A friend recently told me about spending a couple of months renting in a place they’d always fantasized about owning. The surprise wasn’t the view or the weather, but the learning.

Only after living there did they realize what actually mattered: which floor avoids the parking structure roof, why a third bedroom changes everything, and what “ideal” really means in practice.

That’s exactly what I see in AI adoption.

Most leaders want the “purchase decision” immediately: platform standardization, a big rollout, a clean ROI story.

But the real insight happens in the rental phase: the messy weeks when you learn how work actually happens.

In the conversation, we got unusually specific about a step I think belongs in every AI program: asking people to “go meta” on their own workflows.

By "meta" I mean taking a step back and getting a bird's eye view of their work.

Not brainstormy. Not aspirational. Concrete:

• What are the inputs you pull together without thinking?
• Where do they live (email, CRM, meeting notes, spreadsheets, your head)?
• What is the output you’re producing (an email, an investment memo, a model update, a board slide)?
• What does “better” mean here: faster, more accurate, more persuasive, higher quality?

If you’ve ever cooked without a recipe, you understand the problem.

You can make the dish by instinct, but if you had to teach someone else, you’d suddenly need to specify: how many eggs, what heat, what “done” looks like. AI is the same. If you can’t describe the recipe, you can’t delegate it to an agent, a workflow, or even a well-designed prompt.

Two important implications for CEOs (yes, even the ones who “have a team for that”):

1) This work feels slower at the start. Productivity often dips before it rises. It's the classic J-curve. That’s not failure; it’s the cost of making tacit work explicit.
2) The goal isn’t automation theater. The win is competitive advantage: higher-quality decisions, faster cycles, clearer thinking, and a team that can replicate your judgment, not just your tasks.

The practical move: build “recipe time” into your AI rollout (office hours, small-group clinics, or structured check-ins) so the hard part (clarity) actually happens.

Because the organizations that outperform with AI aren’t the ones with the most tools.

They’re the ones who can explain how work gets done.

Alex