Essays, field observations, and long-form research from the Marietta practice. We write when we have something to say — which is less often than we are asked.
Everyone talks about the head (strategy) and the tail (the models). Nobody talks about the enormous, unglamorous middle — the integration layer, the data contracts, the organizational immune system that rejects change. That is where AI programs die. This essay is about why, and what we have learned about surviving it.
Request full essay →Most teams reach for accuracy metrics first. The teams that ship reliably reach for something harder to define — and we have finally found words for it.
Surveying 140 programs across 9 industries. The patterns are stable enough to name. The failure modes are new enough to worry about.
Enterprise buyers are making a choice — perhaps without knowing it — between two fundamentally different theories of what outside help is for.
None of them are what the architecture diagrams suggested. All of them were discovered on a Wednesday at 11pm.
The first pilot worked because three people who deeply understood the problem were in the room. The second pilot failed because they were too busy running the first one.
The year the language models became table stakes. What happened to the 70% of programs that still didn't ship.
We send a note when we publish something we believe is worth your time. That is roughly four to six times a year.
NO MARKETING. UNSUBSCRIBE ANYTIME.