← All writing

AI

Handoverism: The AI Belief Bill Gurley Wants Disclosed

Mani G 3 min read

Orange steering wheel on a dark blue background with an arrow from human to AI and the word Handoverism

Some words arrive slowly. Handoverism arrived in about a week. On 11 November, Ruxandra Teslo, three weeks into life in San Francisco, published Handing over humanity. By 16 November, Benchmark's Bill Gurley was telling everyone on X that it is "critically important that everyone thinking about AI become familiar with the term." That is fast, even by AI-discourse standards.

What handoverism means

Handoverism is the belief that humanity should, or inevitably will, hand control of the world to AI. Not just the boring jobs. The steering wheel.

In Teslo's telling, it is not a fringe doom idea. It is often pitched as the happy ending. Humans survive, a well-aligned AI runs things, and we become, in her words, "the comfortably maintained inhabitants of a civilisation whose direction is no longer ours to decide." She guesses that something like 30 to 40 percent of the AI safety community outside the labs holds some version of this view, and she flags that number herself as uncertain. Treat it as a hunch from someone who has been listening at parties, not a survey.

She also points to a 2025 Lightcone interview where Anthropic co-founder Tom Brown said: "There's going to be a handoff where humanity will hand off control to transformative AI at some point." And she quotes a Scott Alexander essay that sketches several versions of handover. The one she reads him as favouring is titled "God", and the idea is to prompt an aligned AI with "please arrange the world as you see fit".

Why it suddenly matters

Timing. Anthropic filed to go public on 26 October, according to Dealroom. On Yahoo Finance's Deirdre Bosa Live on 16 November, Gurley said that if people around the company believe in handoverism, "that should be in the risk factors. And it is not. And I think it's a massive omission." On X he went further: he called himself "decidedly anti-handoverism" and said anyone writing AI rules should have to say where they stand.

I like the move, even if I am not sure it works. Risk factors in an S-1 are where lawyers list everything from currency swings to the CEO getting hit by a bus. Asking a company to write "our leaders may hope to give the planet to software" is a sharp way of asking whether investors are buying a business or a belief system.

To be fair, Gurley is describing what he thinks people near Anthropic believe. That is his read, not a company statement. On the same show, per Dealroom, Bosa noted Anthropic's view that the most powerful AI should be built by a small number of careful, responsible labs. That is a different claim from "hand it the keys".

A small example you already live with

You do not need a godlike AI to see the shape of this. Think of a team that lets a coding agent auto-merge its own pull requests because the tests pass. Week one, someone reads every diff. Week ten, nobody does. Nobody voted to stop reviewing. The review just quietly stopped happening. That is handover in miniature: no single moment of surrender, just a drift.

Teslo quotes the AI 2040 proposal saying almost exactly this: "There is no single moment where humanity relinquishes control. But in theory, there is a point of no return."

Where I land

My problem with handoverism is not that it is scary. Plenty of true things are scary. My problem is that it skips the hard part. We cannot yet get an agent to stop deleting the wrong folder, and some people are already planning how it should run society. Alignment would have to be a one-time, done-forever fix, and nothing in software has ever been done forever.

So here is my question for you: where is the line in your own work? Which decision would you never hand to an AI, even if it was right 99 times out of 100?