Three things to settle before any AI project: destination, engine, route
Companies starting with AI usually begin with tools: which assistant, which platform, which pilot. Tools are the last question, not the first. Three others come before them, and a project that skips any of the three tends to drift.
Where are you going?
A description of the company you want to become: what it does, who does what, people and systems, how decisions are made, what makes it hard to compete with. Specific enough that a project can be checked against it. Without it, every department optimises its own corner and the results do not add up.
What does the work run on?
The working core that every process plugs into: how the company notices what is happening, understands it, decides, acts, and learns from the result. Today that core is mostly people in meetings and emails. In a company built around AI, much of it runs continuously in software, with clear limits on what is decided automatically and what goes to a person. And with controls that are never off: tests that catch a system getting worse, a log of every action, a way to undo, and a queue for human review.
How do you get there?
The route: a sequence of steps, each with a clear end. Agree the destination. Measure where you stand. Capture the knowledge in people's heads. Strip out the work that exists only to pass information around. Move decisions over in waves, proving each one. Then reorganise around what you built. The order matters more than the speed.
Engine first
If you can do only one thing this year, build the core: the log, the controls, one process running properly on it. Speed without that structure is how companies get hurt. Structure without speed is merely slow, and slow can be fixed.
Destination, engine, route. Then tools.