Why The Last Human Oracle exists.

The thesis

Rationally, AI can already do almost everything. It is the median of human thought and human knowledge, with the added ability to retrieve and give back the whole of what has been written. On that ground the game is over, and there is no reason to regret it.

What remains, however, does not look the way we expected. It is not a list of talents the machines have yet to catch up with, since that list gets shorter every year, on dates that can be written down. What remains are positions. Someone has to decide what is worth wanting. Someone has to consent. Someone has to answer for what was decided when it goes wrong. A machine can produce the best answer. It cannot be the one who is called to account.

Why it matters

The ledger could be read as an inventory of losses. That would be reading it backwards. Each time a task is handed to a machine, something appears on the other side: someone has to decide what to entrust and what to keep, learn to frame the question so that it is worth asking, check what comes back, and sign. These are new responsibilities and new skills, ones nobody held ten years ago because there was nobody to delegate to.

The ledger is kept for that as well. As the frontier moves, it records what is freed and what appears: the positions that delegation creates, the judgments it demands, the gestures that become the heart of human work once the rest is automated. A ceded line often calls for a new one, on the held side.

This is why the site exists. It is meant to be a compass of who we are as a species at this precise moment: what we have placed in other hands, what we still hold, and what we learn about ourselves by holding the reins of tools that can do almost anything. A compass is read again at every step. It does not say where to go. It says where we stand.

Why a ledger rather than a book of opinions

Every advance arrives with a mood, triumph or dread, and the mood is what people remember. The line itself, where it stood and where it stands now, is almost never written. A book changes what you know. It does not say where the frontier is this month, and it does not correct itself.

A ledger does. Each capacity has a status, a date, a note on what exactly changed and the sources that prove it. A status can be wrong in a way that can be corrected: someone points at the line, it is revised, and the previous status stays in the history. This is the discipline of the documents people trust without knowing their author, a table of constants, a court docket, a register of ships. They are dull on purpose.

How it is kept

Three statuses. Held: no machine performs the act, because it belongs to a person by its nature. Contested: machines match or approach the best humans in a narrow or disputed setting, and the question is still open in good faith. Ceded: on the task as defined, the best machine result exceeds the best human result, with a date and a source. Once a line is ceded it stays so. The ledger records, it does not rank.

Every date is checked against a source before it is published. When a claim cannot be checked, the line is marked contested and the note says why. The essays do the rest: they ask what the statuses mean, and they take a position.

Who it is for

This ledger was first kept for its author, who needed to know where the frontier stood before deciding what to learn, what to delegate and what to keep for himself. That is worth saying, because it explains the choices: no advertising, no data sold, no product behind it, nothing for sale, no account to create. New essays and status changes are announced on X and LinkedIn, and that is all.

If it can be useful to others, it might as well be public. That is the only ambition behind putting it online.

Who writes

Quentin Devillechabrolle, formerly of Palantir, modelling what is at stake on the frontier between machines and people.

To correct a line, propose a source or argue a status, the address is in the footer, and he answers on X and LinkedIn.