A board dinner on AI needs one question, sent with the invitation, sharp enough that the table disagrees about it. The question carries the evening. A weak one produces a pleasant dinner and nothing else.
The answer, in one line
Three tests: the people at the table own the answer, the question divides the table into two respectable sides, and nobody present can settle it with technical expertise.
Definition#
A dinner question is the single question a facilitated dinner is built around: named in the invitation, owned by the people at the table, and open enough that informed guests will take different sides.
Three tests for a good question#
- The table owns it. The people eating can act on the answer in their own organisations. A question about what governments should do fails this test.
- It divides the table. If everyone agrees by the starter, the evening has nowhere to go. A good question has two respectable answers.
- Nobody can settle it with expertise. If the most technical person present could end the discussion with a fact, the question is about technology rather than judgement.
Twelve to choose from#
On decisions
- Which decision in this company should a machine never make, and would we notice if one started to?
- What has AI already decided here that nobody at this table approved?
- If a system we rely on made a serious mistake tomorrow, who would find out first, and would it be us?
On people and capability
- If our best junior people never do the junior work, who runs this place in ten years? The argument is on the missing rungs.
- How would we tell work that sounds senior from work a senior person has checked? See synthetic seniority.
- What must this organisation still be able to do if the tools stopped working for a month?
On accountability
- Who here could stop our most important AI system tonight, and how long would it take?
- When did someone last override an AI output in this business, and were they thanked for it?
- What would we have to be shown each quarter to be confident we are in control?
On strategy
- If every competitor has the same AI, where does our advantage come from? See where the advantage goes.
- What becomes more valuable in this business as AI gets cheaper? See what becomes more valuable.
- What are we prepared to tell our people about AI and their jobs, and what are we not? See what to tell employees.
How the question moves through three courses#
Over the starter, a ten-minute provocation puts the question to the table and the table argues. After the main course, while the plates are cleared, the chair plays back what was said: where the table agreed, where it divided, and the sharper question underneath the one it started with. Over dessert the table settles what it now thinks, and each guest names one thing they will do or ask for. The format is described in full on the board dinner.
Questions that fall flat#
"What is our AI strategy?" Too broad for one evening, and it invites a presentation.
"Will AI take our jobs?" Nobody at the table owns the answer, so the talk drifts to forecasts.
"Which tool should we buy?" A procurement question. It belongs to a process with specifications, and the expert at the table wins it.
The facilitated board dinner. Ten to fourteen guests, one question, chaired from start to finish. The board dinner.
The longer formats. The briefing and the away day. Board away days and briefings.
Reference · SS-2026-407
Hirji, R. (2026). Questions for a board dinner on AI. The SuperSkills evidence base, SS-2026-407. https://thesuperskills.com/research/questions-for-a-board-dinner-on-ai. Last reviewed 4 October 2026.
An evidence review by Rahim Hirji, not peer-reviewed research. For a material claim, cite the underlying study as well; every study here carries its own permanent link.
How citations and IDs work