The Techno-Optimist Manifesto and the enemies list

What the document says

The manifesto is a long list of declarative sentences, arranged under headings from Lies to Patron Saints, published on the a16z site on October 16. Its claims about technology are familiar from Andreessen's earlier writing. Technology is a lever on the world. Every material problem humanity has solved was solved by technology. Markets discover prices and lift people from poverty. Energy is life, and fission and fusion are available answers. Population could grow past 50 billion and be fine.

On AI it is more specific and more strident. AI is described as alchemy, as making sand think, and as a universal problem solver. Then comes the sentence that most people quoted this week. Any deceleration of AI will cost lives, and deaths that were preventable by the AI that was prevented from existing are a form of murder. That is a moral claim with a very large denominator, and the manifesto does not attempt to estimate either side of it.

The enemies

The section titled The Enemy insists that the enemies are ideas rather than people, and then lists them. Stagnation. Anti-merit and anti-ambition. Statism, collectivism and central planning. Bureaucracy and vetocracy. Regulatory capture and cartels. Credentialism. Speech and thought control. The Precautionary Principle, named as such. Deceleration, degrowth and depopulation. Nietzsche's Last Man. Somewhere in that list the phrases trust and safety, tech ethics, risk management and sustainability appear alongside the others, and that is the part we want to talk about.

Most of the list is politics and we have no professional standing on it. But three of the items describe our job. Risk management is what you do when a system can fail in ways that hurt people. Trust and safety is the team that reads the abuse reports. Tech ethics, whatever the phrase has come to mean in a university brochure, started as the practice of asking who bears the cost when a system is wrong. Putting those on an enemies list with central planning is a category error, and we think it is a deliberate one.

What accelerationism versus everyone else leaves out

The manifesto's rhetorical move is to fold every objection to a technology into one enemy, deceleration, and then to place the reader on one side or the other. That framing has already shaped how the AI argument is conducted this year. You are either building or you are stopping the builders. There is no slot in it for the person who wants the thing built and also wants to know how it fails.

In practice that person is most of the field. Evaluation, red-teaming, interpretability and safety training exist because the models we ship do things their builders did not intend, and finding those things early is cheaper than finding them in production. None of that work slows deployment in any way that the manifesto's argument engages with. A model that has been tested against the failure modes we know about ships with fewer recalls. The precautionary principle, in its strong form, would stop the model from shipping at all. The engineering practice the manifesto lumps in with it is what lets the model ship.

There is also a factual gap that a working researcher notices immediately. The manifesto's argument for AI rests on AI working, on the model that reads the scan or drafts the treatment being right often enough to save lives. The reliability that makes that true is not a given. It is produced, expensively, by exactly the people who have just been told they are the enemy. The document has no account of where reliability comes from, and so it can treat the people who make it as a tax.

The part we agree with

We do not want to pretend the manifesto is wrong about everything. The critique of vetocracy is fair. Anyone who has tried to build a data centre, a reactor or a clinical trial knows that the number of parties who can say no has grown while the number who can say yes has not. The observation that innovation in a market system is philanthropic by a large ratio, with creators capturing a small share of the value, is a real finding in the economics literature that the document cites, though the 50 to 1 figure is stated rather than argued. And the impatience with degrowth as a serious answer to anything is one we share.

Where we part ways is the assumption that the people who study failure want the failures. The manifesto quotes Feynman on science as belief in the ignorance of experts. Feynman also spent his last public act showing, with a clamp and a glass of ice water, why an O-ring failed. That is what risk management looks like when it is done well, and it is on the wrong list.

What comes next

The manifesto will be read as a founding document by people who want a side, and it gives them one. The cost is that a set of practices with no politics has now been assigned some. We expect the next year of AI argument to be conducted in the manifesto's terms, accelerate or decelerate, and we expect most of the useful work to happen in the gap it left out.

If someone wants a project, we would like to see a counter-document that is equally short and equally declarative, and that says what a builder who also tests owes to the people who use what they build. Not a call for slowing down. A description of the work that makes the acceleration survivable.