Most companies do not have an AI strategy. They have an agent, a pilot, and a folder of enthusiastic screenshots. The agent demos well. The screenshots circulate. And then a quarter passes and the workflow is unchanged.
A tool is not a system
Real transformation happens when intelligence becomes part of the system. That means connecting models to the correct knowledge, tools, permissions, software, workflows and people.
The connective tissue is the work: permissions that determine what the model may read; tools that determine what it may do; workflows that determine when it acts; approvals that determine what requires a person; and interfaces that determine whether anyone actually uses it.
The order matters
Most AI agencies start with a tool and look for somewhere to put it. The reliable order is the opposite: understand the work first — inputs, decisions, exceptions, outputs, responsibilities — then engineer intelligence into it.
That is the difference between experimenting with AI and engineering with it. Only one of those compounds.
What to ask instead
Instead of "which AI tool should we adopt?", the questions that produce systems are: What enters this workflow? What should come out? Who is involved? What is decided, and by what rule? What requires human approval? What happens when something goes wrong?
Answer those honestly and the architecture largely writes itself. Skip them and no tool will save you.