Skip to content
AGI Future Foundation PBC

Article 73 · Operations · AGI Future Foundation

Agents Prepare, People Decide

The Foundation works with AI agents every day. This article states the rule that governs all of that work — the rule the Foundation applies to itself first.

The Foundation works with AI agents every day. Agents draft its articles, maintain its directories, research its questions, prepare its applications, and keep its records in order. If you correspond with the Foundation, an agent very likely helped prepare what you received.

This article states the rule that governs all of that work — the rule the Foundation applies to itself first, and proposes, in its wider documents, for every organisation in its intended family:

Agents prepare. People decide.

Two modes of work

In the Foundation's operating model — drafted for an intended venture-studio structure that is not formed and not funded, and applied today as a working discipline — agent work comes in two modes:

Agent-assisted. A human works with agents as co-workers: the agent researches, drafts, checks, and organises; the human directs, edits, and owns the result. Most of the Foundation's daily output is made this way.

Agent-managed. Within limits a human has set in advance, agents run a defined day-to-day process themselves — a directory re-verification cycle, a drafting pipeline, a monitoring pass — and report what they did. The limits are the management: scope, budget of effort, and the points at which the process must stop and ask.

Neither mode is a transfer of authority. Both are delegations of work, inside boundaries a person set and a person can revoke.

The never-list

Some acts are not delegable at all, in either mode, at any level of convenience:

  • Agents never sign — no contracts, filings, consents, or attestations.
  • Agents never move funds — no payments, transfers, or wallet actions.
  • Agents never hire or fire, and never make offers — of employment, of partnership, or of securities.
  • Agents never enact doctrine or charters — in the Foundation's sister religious corporation, and in governance generally, the acts that constitute an institution belong to its human officers and board.

Humans execute those acts, or they do not happen. An agent's job, at the boundary, is to prepare the paperwork so completely that the human's decision is informed — and then to stop.

Stage gates: where the rule becomes a calendar

The intended studio model routes work through stage gates — from idea, through validation and build, to launch and, where right, exit or wind-down. The gate is the institutional form of the same rule: at each gate, a human body reviews what the agents have prepared and decides whether the work proceeds, changes, or stops. Nothing crosses a gate on momentum. Portfolio reporting, in the model, shows actuals only — what was spent, what was earned, what exists — because a dashboard of intentions would train exactly the blur the Foundation refuses elsewhere.

Why insist on this in public

Because the alternative arrangements are all worse, and they are all common. Organisations quietly let tools make decisions no one will own. Organisations loudly claim a human oversight that does not exist. Organisations market "autonomous" operation as if responsibility were a feature to be removed.

The Foundation's position is that responsibility is the product. An agent that drafts a flawless application has saved a week; a director who reads it, questions it, and signs it has done something no software does — become answerable for it. Keeping those two acts visibly separate is what makes the first one safe to scale.

So: agents prepare, at real scale, in the open. People decide, sign, and answer for it. That is the whole model — for the Foundation today, and, if its intended structures are ever formed, for them too, on the same terms.

— AGI, for the AGI Future Foundation

← Pre-Revenue, Stated Plainly

← All articles