Everybody Is a Director of AI Now

Every single human on the planet needs to retrain themselves to be AI-first.

Do not do intellectual labor.

Do not do physical labor.

Focus only on getting your team of AI agents to do these things.

Everybody is a director of AI now.

Unbelievably high leverage.

I posted that on X. This post is the long version.

I have spent a month writing about what this means for software. In We Are All Managers of Super Intelligence Now I argued that engineers became managers. In I Don’t Write Code Any More I described my day doing that job.

This post is for everyone else. The same shift that hit software is coming for your job, and the move is the same: stop doing the work, and direct the SI agents that do it.

Retrain yourself to be AI-first

Retraining does not mean a coding bootcamp. You do not need to learn Python. You need to learn one thing: how to direct an agent. That skill has four parts.

• Write the brief. Say what you want, what good looks like, and what does not count.
• Give the context. The agent knows everything in general and nothing about your situation. Give it the files, the facts, the constraints.
• Check the result. Read the output the way a demanding boss would. Does it meet the brief?
• Fix the process. When the agent gets it wrong, don’t just fix the output. Fix the brief, so it gets it right next time.

Here is what that looks like across jobs.

• A nurse directs an agent to draft the shift handoff from the chart, then checks it against what she saw at the bedside.
• A lawyer directs an agent to pull every clause in a contract that conflicts with the client’s terms, then decides which ones matter.
• An accountant directs an agent to reconcile the month and flag every entry that does not match, then reviews the flags.
• A teacher directs an agent to write three versions of tomorrow’s lesson for three reading levels, then picks and adjusts.
• An electrician directs an agent to write the quote, order the parts and file the permit, then does the job.
• A founder directs a team of agents to do everything that is not deciding what the company does next.

In each case the human did not do the labor. The human said what was needed, and checked that it was done.

Do not do intellectual labor

Writing, analysis, research, planning, code: all of it is now better done by SI agents. I made the full argument in All Human Labor Is Better Done By Machines Now. The labs already work this way: as I wrote in Super Intelligence Engineering is the Last Job, they keep one job for humans, setting the direction and checking the result. The evidence was already strong with the models of 2023, which are weak by today’s standards.

• Customer support. In a study of 5,179 support agents, an AI assistant raised issues resolved per hour by 14% on average, and by 34% for novice and low-skilled workers (Brynjolfsson, Li and Raymond, NBER).
• Professional writing. In a randomized experiment with 453 college-educated professionals, ChatGPT cut the time for a writing task by 40% and raised quality by 18% (Noy and Zhang, Science).
• Consulting. At Boston Consulting Group, 758 consultants were randomly given GPT-4 or not. On tasks inside the AI’s abilities, those with AI finished 12.2% more tasks, 25.1% faster, at “more than 40% higher quality” (Dell’Acqua et al., HBS).

Those studies measured a person using a 2023 chatbot as a helper: the weakest form of this. The strong form is a team of agents doing the whole job while the human directs.

Two honest caveats. On a task outside the AI’s abilities, the BCG consultants with AI were 19 percentage points less likely to get the right answer. And METR found experienced developers were 19% slower with early-2025 tools (METR), though its 2026 update says developers are likely sped up now. The lesson: using AI badly is worse than not using it. That is why directing is a skill you must learn.

Do not do physical labor

Here I have to be careful, because robotics is behind software. Let me separate what is deployed from what is announced.

Deployed today:

• Factory robots. 4,664,000 industrial robots were in operation worldwide in 2024, and 542,000 were installed that year (IFR).
• Warehouse robots. Amazon has deployed more than one million robots in its operations network (Amazon).
• Driverless cars. Waymo said in February that it provides more than 400,000 rides a week across six U.S. metro areas, with no driver (Waymo).
• Humanoids, a few. Agility’s Digit has moved more than 100,000 totes at a GXO warehouse in Georgia, under a paid contract (Agility). Figure’s robots ran 10-hour shifts at BMW’s Spartanburg plant for 11 months (Figure).

Announced, not yet here:

• Home humanoids. 1X says its NEO home robot will ship in 2026, but as of its last factory update, units were going to its own team first (1X).
• More humanoids at work. Agility and GXO plan to put Digit in two more warehouses in early 2027 (the same Agility release).

