Skip to main content
September 2026
← All 28 pairs

Pair 3×5 · structured conjecture

AI autonomy × Institutional adaptation

Work allocation shifts continuously as AI absorbs more of the job; accountability only ever moves in discrete, late jumps forced by scandal — and the gap between the two is where blame, not judgment, ends up living.

The full 2×2. Click to enlarge.

The four scenarios

Two questions: how much of the actual work has quietly moved to AI, and whether the institution around it can rewrite who's accountable fast enough to notice. Agency moves continuously; accountability only moves in discrete, late jumps, usually forced by a scandal rather than chosen ahead of one.

What do these axes mean? ▸

Each axis is a spectrum. The card takes the two poles of each and reads off what the corner where they meet produces downstream.

Axis 3 · AI autonomy

Low
Advisory AI drafts and recommends
High
Delegated AI acts and binds others

Whether AI advises a human who decides and executes, or acts directly on the world with little mediation.

Axis 5 · Institutional adaptation

Low
Inertial slow to metabolise change
High
Reconfigurable can rapidly restructure

Whether institutions can metabolise change quickly, or break under it. Fast adaptation is not automatically benign.

Advisory · Reconfigurable

The Rented Rulebook

Responsible advisory use turns out to cost more than the software: real accountability requires records, training, escalation paths, and audit staff, so this is the one corner where cheap AI is expensively surrounded. That expense recreates a new power center — governance leads, prompt stewards, model-risk teams — who don't do the work, they decide what counts as legitimate work, and smaller organizations either can't afford the wrapper or rent it from the same vendors selling the tools it's meant to govern.

Delegated · Reconfigurable

The Insurable Machine

This isn't a frontier-agent scenario — credit scoring, welfare clawbacks, and fraud detection have run delegated-and-processed for years, appeal rights and audit trails bolted on after the fact. Formalizing delegation buys legitimacy, and legitimacy is what gets sold. Certified agents, warranties, and liability products turn admitted delegation into an insurable asset class — and whoever defines the insurable form quietly decides what delegation is allowed to look like everywhere else.

Advisory · Inertial

The Backlog Behind the Signature

Comprehension stopped scaling with throughput long before anyone admitted delegation had happened: the human still signs everything, but AI raised the volume faster than review capacity could follow, so the signature increasingly certifies a document nobody actually read. The real cost lands a cohort later — reviewers trained entirely on already-drafted work never build the judgment the signature is supposed to guarantee, so the fallback competence disappears exactly when it would be needed to catch a bad call.

Delegated · Inertial

The Unadmitted Delegation

The tell isn't a policy that says delegated — no one admits that — it's a worker who clicks accept, forwards, or fails to override fast enough for the override to matter, absorbing blame for a system they didn't design or staff. Frontline workers usually clock the fiction before executives or regulators do, because they're the ones being blamed for it; but the arrangement persists because it suits everyone with the power to name it, so it tends to end only when litigation, scandal, or an insurer's refusal forces the question into the open.

Related pairs

Other cards that share one of these variables.

More pairs with AI autonomy

More pairs with Institutional adaptation