Judgment Is the Scarce Input
Why the businesses that compound in the AI era put their sharpest humans in front of the machine, not behind it.
For most of my career, the binding constraint in any business was execution. Want a model built, a market scanned, a support queue cleared? You needed people who could do the work, and good people were expensive and slow to find. That was true at KPMG, true running restaurants, true building kitchens.
That constraint has quietly disappeared. The marginal cost of capable work is close to zero now. What’s still scarce — scarcer, actually, because everyone’s chasing it at once — is judgment: knowing which number actually decides the outcome, which deal to walk from, where the unit economics break under real load instead of in a spreadsheet.
Most firms are responding to cheap capability by trying to remove humans from the loop entirely. I think that’s backwards. The businesses that compound do the opposite — they put their sharpest people in front of the machine, pointing it and editing it, and let the machine absorb the volume behind them.
This is the whole design behind how I run advisory work now: a small senior pod making the calls that actually matter, with AI carrying everything that doesn’t require a call. Not humans reviewing what a model already decided — that’s the same arrangement wearing different words, and it just moves the fatigue downstream. The pod decides first. The machine executes at scale. That order is the entire argument.