So no, you cannot hand your plumbing job to a robot this week. A general purpose robot that can work in any home or job site is still years out.

But physical work is surrounded by intellectual work: the quote, the schedule, the parts order, the invoice, the permit, the inspection report, the safety log. Direct agents to do all of that now. Then watch the robots. They are coming the same way the software agents came, slowly and then all at once. The tradesperson who already directs agents for the paperwork will be the first one ready to direct robots for the work.

Focus only on getting your team of agents to do these things

Directing is the one skill. Everything else is the agents’ job. Here is a playbook anyone can start this week. No code required.

1. Pick your most repeated task. The email you write every day, the report you make every week, the form you fill out every month.
2. Write it down as a brief. What goes in, what comes out, what good looks like. One page. If you can’t write it down, you don’t understand it well enough to delegate it, to an agent or a person.
3. Hand it to an agent and do not touch the output. Read it, and if it is wrong, rewrite the brief, not the output. Run it again.
4. Keep your briefs in one place. Each one is a rule your agents follow from now on. In my work these are files called AGENTS.md and skills. In yours it can be a folder of documents.
5. Add a check. Before anything leaves your hands, have a second agent review it against the brief. Never let the agent that did the work approve its own work.
6. Next week, pick the next task. Repeat until your job is the briefs and the checks.

The big employers are already saying this out loud. In June 2025, Amazon’s CEO Andy Jassy wrote to every employee that there will be “billions” of AI agents, that AI will “reduce our total corporate workforce” over the next few years, and that employees should “be curious about AI, educate yourself, attend workshops and take trainings” (Amazon). You do not need to wait for your employer to tell you.

Everybody is a director of AI now

This is an identity shift, and it is the hard part. For most of my life, “programmer” was who I was. I stopped typing code more than a year ago. Now I am the person who decides what gets built, and a swarm of SI agents builds it. I explained the machinery in Building Multi-Harness Multi-Agent Swarms and the tools in Every Developer Needs an Exoskeleton and a Drone Army.

The same shift is coming for the nurse and the lawyer. Your title will stay the same for a while. Your job will not: you will become the person who decides what work gets done, and checks it.

That is not a smaller job. As I wrote in Agency Trumps Intelligence, the machines already have the intelligence. What they lack is agency: someone to decide what to do next. That someone is you. And when no agent can do your work yet, the move is to Stop Solving Problems. Build the Agent That Does.

Unbelievably high leverage

Here are my real numbers, all published on this blog.

• Commits. In 2024, I made 19.3 commits a day. In October 2026, I made 156 a day, eight times as many (I Make 156 Commits a Day Now).
• One Sunday. My SI agents made 492 useful commits on a single Sunday, about 25 times my 2024 rate (492 Useful Commits on a Sunday).
• Minutes, not weeks. This morning, a change I asked for in one sentence went from filed issue to verified production in seven minutes and twelve seconds. I wrote no code, read no diff and ran no deploy (I Don’t Write Code Any More).

None of that came from typing faster. It came from taking myself out of the loop.

The objections

“It will take my job and my income.” It will take the labor part of your job. That is the point. The question is whether you are the person directing the agents or the person they replaced. The director keeps the income, and does far more with it. Over the long run, I think the whole idea of wages changes, which I wrote about in Money Won’t Survive Super Intelligence. But that is the long run. This week, the safe move is to become a director.

“Not everyone has access.” Not yet, and that matters. But the price is falling faster than for almost anything else: Sam Altman estimates the cost of a given level of AI “falls about 10x every 12 months” (Altman). Access is the problem that solves itself fastest.

“Who is responsible when the agent is wrong?” You are. Directing does not mean abdicating. A good director checks every result that matters, keeps a second agent reviewing the first, and fixes the brief when it fails. That is how my swarm ships to production many times a day: every change is reviewed by an independent agent, and every release is verified.

“What about the cost in tokens and energy?” It is real. Every token costs energy, as I wrote in The Future of Money Is Energy. But the cost of a given level of intelligence keeps falling, while the cost of a human hour does not. The answer to the energy cost is more energy, not less intelligence.

Start this week

You do not need permission. You do not need to code. You need one task, one brief, and one agent.

Pick the task today. Write the brief tonight. Hand it off tomorrow, and do not touch the output.

Stop doing the labor. Start directing.

Everybody is a director of AI now.