Institute

The centre of gravity of cognitive production is moving from human labour to autonomous agents. What follows is not that people become obsolete. It is that human authority — the capacity to legitimately mandate, verify, hold accountable and revoke — becomes the scarce, priced and credentialled resource.

That authority requires a discipline that does not yet exist as one: an engineering-economics of mandate, verification and residual risk over synthetic systems. Its objects are the mandates issued to autonomous systems; the verification of machine-produced work by verifiers less capable than the producer; the allocation of residual risk across composite agentic workflows; the identity and revocable authority of a non-human actor; and the population-scale effects that The Synthetic Economy research programme studies.

Why the adjacent disciplines each decline part of it

Control theory owns feedback but not authority. Systems engineering owns lifecycle but assumes specifiable requirements. Mechanism design and agency theory own delegation but assume agents that can be incentivised, sued or fired. Corporate governance owns accountability but only over people and capital. Regulatory science owns verification but is institutionalised for drugs and pollution, not cognition. Each answers one question and declines the rest.

What the Institute does not claim

It does not claim that nobody teaches AI agents. The Technical University of Munich has taught agents as economic actors under its AGENTONOMICS framework since 2026; HKU, Stanford, MIT, Singapore institutions and others occupy major individual pillars; new governance master's degrees are launching at Oxford, Imperial and Birmingham. The Institute's claim is architectural: the pieces exist in separate buildings, separate disciplines and incompatible levels of formality, and no institution is organised to form the professional who holds them together.

Where people go in the hierarchy

The thesis is not that humans disappear from the loop. It is that human participation can move upward in the control hierarchy: from continuous task execution toward objective-setting, authority, exception handling, verification and ultimate revocation. One published framework cited in our research reported human intervention declining from 21.0% to 11.5% of steps across five epochs while task success remained high. That is one published result under one framework. The Institute does not present it as proof of a universal intervention rate.

The pieces already exist. The profession does not.

The Institute intends to be judged on evidence, and to say only what is true about itself.

Literal truth on the public record

Every public sentence about the Institute's own state must be literally true. Proposed structures are labelled proposed; work in development is labelled in development. This site was audited against that rule before publication and will be again with each revision.

Evidence over information

The Institute's intended pedagogy is build, operate, govern — not know, explain, recall. The eventual credential is meant to attest demonstrated authority and performance under constraint, with a verifiable evidence dossier, rather than attendance.

Adversarial revision

Founding claims were tested by commissioned research that deliberately argued against them. The strongest version of the original thesis did not survive and is not used. The narrower version that did survive is the one on this site. Later work will follow the same pattern: claim, challenge, evidence, decision, recorded.

Source discipline

Where the site cites a fact about the outside world — a job posting, a regulation date, a published intervention rate — it comes from a primary source fetched during the September 2026 review. Job postings are perishable; regulatory dates can move. Where a claim is an inference rather than a sourced fact, it is worded as one.