Essay · 2025
Enterprise AI Problems Don't Look Like AI Problems
Customers, partners, and employees complain in different vocabularies. Read any one channel and you get a list of local irritations. Read all of them together and the same handful of structural problems keeps appearing, described from whichever seat the person happened to occupy.
Ownership is the usual root cause
The breakdowns that persist longest are the ones that sit between teams. No single group is failing; the handoff is. That is why they survive reorganizations, tooling changes, and model upgrades — nothing in the org chart is responsible for the seam.
An AI layer placed over an unowned seam inherits the seam. It also makes it harder to see, because the failure now happens inside a system nobody can inspect.
Synthesis is the method
Cross-channel synthesis is slow and unglamorous and it is the only technique I know that surfaces this class of problem. Five hundred pieces of feedback, coded consistently, will collapse into a small number of themes that no individual study could have produced.
The output is not a report. It is a map that lets a roadmap conversation happen at the level of the system rather than the ticket.
What changes when you name the system
Once a breakdown has a shape and a location, sequencing arguments get easier. Teams stop debating whose fix is more urgent and start debating which dependency has to move first. That is the point at which research starts influencing strategy instead of documenting it.