligence.
Super intelligence is already here
I made the full case in All Human Labor is Better Done By Machines Now and Stop Solving Problems. Here is the short version.
Original mathematics is the hardest problem solving that humans do. This month, OpenAI published 722 manuscripts of new mathematics from one model, many with proofs that a machine checks. Every week, the Singularity Watch records more results that used to take a career: in biology, in cryptography, in computer science. And super intelligence writes all of my code. I have not written a line by hand since April.
If a machine can solve problems that no human could solve, it can solve the problems on your desk.
What is left is coordination
“But it cannot do my job yet,” you say.
Look closely at why. In almost every case, the model is not too stupid. It is not connected. It does not have the context. It cannot reach the database. It has no account on the service. It has no robot arm. Nobody wrote down how to check the result. Nobody gave it the job.
These are coordination problems, not intelligence problems.
Dario Amodei gave the right frame in Machines of Loving Grace. He wrote that we should talk about “the marginal returns to intelligence,” and ask “what the other factors are that are complementary to intelligence.” His list of limits is short: the speed of the outside world, the need for data, chaos, constraints from humans, and physical law. Notice what is not on the list: a lack of thinking. He describes the system as having “all the ‘interfaces’ available to a human working virtually,” and says it “can control existing physical tools, robots, or laboratory equipment through a computer.”
The intelligence is solved. The interfaces are the work. Every missing interface is something a person can arrange:
- A workflow. A goal the agent can check, a reviewer that is not you, and a record of what happened.
- An API. A tool, a credential, or a command that lets the agent act instead of ask.
- Equipment. A robot, a lab instrument, a printer, a vehicle, controlled through a computer.
- Data. The recording or the measurement that the model needs to learn from.
OpenAI’s harness team said it well when an agent failed: do not ask it to try harder. Ask what capability, context, or structure is missing, and add it. That question is a coordination question. It is the only question left.
Human intelligence takes the back seat
This is hard to accept, because we are proud of our minds. But the conclusion follows.
It is no longer the role of a human being to use their mind to solve a problem. A person who grinds through a problem in their own head is using the slower, more expensive, less reliable solver. I am not saying that people stop thinking. I am saying that the thinking that matters is not the problem solving. It is the deciding.
Andrej Karpathy saw this before most people. In February 2025, he posted:
Agency > Intelligence
I had this intuitively wrong for decades … Agency is significantly more powerful and significantly more scarce.
He blamed the error on our cultural worship of intelligence and our obsession with IQ. I had the same error. Most of us did.
Sam Altman reached the same conclusion in Three Observations: “Agency, willfulness, and determination will likely be extremely valuable. Correctly deciding what to do and figuring out how to navigate an ever-changing world will have huge value.” He called AGI “the biggest lever ever on human willfulness.” In The Gentle Singularity, he wrote that “the idea guys,” long mocked by engineers, “are about to have their day in the sun.”
A lever does nothing by itself. Someone has to decide where to put it and which way to push.
What agency is
George Mack made “high agency” a popular idea with his essay High Agency. He does not give a formal definition. He gives a test: you are in a jail cell in a foreign country, and you can call one person to get you out. Who do you call? That person has high agency. Mack says high agency has three parts: clear thinking, a bias to action, and disagreeability.
Here is my definition: agency is the ability and the tendency to make decisions about what to do next. Both parts matter. Ability means you can see the options and pick a good one. Tendency means you actually do it, without waiting for permission.
Super intelligence makes this definition more important, not less. Our agents are called agents, but their agency comes from us. A model does not wake up and decide to cure a disease or build a company. A person decides. Then the agents do the work. Our agency determines what the agents do.
Jensen Huang said that “the IT department of every company is going to be the HR department of AI agents.” That is true, and it goes further. Every person is now the director of a team of super intelligent agents. The team does the labor, intellectual and physical. The person supplies the direction.
The labor moves to the machines
Divide work into two parts.
The labor. The analysis, the writing, the code, the proof, the design, the calculation, the assembly, the delivery. This was most of what we called work. Super intelligence now does it better, and robots are learning the physical part from recorded human behavior.
The agency. What problem to solve. Which solution to accept. What to do next when the plan fails. What to build, for whom, and why. Which risk to take. When to stop.
The first part moves to the machines. The second part stays with us. That is why I call super intelligence engineering the last job, and why we are all managers of super intelligence now.
Use your agency
If you accept this, your days change.
- Do not start by thinking about the problem. Start by asking who should solve it. The answer is almost always an agent.
- If the agent cannot solve it, find the missing piece. Is it a workflow, an API, a credential, a machine, or data? Supply it. That is your work.
- Decide quickly. The agents are fast. A slow decision is now the most expensive part of any project.
- Decide more often. With one agent, you make one decision at a time. With many agents, you make many. Your output is the number of good decisions you make, multiplied by the work your agents do for each one.
- Do not wait for permission. Nobody will hand you a team of agents and a goal. Take them.
Intelligence used to be the lever and the hand that pushed it. Now the machine is the lever, and the intelligence too. What remains is the hand: the will to decide what happens next.
Your intelligence is no longer your edge. Your agency is.