The baseline expectation that every PM can map, at a structural level, which systems drive which outcomes in their product — literally sketch the boxes and lines connecting cause to effect. AI tools (AI-Enabled PM Self-Service: Data, Code & Debugging, Hex Threads (Self-Serve PM Data Tool)) raise the bar past this baseline: once a PM has the structural map, they're expected to push into a deeper layer of understanding using AI directly rather than stopping at the diagram.
Require every PM to be able to draw how their product's systems connect before anything else; then use AI tools to go one layer deeper than the diagram, into actual mechanism and code.
Before shipping a local fix, trace its second-order effects on other parts of the system and other stakeholders — not just whether it solves the immediate problem. Verrilli's example: forcing all sellers on Whatnot to create listings (a plausible fix for a listings/search problem) has downstream effects on seller throughput, so the fix has to be evaluated against that knock-on cost before being rolled out broadly, not just against whether it improves search.