The Organizational
Intelligence Canvas

Ten questions for designing an organization where human and machine intelligence compound rather than substitute. Successor to the Human-Centered AI Canvas (2019).

Tap any tile to reveal its provocations — then reveal a worked example.
Layer 01 — Direction
01

Intent to Amplify ◎

What this organization is for, and the human capacity it exists to enlarge.

+ Reveal provocations
  1. What is this organization for — and which human capacity does it exist to amplify?
  2. If intelligence were free and abundant, what would still be scarce here?
  3. Which problems are you choosing not to hand to a machine?
Worked example — a design team inside a creative agency
  1. Framing. The team exists to make a client's problem legible before it makes anything usable.
  2. Taste, judgment under ambiguity, and the standing to tell a client they are solving the wrong problem.
  3. The framing conversation with the client — and the decision about whose needs count.
02

Value Portfolio ↗

Falsifiable bets, not ambitions. Where value is migrating, and what you are wagering.

+ Reveal provocations
  1. Where is value migrating — are you defending the core or inventing the next one?
  2. State the bet: what value, for whom, measured how, by when?
  3. Which AI investment would you kill tomorrow if forced to prove impact?
Worked example — a design team inside a creative agency
  1. From deliverable volume to framing and orchestration. Inventing, not defending — the hours-based core is deflating.
  2. 30% of engagements sold as system design rather than deliverable count by end of 2027, measured on revenue mix.
  3. Per-seat licences adopted designer by designer: real cost, zero compounding, no shared context.
03

Human Stakes ⚖

The honest ledger: pains relieved on one side, pains created on the other.

+ Reveal provocations
  1. Which pains does this relieve — and which does it create: intensification, surveillance, loss of craft?
  2. Who bears the cost of this transformation, and who captures the gain?
  3. What would your people say if they could speak entirely freely?
Worked example — a design team inside a creative agency
  1. Relieved: spec grinding, competitive audits, deck production. Created: three concepts now expected in the time one used to take; “I made this” quietly becoming “I approved this.”
  2. Juniors bear the cost, margin captures the gain. Saying that out loud in the room is the entire point of this block.
  3. “I'm faster, and I'm prouder of less.”
Layer 02 — The Human-AI Arrangement
04

Delegation Spectrum ◑

The autonomy dial: assisted, supervised, autonomous — and who turns it.

+ Reveal provocations
  1. For each stream of work: assisted, supervised or autonomous — and who decides when it moves?
  2. What earns an agent more autonomy, and what revokes it?
  3. Which delegations are reversible — and which have quietly stopped being so?
Worked example — a design team inside a creative agency
  1. Desk research → autonomous with audit. Synthesis → supervised, human owns the claims. Concept generation → co-pilot. Client framing and interviews → human only.
  2. Earned by a clean audit trail across ten engagements. Revoked by a single undisclosed error reaching a client.
  3. Research synthesis looks reversible. Once nobody reads a raw transcript, the skill is gone inside eighteen months.
06

Teaming Contracts ⇄

Rules of engagement between people and agents: ask, act, hand off, escalate.

+ Reveal provocations
  1. When should an agent ask, and when should it act?
  2. How does work hand off between people and agents without losing context or ownership?
  3. What does a good escalation look like — and who is reliably on the other end?
Worked example — a design team inside a creative agency
  1. Ask before anything client-facing, any claim about a user, any commitment to a spec. Act freely everywhere else.
  2. Every AI-produced artifact carries provenance: what made it, what was verified, and by whom.
  3. Escalation goes to a named design lead — never to “the team” — answered inside one working day.
05

The Human Agency Boundary

What stays human on purpose

  1. What stays human not because machines can't, but because they shouldn't?
  2. Where is human judgment constitutive of a decision's legitimacy — not merely its accuracy?
  3. How will you defend this line the day efficiency pressure comes for it, and who holds the pen?
Worked example — a design team inside a creative agency
  1. Critique. Not because AI cannot critique, but because critique is how taste is formed. Automate it and you stop producing designers.
  2. A synthesis can be entirely defensible while nobody is accountable to the participant whose words were used. Only a person can carry that.
  3. Written into the staffing model: every engagement has a named human framer, billable, visible on the SOW. A boundary that lives in the values deck is decoration.

Draw this line before you design the system beneath it. A boundary declared under pressure is not a boundary — it is a concession.

Layer 03 — The Work System
07

Collective Intelligence ⟳

How what one team learns becomes what the whole organization knows.

+ Reveal provocations
  1. How does what one team learns become what the organization knows?
  2. Where does AI help you think together — and where does it flatten dissent into consensus?
  3. What can this organization now sense that it could not before?
Worked example — a design team inside a creative agency
  1. A shared context layer — accumulated client knowledge, patterns, past critiques — that every agent reads from and writes back to. The context compounds; the tooling doesn't.
  2. Observed failure: a critique session where everyone arrived at the same AI-shaped concept and mistook it for consensus.
  3. Patterns across every engagement at once, which no single designer was ever positioned to see.
Layer 04 — Foundations
08

Capability & Craft ✦

Which capabilities are expanding, which are quietly atrophying, and where craft is learned.

+ Reveal provocations
  1. Which capabilities are expanding, and which are quietly atrophying?
  2. Which skills must people keep practising even when AI could do them faster?
  3. Where do newcomers learn their craft once the entry-level work is automated?
Worked example — a design team inside a creative agency
  1. Expanding: breadth, exploration speed, code fluency. Eroding: sitting with ambiguity, defending a choice under pressure.
  2. Sketching before prompting. Writing the research question by hand.
  3. Apprenticeship used to run through grunt work. Making juniors the verification layer is a trap — they cannot yet judge what they are checking. Pair them into framing earlier instead.
09

Oversight & Accountability ▣

Seeing in time to intervene — and knowing who answers by name when an agent acts.

+ Reveal provocations
  1. Can a person reconstruct why the system acted as it did — in time to intervene?
  2. When an agent acts, who is answerable — by name, not by committee?
  3. What must this system never do, encoded where it cannot be argued away?
Worked example — a design team inside a creative agency
  1. Not yet. We cannot reconstruct which parts of a deliverable were generated — that is the first gap to close.
  2. The named human who signs the deliverable. One name, on every document that leaves the building.
  3. Synthetic participants are never presented as real research. Written into the contract, not the handbook.
10

Narratives & Readiness ↻

The story the organization tells itself, the identities it unsettles, the futures it rehearses.

+ Reveal provocations
  1. What story does this organization tell itself about AI — and who does that story leave out?
  2. Which identities and sources of pride are being destabilized, and what replaces them?
  3. How will you rehearse a future you are not ready for, before it arrives?
Worked example — a design team inside a creative agency
  1. “AI makes us faster” leaves out the people who suspect their craft is the thing being made faster.
  2. “I am a maker” is destabilized. What can replace it: “I am the one who decides what is worth making.”
  3. Run one full engagement inside the new model before rolling it out — and let it fail somewhere cheap.