In our view, the harder obstacles to adoption are now organizational: who owns the budget, who decides which model runs which workflow and how many overlapping tools were bought in the meantime.
The technology is no longer the hardest part.
For most of what enterprises want from AI, the capability is, in our view, sufficient for many uses. The difficulty has moved to questions that capability was never going to answer.
Tool sprawl is the visible symptom. Several teams each procure something reasonable, the functions overlap, nobody owns the total and the combined bill arrives without a corresponding combined benefit.
Decision rights come first.
The organizations best placed to handle this are those where somebody can say, without convening a committee, which workflows are in scope this quarter and which model serves them.
That is a governance question before it is a technology question. It is also why an assessment should start with the people accountable for an outcome: if nobody owns the outcome, the choice of tools has no criteria against which it can be judged.
What we would do first
Inventory what is already bought, attribute the spend to teams and find the overlaps. This work is routine, and it reveals duplicate tools that nobody has counted. It also produces the list of decisions that need an owner, which an organization needs before it can expect any return on its spending.