Which work goes in.
Each member arrives with one frustrating workflow. The room maps them, applies the delegation ladder and draws the red lines.
Decided: three to five workflows enter the portfolio, each with a target rung, and the red lines are endorsed.
Better, not fewer.
Are we making people better, or trying to remove them?
Decided: an augmentation test is added to the scoring, and any candidate that starts from replacement is redesigned or dropped.
What will stop us.
Readiness is scored across four layers: data, collaboration, governance and incentives, implementation discipline.
Decided: the baseline is recorded and the single binding constraint is named, with an owner.
Where the value is.
The portfolio is scored on seven criteria. Every piece of AI already running inside your vendors' products is found.
Decided: the Opportunity Scan is signed off; each piece of vendor AI is switched on, off or gated; nothing is switched on without a register entry.
How we govern it, and how we encourage it.
The draft governance framework has been circulating and marked up since decision one.
Decided: a four-layer stack is adopted (ethics, operating rules, governance and oversight, incentives), an accountable executive is named, the green list is written, and five incentive commitments are agreed with measures. Three layers govern. One encourages.
Which two we back.
The scored portfolio is circulated beforehand.
Decided: two business cases signed off, each with what good looks like, who judges, the baseline and the stop condition; the measurement framework; the 90 and 180 day roadmap; and the readout booked. Two, because two is resourceable and comparable.
Six decisions. The AI activation program.
Decide what AI is for before someone decides it for you.
Six leadership decisions about AI, made in the right order, by the people who own the outcome. Most executive teams have been told to do something about AI and handed a vendor demo and a risk register. This is four months in which your senior people make the six calls that determine whether AI helps your organisation or merely arrives in it. One decision every three weeks. The evidence is prepared in the fortnight before, so the room argues about what it means rather than collecting it.
Send the brief01 The six decisions
One every three weeks, in this order
02 How it runs
The room decides. It does not collect.
A Council of eight to twelve people across operations, commercial, finance, technology and data, risk, and people. The sponsor never chairs. One named sceptic sits on every Council.
Six sessions, one every three weeks, each the same shape: a quarter teaching, half working on your own material, a fifth deciding, ten minutes writing it into the registers. Seventeen hours of Council time across about four months, which is roughly 130 to 200 person-hours of senior people, said plainly before it starts. The sponsor is briefed separately after each session and sits in on the sixth.
Between sessions the work does not stop. The governance framework is drafted after decision one and is in your hands within a fortnight, so by decision five it is already yours.
03 What you have at the end
Decided, written down, and owned
A portfolio of work with its red lines. A readiness baseline and a named constraint with an owner. An Opportunity Scan with every piece of vendor AI accounted for. A governance stack with an accountable executive and a green list people can actually use. Two business cases with their measures, a roadmap, and a readout to the board, the Secretary or the executive. And a room that has learned how to decide about AI, which is the asset that outlasts the pilots.
04 Where it sits
One of five steps
Six decisions is the launch step in a family of five. Each has a start condition and a leave condition, and nothing is sold as a workshop with optional extras.
The prerequisite day: mental models and personal capability.
How the work thinks
A six-week discovery program for organisations that want to map the work before committing a Council to it. Work map, red lines, one redesign brief, a shortlist, a readiness read. Its outputs stand as the Council's first outputs; nothing is redone.
Six decisions
The launch. About four months. This page.
Prove it
The activation. Six months: nine sessions, pilot oversight, sponsor sessions, champion development, a vendor review, scale-stop-invest papers and an operating plan.
The appointment
An expert AI colleague designed and launched with its governance built in, three weeks supervised. Proposed only once decision six has chosen a pilot, never before.
Technical build sits outside the program, with partners, and is quoted only after the room has chosen. Only what originates in the Council gets built.
05 For commissions and departments
A note on the word decision
At a commission or a department the word "decision" means something specific. These are six leadership decisions about AI, and they are the only decisions in the program a system ever goes near. Where there is no board, the readout goes to the Secretary, the President or the executive leadership team. Much of the third governance layer is handed to you by NSW policy already, so the room spends its time on the two layers that are not: operating rules, and the incentives that make anyone follow them.
06 Before you commit
Questions people ask
Who should be in the room?
Eight to twelve people who own work, not people who represent it. The sponsor does not chair. The head of technology is not the champion by default. One sceptic, named.
How much of our time?
Seventeen hours each in the room across four months, plus the evidence prepared in the fortnight before each session. The sponsor gives two hours in week zero and about an hour after each session.
Do we need to have done Practical AI first?
Yes. It is the day that gives the room its vocabulary and its mental models, so the Council can start by asking rather than retelling.
Which tools?
None are recommended. The program is technology agnostic. The vendor audit covers the AI you already own, which is usually more than anyone in the room knows.
What does it cost?
Fees are quoted per engagement once the brief and the room are clear. Effort is stated plainly up front, and nothing is built, bought or switched on for you without a named human saying yes.
What if we are not ready?
That is what decision three is for.
07 Provenance
Where the design comes from
Designed on the AI-Driven Leadership: Strategies for the Future program at Stanford Online, Stanford Engineering (December 2025), and on running Pathos Services itself on a team of eight AI colleagues under exactly this kind of governance. The Practical AI day that opens it has a 93% participant satisfaction record across six NSW public sector cohorts in 2026.
Start with the sponsor
A two-hour conversation. We will say plainly whether Six decisions is the right next step, or whether a day, or six weeks of How the work thinks, would serve you better.
Send the brief