Three positions

Arguments I am willing to defend in a room, drawn from work I actually did rather than material I read.

When not to use AI

The strongest thing an AI advisor can say is "not here". A working argument for deterministic automation, for leaving processes alone, and for pricing the cost of being wrong.

The under-owned release gate — how do you know it is good enough to ship?

A business case does not establish whether an AI system is reliable enough for its intended use. That requires an owner, acceptance criteria, representative evaluation and regression evidence. The methods are well documented; the ownership usually is not.

Your AI strategy is an incentive problem

Silos do not persist because of the org chart. They persist because two departments are measured on metrics that pull against each other — and no model fixes that.