Adaptation speed and access breadth don't change how much power exists in a system — they relocate it to wherever the check on it is missing.
The full 2×2. Click to enlarge.
The four scenarios
Two questions: how many actors can use frontier capability, and whether institutions can bend to absorb what that capability does to them. It's tempting to read reconfigurable as good and inertial as bad — the fastest-adapting institution in a shock is often the one that captures the new capability before anyone can organize against it, and slowness sometimes just buys legitimacy the time it needs to form.
What do these axes mean? ▸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 1 · Access to capability
Low
Chokepointed — a few controlled actors
High
Proliferated — frontier capability widely available
Who can use or deploy high-end capability — a few controlled actors, or anyone with a laptop.
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.
Chokepointed · Reconfigurable
The benevolent version is the more durable one
Institutions adapting fast while capability stays scarce doesn't read as good governance so much as the fastest route to durable, legitimated, hard-to-reverse concentration — because no outside actor has comparable capability and comparable speed to contest the rewrite before it's already in force. The unsettling version isn't the coercive one: a fast-reconfiguring chokepoint-holder that delivers real, excellent service earns genuine consent, which locks in harder than force does, because people don't want to leave. Worse, each restructure erases the procedural memory of the version before it, so within a few cycles there's nothing left to restore to even if someone wanted to.
Signs it isn't
The corner's own redemption is conditional and actively undermined by its incentives: a self-binding, genuinely accountable fast-reconfiguring holder could mount the coordinated emergency response a proliferated or inertial world can't manage — but the same speed and monopoly that would let it self-restrain are exactly what's being used to outrun contestation.
Proliferated · Reconfigurable
No steady state, only turbulence
The corner that looks like the dream — everyone empowered, institutions keeping pace — has no steady state at all, and it shows up as continuous regulatory arbitrage: firms, agencies, courts, and insurers all update their rules and models in near-real time, so compliance itself becomes a moving battlefield and a signed contract loses its settlement value before the ink dries. The scarce resource becomes stable expectations, not power. It isn't a destination so much as a launchpad too costly to sustain: it decays toward the fast-adapting chokepoint if one actor freezes the music, or toward the frozen-institution corner if everyone exhausts their capacity to keep adapting. What it protects, if it can afford the coordination tax, is real — because both concentration and paralysis are structurally resisted here, local failure can stay local and exit stays open.
Chokepointed · Inertial
Rental relationships become governance relationships
The danger isn't that a few actors hold the AI, it's that institutions too slow to build their own capability are forced to rent it — and rental relationships quietly become governance relationships. Within a few budget cycles the vendor's roadmap becomes the agency's roadmap, model deprecation becomes involuntary institutional amnesia, and nobody voted for the procurement contract to function as a constitution. The genuinely irreversible part isn't the dependency itself — it's that staff who could once do the work without AI deskill or retire, leaving the institution unable to perform the task either with its own people or without the vendor.
Signs it isn't
The same slowness and scarcity that produces vendor capture can instead function as a legitimacy incubator: few actors and little fast motion leaves time to negotiate durable rules and preserve a human fallback that a proliferated free-for-all wouldn't allow. Whether this corner reads as quiet capture or patient settlement turns on whether that fallback capacity is deliberately kept alive or left to atrophy.
Proliferated · Inertial
The bypassed institution keeps standing
Capability everywhere and institutions frozen doesn't produce revolution, it produces something quieter and stranger: people stop fighting the slow regulator and just use cheap frontier tools to do privately what it used to mediate — adjudication, verification, certification — until the institution is still funded, still staffed, and increasingly decorative. Because individuals can patch around institutional failure, the institution stops receiving clean evidence of where it's broken, so AI makes dysfunction survivable and therefore durable rather than forcing a fix. What eventually fills the vacuum is a set of parallel private institutions doing the bypassed work without any of the public accountability the original had.