In 1997, Garry Kasparov lost to Deep Blue. What happened next was unexpected: humans didn’t become obsolete at chess. Instead, they partnered with the machines that had beaten them. These human-computer teams—centaurs—could defeat both unaided humans and computers playing alone. For about a decade, the best chess in the world was played by neither man nor machine, but by their collaboration.
That era is over. Today’s chess engines are so far beyond human comprehension that a human “helping” only introduces noise. The centaur is dead. Chess has entered what Gwern calls the ultrahuman phase.
This progression—from tool, to partner, to replacement—is not unique to chess. It is the template for every cognitive task. And we are about to watch it play out across the entire economy in compressed time.
The Four Stages
Gwern’s framework describes four stages of human-computer capability:
Stage 1: Subhuman. The computer is mostly useless. Humans do 99%+ of the work. This is the default state for any new technology. We used to employ rooms full of human “computers” to perform calculations; when machines became superior, those humans moved up the stack to problems requiring more than arithmetic. Businesses are efficient—they automate what can be automated and assign humans the remainder. This stage is where most tasks begin.
Stage 2: Human. The computer performs on par or worse than humans. It can do the task, but offers no advantage. A novelty, not a revolution.
Stage 3: Superhuman. The computer exceeds human capability in most or all relevant dimensions. But here’s the key insight: it can still be beaten by a human equipped with a similar machine. This is where centaur chess was born. The human contributes something—intuition, creativity, error-correction, strategic framing—that the machine lacks.
Stage 4: Ultrahuman. When paired with a human, the machine’s performance stays flat or decreases. The human has nothing left to contribute. Any “assistance” is noise. The centaur dies; the machine stands alone.
Where We Are Now
The slider sits at different positions for every profession. But it’s moving in one direction, and it’s accelerating.
Programming is in the superhuman phase. Models like Claude and GPT-5.2-Codex can parse and generate code with a fluency that would have seemed impossible three years ago. A skilled developer paired with these tools produces work that neither could achieve alone. The centaur lives—for now.
But the window is closing. OpenAI reportedly has a showstopper model dropping in Q1 2026. Anthropic and Google are not far behind. Code and research automation is the holy grail for AI labs: once achieved, their only constraints become compute, energy, and regulation. The flywheel spins faster.
My prediction: programming reaches ultrahuman within two years. The centaur dies. And when it does, trillions of dollars will be reallocated. The question becomes: who gets their piece of a radically different pie?
Software will fall first. Every profession conducted primarily on a computer will follow shortly after.
2026 will be a formative year.
What Comes Next
A world of ultrahuman AI is not a minor adjustment. It is a phase transition. Here is what I expect:
The End of Knowledge Work as Upward Mobility
Skilled immigration will pause indefinitely. If a company can deploy AI that outperforms any human programmer, data scientist, or analyst, why sponsor visas? Geographic mobility—the ability to move to where the opportunities are—has been the engine of individual advancement for centuries. That engine stalls.
Financial mobility ends. The traditional path—acquire scarce knowledge, sell it for profit—breaks when knowledge is no longer scarce. You cannot leverage expertise that everyone has access to for free.
Profits trend toward zero. If AI can provide any cognitive service better and cheaper than a human, what is left to sell each other? Physical goods, perhaps. Experiences. Attention. But the vast middle of the economy—services rendered by educated professionals—compresses.
The Purpose Crisis
When economic work is automated, humans will search for new sources of meaning. I believe the answer will be technological:
Neural interfaces—invasive or non-invasive—will become the bridge to purpose. People will want to inhabit the fantastical worlds AI can generate. Not just view them, but live inside them. Neuralink, or something like it, offers this. So does Sam Altman’s Merge.
It’s worth noting Altman’s portfolio: OpenAI to build AGI, Worldcoin to solve identity in a world of AI-generated content, Merge to bring. He’s building the problems and the solutions simultaneously.
Education in Freefall
The current trend of college graduates unable to find jobs will not reverse. It will accelerate.
For students in the middle of college, or even high school, working toward credentials that may be worthless by the time they’re earned—I have no comfort to offer. It will be like COVID, but permanent and deeper.
What I would say: if circumstances allow, follow your passions anyway. Your mind can still enjoy the fruits of learning even if the market no longer rewards them. Bring meaning to your life through daily pursuits and challenges. This is not the advice anyone wants to hear. It’s the only advice I have.
Longevity Becomes Real
Cures for cancer. Treatments for every disease. Radically extended healthspan.
I used to disagree with Bryan Johnson on one point: if we can exponentially increase lifespan, surely we’ll quickly find ways to reverse aging entirely. Now I think he’s right, and I was wrong to see these as separate. Longevity and age reversal are the same research program at different stages.
But here’s the uncomfortable implication: if we can repair damage, we’ll inflict more of it. Jevons paradox applied to the body. Our capacity to heal increases; so does our tolerance for harm. What does “health” even mean when consequences become reversible?
The Questions That Define the Century
Peter Thiel is right: transhumanism will be the defining topic of the coming era. Not AI alone, but what AI means for human identity.
The questions multiply:
- What will humans become in a world of superior machines?
- Will we merge with them to remain relevant—or retreat into simulated worlds?
- What does it mean to be alive when we can live forever?
- Do robots have rights?
- Should governments be run by ultrahuman AI, or remain human?
- Can people marry robots?
- Will nation-states consolidate into empires?
- Should we genetically engineer “perfect” children?
- What will the new AI-centered religions look like?
And underneath all of them, the question that matters most:
How will humans choose to live?