The problem it solves
Most AI failures aren’t caused by one bad decision. They’re the accumulated result of a dozen smaller decisions that each seemed reasonable at the time and were never challenged from outside.
The vendor assessment that nobody pushed back on. The internal roadmap that went to the board unchallenged. The feature that got added because a competitor had it. The project that scaled before it was proven.
Each of those was a moment where one uncomfortable question would have changed the direction. The question wasn’t asked — not because nobody was smart enough, but because nobody in the room was positioned to ask it.
What the sessions look like
Two sessions per month, sixty minutes each. No fixed agenda. You bring whatever is live: an AI vendor you’re evaluating, an initiative that isn’t landing, a decision the board is pushing, a pattern you’ve noticed that you don’t quite trust yet.
I ask the questions that don’t get asked in your regular meetings. We find the assumption underneath the presenting situation. We test whether the current reasoning is solid or whether it’s carrying weight it wasn’t built to carry.
Sessions are not recorded.
What changes over time
Leaders who stay with this engagement for 3–6 months typically describe the same shift: they stop being surprised by AI initiatives that fail to deliver. Not because they saw the failure coming — because they had been asking the right questions along the way.
The pattern-recognition that develops isn’t about AI specifically. It’s about how vendor claims are constructed, how internal proposals build consensus, and where the reasoning tends to go untested. Once you’ve seen it across a dozen decisions, you develop an instinct for it.
What this is not
Not coaching. Not AI mentoring. Not a forum for vendor updates.
This is structured adversarial scrutiny of AI investment decisions — ongoing, not episodic.
"The most dangerous AI decisions aren't made in a single meeting. They accumulate through dozens of small choices nobody challenged."
What you leave with
Not deliverables. Capabilities and clarity.
- A challenge function built in Ongoing adversarial scrutiny of AI decisions as they happen — not after the budget is committed and the problems are visible.
- Faster diagnosis under vendor claims The habit of identifying what a proposal depends on being true becomes faster with each session. You stop being surprised by what wasn't in the pitch.
- Reduced exposure to consensus-driven AI commitments When every significant AI decision gets pressure-tested before it's made, the failure rate drops — not because decisions are more conservative, but because they're better examined.
- A different relationship with AI uncertainty After 3–6 months, most leaders report a materially sharper instinct for which AI claims deserve confidence and which are performing confidence.
Good fit / not a fit
- Good fit: AI decisions arrive faster than you can evaluate them Vendors. Internal proposals. Competitor announcements. New tool categories. The pressure to keep up is constant — and the time to think critically about each decision is not.
- Good fit: Your team is aligned on AI and you're not sure you should be Everyone is excited. The roadmap makes sense. But you've noticed that the organisations that move fastest on AI aren't always the ones that get the best results.
- Good fit: You want the scrutiny built in, not occasional A one-off assessment addresses one decision. The retainer changes how you approach all of them.
- Not a fit: Not a fit AI implementation support, vendor management, or technical architecture. This is adversarial advisory on AI investment reasoning — it does not execute the technology decisions it examines.