Three Questions to Test the AI Architecture

On Monday, I argued that almost every institution has an AI strategy, and far fewer have a coherent one, and that the work of the Architect imperative is to ensure the coherent strategy exists in writing for the institution itself, not just for the audience the published strategy is written for. Today I want to give you the tools.

The tools are three questions. They are not technical questions. They are leadership questions, and they are uncomfortable to ask out loud in a senior team that does not yet know whether its strategy is coherent or only published. The discomfort itself is the signal.

Question one: Name a use case the strategy refuses.

Every coherent strategy refuses something. The institution has decided that there are AI use cases it will not pursue, even when peers are pursuing them, even when vendors are offering them, even when middle management is asking for them. The reasons for refusing vary. The institution may determine that the use case is inconsistent with its trust posture. It may judge the resource cost to be misaligned with the strategic priority. It may judge the institutional readiness to be insufficient. The reasons matter less than the fact of refusal. A senior team that cannot, in the room together, name a single AI use case the strategy refuses to address. They have an ambition list with no boundaries.

The senior team that can name a refusal will usually disagree about whether it is the right refusal. That disagreement is healthy. The senior team that names no refusal at all is the team in trouble.

Question two: Identify a decision the strategy has deferred, and name the trigger that will reopen it.

Coherent strategies defer decisions. They have to. The technology is moving faster than the institution can absorb, the geopolitical context is unsettled, and some of the decisions worth making this year cannot yet be made well because the underlying conditions are still forming. A coherent strategy is comfortable with deferral. What a coherent strategy is not comfortable with is silent deferral.

The trigger that reopens a deferred decision is the test. A coherent strategy says: we are deferring the decision on this category until the regulatory position in the EU clarifies, our data foundation can support it, or the cost curve crosses a specific threshold. A senior team that can name both the deferred decision and the trigger has done the work. A senior team that knows decisions are being deferred but cannot name the triggers is letting the deferrals harden into avoidance, which is the second most common failure mode of the Architect imperative after the absence of refusals.

Question three: When one of the strategy’s commitments turns out to be wrong, what is the path back?

The strategies that survive are those with a path back built in. The strategies that fail are those written as if they will not be wrong. In the AI era, every strategy will be wrong about something within 12 months. The question is whether the institution has the architecture to revisit the wrong commitment without political cost, without losing face publicly, and without freezing while the revision is being made.

The senior team that can describe the path back, with specifics, has a coherent strategy. The path back usually involves a quarterly review with explicit permission to change commitments, a small standing group with the authority to recommend changes between reviews, and a public posture that names the institution’s willingness to revise as a strength rather than a weakness. The senior team that cannot describe the path back is writing strategies that will become brittle the moment they meet the world.

If you ask these three questions in your next senior strategy review, the most useful thing you can do is sit with whatever the answers reveal, including the silences. The silences are themselves answers. Where the room goes quiet, the Architect imperative has work to do.

Friday’s post is the sharper-edged version of this argument: what to do when the strategy is, on inspection, only a slide deck. It is not the most pleasant post in the eight-week sequence. It is one of the more useful ones.

The AI Fluent Leader is available on Amazon worldwide.