There's a pattern that's become almost predictable. Walk into a startup or a fast-growing mid-size company and AI is already in the work — not as a rollout, not as a strategy, but as something individuals quietly started doing and never stopped. A tool someone tried on their own. A prompt that made something faster. A task that changed without anyone naming the change. And even inside large enterprises — where governance policies and steering committees exist — individuals are doing the same thing quietly, outside the official rollout entirely. Nobody decided this. It just happened at the individual level, person by person, and mostly it's working, until suddenly it isn't.
Walk into a large enterprise and there's something else running in parallel: a steering committee, a governance policy, a rollout plan with an executive sponsor. Declarations that the company uses AI — in board presentations, earnings calls, press releases — less to describe how work is changing and more to signal to shareholders, directors, and competitors that they are not being left behind. Someone decided this. And mostly it's not working either.
Both instincts are defensible. Small organizations move fast because they can. Large ones centralize because the risk of experiments that don't add up to anything is real. The problem isn't the instinct. It's that neither approach has answered the question underneath it: who should own which decisions, over what time horizon — and in either case, what happens to the managers caught in the middle?
John: The decisions only leadership can make
There are three versions of this conversation, and they tell you everything about where an organization actually is.
The first happens in the board deck. The company uses AI. It's in the slide. It's in the earnings call. It's in the press release. The message is aimed at shareholders, directors, and competitors: we are not being left behind. Whether it's true in any meaningful sense is a secondary concern.
The second happens at the conference dinner — somewhere with good wine and people trying to sound like they have this figured out. This isn't the board deck. Nobody there needs to be reassured the company is keeping up. This is something more personal: the quiet competition between people who are supposed to have answers. Someone asks how the AI rollout is going. The last thing you want to say is "we're monitoring it." You want the number. Leaders are asking "how many people are using AI" — they crave a number, any number, that says they're moving.
The third version happens over coffee, with someone you've known long enough that they don't have to perform. I've had versions of it enough times now that they've blurred together — different people, different companies, same admission: I don't actually know if this is working. I know we're doing it. I know we're spending on it. But I couldn't tell you whether it's changing anything that matters.
That gap — between the board deck and the coffee — is where the real diagnosis lives.
The question that only leadership can answer is not "how many people are using the tools." It's four harder ones: What is our actual competitive advantage — our moat — and how does AI either protect or threaten it? What do we believe about how our industry will be structured in five years? What work, if we automate it, makes us stronger — and what work, if we automate it, hollows us out? And what do we owe the people in the middle of this?
These are not questions the front line can answer. They require someone who can see the whole structure. The front line can tell you what the work feels like. They cannot tell you what the work is for.
This is not a new failure mode. When CRM arrived, most sales organizations used it to log what they'd already done, in the categories they'd always used, at the end of a process that hadn't changed. When email arrived, nobody redesigned around it — it just sped up the memo and became the accidental dictator of our daily priorities. The technology became a faster version of the existing mental model. That's the same thing happening now with AI, at every level of the organization.
When Ben pushed back in our conversation and said the four questions I was laying out "still sound like bottom-up structures," I'd resist that framing. The answers have to come from the bottom up — the people closest to the work know things that never reach the C-suite. But knowing what's happening in the work is not the same as knowing what the work is for. Those are different kinds of knowledge. And someone has to be accountable for the second one. If that accountability doesn't sit at the top, it doesn't sit anywhere — and organizations without it end up with a thousand small optimizations that never add up to a direction.
Ben: What the front line knows that leadership can't see
John's right that someone has to own the question of what the work is for. I'm not arguing against direction. I'm arguing against the assumption that direction is enough.
Here's what I keep seeing. Leadership sets the strategy, picks the tools, funds the rollout — and then treats adoption as an execution problem. As if the only thing standing between the plan and the result is whether people do what they were told. But the people closest to the work are making hundreds of small decisions every day that no strategy deck can reach. Which tasks are actually worth automating. Where the tool helps and where it quietly makes things worse. What the customer actually needs that the dashboard doesn't capture. That knowledge doesn't travel upward on its own. It has to be pulled, deliberately, and most organizations have no mechanism for pulling it.
So you get a strange thing: leadership confident it has a plan, the front line confident the plan misunderstands the work, and a layer of managers in between trying to translate between two groups who don't quite believe each other.
The manager in the middle is where this actually breaks. They're handed a mandate they didn't shape and a team whose real concerns they can't fully answer. They absorb the pressure from above and the anxiety from below, and they're given almost no authority to change either. We keep asking that layer to carry the transition without giving them any say in designing it.
Where we land
We didn't fully resolve this, and that's the honest part. John holds that accountability for direction has to sit at the top or it sits nowhere. I hold that direction without a real channel from the front line is just confident guessing. Both are true.
What we agree on is the part most organizations miss: the manager caught in the middle is not an implementation detail. They are the place where the two kinds of knowledge are supposed to meet — and right now, in most companies, they're the place where both kinds go to die. Before you decide who owns which decisions, look at whether the person you've put in the middle has been given anything to decide at all.