Key Concepts

The vocabulary this site assumes

Thirteen ideas that recur across the books, the tool, and the site generally — grouped by what they're about rather than the order they occurred to us. Jump to any of them, or read straight through.

Five capabilities

What AI makes possible

Pervasive Intelligence

Reasoning is no longer confined to a specialist function or a handful of senior minds — it's available throughout the organisation, all the time. That's a gift and a governance problem in equal measure. The systems doing this reasoning can generate insight, spot patterns, and propose conclusions with real fluency, but they can never stand behind a decision or bear its consequences. Pervasive intelligence means separating what can be delegated — exploring, sensing, challenging — from what can't: judgement, ownership, accountability. Get that separation wrong, in either direction, and the governance fails.

Perpetual Sensing

Governance used to move in quarterly rhythms: information arrived curated, weeks old, at scheduled meetings. AI collapses that lag. The organisation can now sense itself continuously — repairs, sentiment, turnover, compliance — as patterns form, not after they've calcified into problems. That doesn't remove the need for judgement about what actually matters; it changes the question a board asks from "did we comply?" to "are we still coherent with what we said we stood for, as the picture shifts in real time?"

Generative Coherence

No single person, or system, holds the whole organisation in view at once — which is exactly why commitments quietly contradict each other over time, not through anyone's incompetence, but because the connections were never visible when the decisions were made apart. Generative coherence is the capability that holds the pattern steady and asks whether this is still what was actually meant. It can't decide which commitment should win when two conflict — that's irreducibly human work — but it can stop a board from discovering the contradiction only after acting on it.

Radical Transparency

Boards used to control the explanation: what got disclosed, when, and how it was framed. AI makes that arrangement hard to sustain, because stakeholders can now question the reasoning directly rather than waiting for an account to be volunteered. This isn't exposure without limit — residents, regulators, and funders each see what relates to their legitimate interest — but it is a shift from transparency as something offered to transparency as something that can be demanded and inspected. The board's authority doesn't shrink; what disappears is the option of claiming an understanding nobody else can verify.

Integral Ethics

Ethics applied after a bad outcome is damage control, not governance — and algorithmic systems remove the luxury of correcting course afterwards, because they execute at a speed and scale that outpaces retrospective review. Integral ethics means values have to be designed in before deployment, not bolted on once something's gone wrong. It's harder than it sounds: AI systems arrive with values already embedded by their developers, often without anyone having written them down as a specification, and a board has to test for that, not assume good faith. The real test is what the framework does under pressure — when it's inconvenient, does it still hold?

The architecture

How it's structured

The 1+5 Architecture

Governance Intelligence, plus five sensing domains: Asset, Resident, People, Process, and Boundary Intelligence. Each domain senses continuously and reports; none of them decides. Governance Intelligence is categorically different — it doesn't sense, it holds the constitutional conditions under which the sensing can be trusted and acted on. The "1" isn't a sixth equal component sitting alongside the other five. It's the conscience the other five operate within.

Mode 1 / Mode 2

The most consequential design decision in the whole architecture: what can the organisation's AI do on its own, and what must always come back to a human being. Mode 1 is operational — sensing, monitoring, the things that are, in principle, progressively automatable. Mode 2 is ethical — adjudicating between competing goods, the things that are not automatable now and will not become automatable later, because they require a form of authority that only comes from having to live with the consequences. Mode 2 isn't a placeholder for "not yet automated." It's a permanent commitment.

The Membrane

A cell membrane doesn't run the cell's metabolism — it maintains the conditions under which that metabolism can function coherently: what passes freely, what requires active transport, what's excluded outright. Governance Intelligence is the membrane of the organisation in exactly this sense. It doesn't direct the five sensing domains; it holds the shared values, the pre-authorised parameters, and the escalation routes that let them operate without drifting from what the organisation is for. Unlike a biological membrane, this one didn't evolve — its selectivity is a constitutional choice, which means it has to be actively maintained rather than assumed to be working.

The Dynamic Risk Perimeter

Risk management has always assumed a fixed boundary: here's the organisation, here's the outside world, risk comes from out there. That assumption doesn't survive contact with an AI-native organisation. The real risk perimeter isn't the legal or physical edge of the organisation — it's the edge of what it's currently watching. An organisation actively tracking the regulatory pipeline has a materially different risk horizon than one that reads the sector risk profile once a year and files it. And because that perimeter can widen or deepen on its own initiative, deciding where it's allowed to expand into new territory is itself a governance decision, not a technical setting.

The Governance Intelligence Tests

Twenty-nine tests — fourteen operational, fifteen ethical — that exist to falsify the architecture's own claims rather than confirm them. The operational tests are mechanistic: did the designed process actually execute as designed. The ethical tests need a human being in the room, capable of noticing what the minutes smoothed over. As AI takes on more of the operational assurance work, the ethical tests don't become less important — they become the thing that was always going to matter more.

Read all 29 tests →

Foundations

What it rests on

Telos

Human beings have telos — an intrinsic orientation toward flourishing, built into what they are. Organisations don't have this naturally; their purpose is conferred on them by the people who create and govern them, which is exactly why organisational purpose can drift, get captured, or quietly get forgotten if nobody's holding it. AI systems have neither telos nor conferred purpose — only function, defined externally and executed without any orientation toward flourishing at all. A board's constitutional job is custodianship of conferred telos: making sure function, whether habitual or algorithmic, never quietly substitutes for purpose.

Incommensurable Goods and Trade-offs

How many units of efficiency equal one unit of dignity? The question isn't hard to answer — it's wrong to ask. Some goods belong to different orders of value entirely, and forcing them onto a common scale doesn't resolve the tension between them, it dissolves one into the other, which is a moral choice dressed up as a technical one. Safety versus autonomy, individual need versus collective resource, a current resident's kitchen versus a family still waiting for a home — these aren't difficult sums. They're judgements about what kind of organisation this is. An algorithm can show the frontier of possible trade-offs with real clarity; it cannot tell you which point on that frontier to choose, because that choice is where governance actually happens. Optimisation earns its keep by making the trade-off visible. It cannot make it for you.

Compliant by Design

The shallow reading is straightforward: encode the regulatory requirements so the organisation can't produce an output that breaks them without the breach being immediately visible. Useful, but not the point. The deeper reading is that the organisation's declared purpose is held alongside the regulatory frame, and everything is tested against both at once — not just "is this compliant" but "is this still coherent with what we said we were for." An organisation can pass every regulatory test and still be drifting from its own founding purpose. Compliant by design is what catches that drift before it becomes structural.

See how this is built out in the books