FeaturedMethod · 9 min
The AI-native test: take the AI out and see what's left standing.
Most "AI initiatives" are AI-adjacent — the product still works without the model, so the model never becomes the moat. Here's the two-minute test we run in every intake, and why adjacent work plateaus at exactly the point it starts costing real money.
Read the piece →
Compounding · 6 min
Why your pilot plateaued
Automation is a cost you cut once. The loop is the part everyone skips — and the only part that appreciates.
Read →
Data · 7 min
The five levels of data readiness
What each level actually unlocks, and the one you can't build a learning system below.
Read →
Spend · 5 min
Toolitis, itemized
What fourteen overlapping AI subscriptions cost a 200-person company — in dollars, and in decisions.
Read →
Method · 8 min
Scoring where to start, defensibly
The five factors we rank opportunities on — and how to defend the order to a board that wants the shiny one first.
Read →
Adoption · 6 min
The system nobody opens
Deployment is not adoption. What we changed after watching a good build sit unused for six weeks.
Read →
Buying · 5 min
Build, buy, or rent 70% of a fit
A plain decision rule for when SaaS is still the right answer — and when owning the workflow wins.
Read →Last updated Feb 2026
