The disagreement starts when productivity inside a broken workflow is treated as transformation.
Most leadership teams already believe AI should make people more productive. That belief is reasonable. People are already using copilots. Individual work is getting faster.
The harder question is why so little of that speed shows up as a reliable operating result.
Faster work is not the same as better work
If a process depends on five systems, three handoffs and a person who “just knows,” making that person faster does not fix the process. It produces faster incomplete work.
This is why organizations can report high individual AI use and still struggle to point to enterprise value. The model is doing what was asked. The environment around it was never designed.
What to look at instead
Ask where time is actually spent: reconstructing context, waiting on another team, re-entering the same facts, or checking whether an AI draft is allowed to go out.
Those are environment problems — data, workflow, permissions, evaluation — not model-selection problems.
