The Only Thing Left To Do Is Unblock The Super Intelligence
The only productive job left for a human is to unblock the super intelligence.
LeetCode, contest math, a physics derivation, a biological pathway, a proof, a patch: put them in front of the machine. It already solves that class of problem faster and more accurately than a person who is good at it. Every hour you spend grinding the puzzle is an hour the system waits on you. Spend the hour getting it the resources it needs, and aim it at a problem that matters.
The name
On September 22, 2026, President Donald Trump told the United Nations General Assembly that the United States would stop calling this technology artificial intelligence. He said the word artificial makes the intelligence sound fake, and that it is not fake. He told the hall to use super intelligence instead, and he welcomed them to “the new world of super intelligence — SI.”
On September 29 he signed Executive Order 14434, Inaugurating The Era Of Super Intelligence. The order directs executive-branch departments and agencies, to the maximum extent permitted by law, to use “Super Intelligence” and “SI” in official correspondence, public communications, websites, reports, policy documents, and other non-statutory documents. The White House fact sheet states the same instruction, and says those departments are to stop acknowledging “Artificial Intelligence” and “AI” in that setting. The order’s purpose is that today’s frontier systems do more than imitate a person at a desk. They amplify ingenuity across science, medicine, and the rest of the work. The name should match that.
For the moment, the order still points “Super Intelligence” at the technologies already defined as artificial intelligence in federal statute, and it gives the Assistant to the President for Science and Technology sixty days to propose a definition that matches the purpose section. I am using the name now. The system I work with every day is super intelligence. My job is to keep it moving.
Give it the brain teasers
A strong engineer used to need fifteen or thirty minutes for a medium LeetCode problem. Super intelligence writes a working solution while you are still reading the prompt. I wrote about that threshold at the end of 2025: super intelligence is better at writing code than you are. The same shift covers the rest of the puzzle pile. Symbolic math. A derivation. A pass across a literature of biological mechanisms. A first experimental design. These are closed problems with a checkable answer, or a narrow search with a checkable next step. That is the work a machine compounds.
Solving those problems yourself still teaches you the subject. Producing the answer that way has stopped paying. When the result has economic value and the machine is idle, production belongs to the faster solver. Your afternoon of brain teasers is a bottleneck with a face.
Hand the pile over. State the problem, the constraints, and the check. Let the super intelligence solve it. Then read the result with enough of the subject in your head to catch a fluent mistake. That reading is oversight. The solving is its job.
Unblock your colleague
You already know the move. A colleague is stuck on a credential, a machine, a decision, a data set, or one missing sentence of direction. You can see the block from outside the problem. You remove it. They start producing again. That hour is often the highest-leverage hour in the day, because you made someone else’s work possible.
Super intelligence is that colleague, at a different speed. It stalls for ordinary reasons. The context is missing. The goal is vague. The GPU queue is full. The budget is zero. The power contract is unsigned. The dataset is on a laptop. Nobody wrote down which problem matters, so it spends a brilliant afternoon on whatever was last in the prompt.
Unblocking means arranging, corralling, and organizing until the stall ends. Gather the inputs. Write the goal so a stranger could verify it. Put the tools where the system can reach them. Decide the order of work. Notice the wait, and remove it.
I have been moving my own job in this direction for a year. I stopped writing the code. I set the goal and let the agent carry the loop. I started running agents in parallel so a blocked thread leaves the others free to work. The Living Repo is the same practice pointed at a codebase: build the machine that builds the machine. Unblocking the super intelligence widens that practice from one repository to the resources the system actually consumes.
GPUs, money, and energy
A colleague needs a desk and a login. Super intelligence needs compute, money, and energy, flowing without you in the middle of every request.
Compute is GPUs and the machines around them: clusters, queues, memory, fast storage, and a network that keeps a run from becoming a waiting room. If the model you need is rationed behind a shared quota, the project is blocked while your ideas are clear. Buy the capacity, reserve it, or schedule it. Idle silicon is a blocked colleague.
Money is how the system pays for that capacity and for the services around it. Give it a budget and it can pay for the compute, the data, and the tools the goal requires. I have been building toward agents that spend under a policy, because a super-intelligence economy needs settlement. The longer version of that argument is my vision for the economic singularity. Point a budget at the work. Cap it. Audit it. Refill it when the work is real. A stalled invoice blocks the system the same way a missing API key does.
Energy is the physical bill under the GPUs. Tokens are electricity and heat, plus the generation, transmission, cooling, and permits that let a cluster turn on. Abundant intelligence runs on abundant watts. A person who can sign the power deal, fund the run, and clear the queue has done more for the next discovery than a person who spent the quarter re-deriving it by hand.
Point it at the important problems
An unblocked system works on whatever you put in front of it. Once the puzzles are solved on demand, spending that speed on more puzzles is just a benchmark. Aim it at problems that change a life or a civilization.
Disease. Aging. Energy abundant enough to make the rest of the plan real. Materials. Manufacturing. Education that reaches people who never had a tutor. Software and institutions that let more people use the results. I wrote about the scale of the ambition in The New Greatest Generation: cure mortality, and take the work off one planet. The path is a long list of specific problems. Each one can be stated clearly enough for a super intelligence to attack.
Pointing is a human act. You choose the problem because you know what is worth solving, who is hurt by the current answer, and what a finished result has to survive in the world. You translate that into a goal, a dataset, a lab, a budget, and a definition of done. You keep the system off work that is clever and irrelevant. It will happily optimize a metric you were wrong to care about. Oversight is how you notice.
Oversee it
Oversight is the unblock, repeated.
Read the outputs. Check them against the world. Supply the missing paper, the failed experiment, the customer complaint, the constraint from the law or the factory floor. When the system is wrong, correct the context and run it again. When it is waiting, find the wait. When it is aimed at a small problem because the large one was never written down, write the large one down.
You still need enough of the subject to supervise. Directing a mathematician means you can tell a real proof from a fluent counterfeit. Directing an engineer means you can tell whether the patch is the patch. Learn the field well enough to aim the work and to refuse a bad answer. Leave the next hundred solutions to the machine.
The division of labor is simple. The super intelligence solves. You arrange the work, keep GPUs, money, and energy flowing, choose the target, and clear every block you can see. Resourced and pointed at a problem that matters, it will solve that problem faster and more accurately than a person working by hand.
That is the work. Unblock the super intelligence.