Nearly every AI programme I see starts with a plan. A list of pilots, a budget, some dates. Martin's ten minutes explain why that list tells you almost nothing about whether any of it will matter.
A strategy makes a bet about the outcome and lays out its logic, so you can see when the evidence turns against it. A plan lists the activities you control. That's why plans feel safe, and why they rarely say what you will leave alone.
The same test works on any AI roadmap. Ask what choice it makes, what it deliberately doesn't do, and what would have to be true for it to win. If every answer is another activity, you have a plan.
