A decision-making framework for the AI era: who decides, the person or the machine?

The 9Q Grid was designed after Helix's work with RAA; we have since mapped RAA, City of Mitcham and our other client work onto it.

Classic decision-making frameworks sit in familiar families. Rational models move from define the problem to gather evidence to choose. Intuitive approaches trust pattern recognition when time is short and the situation is familiar. Matrix methods score options on two or more axes so a group can compare bets without pretending every criterion is equal. Each of those families still assumes a human team is the only actor that decides.

None of them asks the AI-era question: should a person keep the call, share it with a machine, or hand the act to the machine under governance? That is the gap the 9Q Grid fills. It is a decision-making framework for work that is already shared with AI — not a replacement for every prior model, and not a maturity score that ranks Adapt above Assist.

Lists such as “four Cs of decision-making” circulate widely. Treat them as memory aids for human process. They do not place the machine on the map, and this page does not invent a four-C list. For use-case scoring matrices — value, feasibility and then grid placement — see how to prioritise AI use cases.

Here the question is operational: who decides when AI is in the work?

The agency axis of the 9Q Grid names three human roles for any decision shared with AI. The wording is from Chapter 3 of The Co-Intelligent Organisation.

Assist keeps the human at the helm. Augment keeps the human in the loop. Adapt puts the human on call. The higher the agency, the more trust must already be earned. Position is not permission. For how those roles scale from organisation practice to agent systems, see AI autonomy levels and the Assist / Augment / Adapt guide on Agent Watch.

A practical tell: if a person must clear every option, you are still in Assist or Augment. If the system acts and people only hear about exceptions, you have already moved into Adapt — whether the slide deck admits it or not.

The same agency level is a different decision on different terrain. The book’s work axis has three columns, each with its own example class.

• Structured work — defined, repeatable, rules-heavy. Example class: an accountant with Copilot (Q1 Operator) still owns the filing decision; the machine accelerates a known line of work.
• Dynamic work — iterative and feedback-rich. Example class: a strategist with a model (Q5 Partner) decides together with the machine across rounds of refinement; neither is the sole author.
• Complex work — emergent and hard to specify. Example class: emergency coordination with signal support (Q7 Navigator) keeps wayfinding human while AI improves the map.

Crossing agency with work gives nine quadrants. A Q2 Tactician decision (Augment × structured) — AI ranks or flags, a person clears each call — is not a Q8 Explorer decision (Augment × complex), where human and AI co-search ambiguous futures. Same “share with AI” story; different evidence, override path and trust budget.

UiPath-style claims automation sits in Adapt × structured (Q3 Automator): the machine runs the workflow; humans govern rules and exceptions. That is a different decision from a multi-agent orchestration claim in Adapt × complex (Q9 Conductor). If you fund the second with the trust of the first, the currency of trust is already overdrawn.

Use this check before you fund or escalate an AI-backed decision. It builds on the book’s “Before We Move On” questions — plot position, then ask what is coherent next — turned into a sequence a working team can run in one sitting.

1. Name the decision: what call is being made, who answers for the result, and what evidence counts as enough.
2. Place it on the grid: Assist, Augment or Adapt crossed with structured, dynamic or complex work.
3. Check the trust it needs: does the organisation already govern this agency, or are you buying a high-agency outcome with a low-agency trust budget?
4. Set the human override: at the helm, in the loop, or on call — and what triggers escalation.
5. Review: after live use, re-plot. Has the decision drifted into a different quadrant or failure mode?

Jeff Bezos’ roughly-seventy-percent information rule is a useful companion for Partner-style work: decide when you have most of the information you wish you had, then commit and iterate. It ties into Q5 Partner — decide together, then iterate. It is not a licence for Adapt-row automation without boundaries. See Bezos’ 2016 shareholder letter for the original framing.

The book names failure modes by quadrant. Three matter most when the decision is mis-shared with AI:

• Recommendation capture (Q2) — the human starts rubber-stamping the machine’s options.
• Silent drift (Q3) — automation keeps doing yesterday’s right thing in today’s changed world.
• Dyad dependency (Q5) — the human stops operating without the AI partner.

Each trap is a decision failure first: the wrong human role for the work, or no review loop when the terrain changed. Tool dependence on the Assist row and orchestra drift on the Conductor cell are cousins of the same problem — the human role was never named honestly. Full archetype detail lives in Chapter 3 of The Co-Intelligent Organisation.

One well-placed decision is not a portfolio. Plot the top initiatives, watch which quadrants are crowded, and fund by trust — not by the loudest vendor story. For the scoring-plus-grid method, use how to prioritise AI use cases. For a short self-check, take the 9Q diagnostic. Public company readings in the 9Q case library are labelled public evidence, not Helix clients.

A Helix Clarity Scan helps you see your top opportunities clearly. It does not deliver a Strategic Shape and does not plot your portfolio on the grid.