I've been thinking about a pattern in the conversations I've had over the last few weeks with leaders working through their AI rollouts.
Many of them brought up "breaking down silos" or "being more productive" as the goal (e.g. get rid of specialist work and let generalist decision-makers work faster). Almost all of them, however, are stuck in the same spot. Sure, every function can be faster on their own. Finance can be faster. Recruiting can be faster. Ops can be faster. But none of it adds up to the enterprise moving faster, improved quality, improved productivity, or improved SG&A. Sometimes the gains even work against each other.
Here's what that specialist-to-generalist trade actually costs, and it rarely shows up in the rollout readout. A generalist working faster is still a generalist, now making calls a specialist used to catch before they got expensive. Recruiting's new AI tool cuts time-to-screen in half and hands a generalist hiring manager a shortlist by lunch. Great news, until that manager, moving faster now with less specialist judgment in the loop, makes an offer that skips a compensation-band check a specialist would have flagged on reflex. The function got faster. Eight weeks later the enterprise picks up the cost in an SG&A line nobody traces back to the rollout that made everyone so much more productive.
This isn't the silo or specialist problem people think it is, nor is it an AI problem. It's a disconnect between two things that get talked about as if they're the same: breaking down silos and orchestrating an enterprise.
I've watched the identical pattern play out with no AI involved at all. A culture initiative gets written into every manager's scorecard, and a year later engagement hasn't moved in three business units, because nobody orchestrated what happens when a manager's actual day-to-day behavior contradicts what the scorecard says. A reorg draws three departments into one box on an org chart and calls it collaboration, and none of them have agreed on who breaks the tie when their priorities conflict. Put an AI label on the pattern and it feels new. It isn't. It's the same leadership gap, wearing whatever this year's initiative happens to be wearing.
Integration is what most of these rollouts have actually solved for (or at least tried to). Systems talk to each other, data moves, dashboards populate, tasks get automated. That's real work, and it matters, but what does it solve at an enterprise level when executed at department or team levels?
Usually, less than the rollout readout suggests. Disconnected work creates activity. Integrated work creates alignment. Orchestrated work creates execution. Most organizations doing this work right now are still fighting the first fight, tearing down silos that took a decade to calcify, and that fight is real and worth having. But alignment achieved at the department level, however well executed, doesn't automatically convert into execution at the enterprise level. Something has to carry it across that boundary, and that something is a decision, not a dashboard.
Orchestration is different. It's deciding which parts of the business need to move together right now, which parts should stay separate on purpose, and who owns the timing and tradeoffs between them. That's a leadership call, not a systems one, and in most of the conversations I've had, nobody's actually been assigned to make it.
It's also worth being precise here. There's real software doing real orchestration work, coordinating agents across systems, and that matters too. It's just a different job than the one I'm describing. Software can orchestrate agents. Leaders must orchestrate outcomes, and own what happens when something breaks.
The other thing I keep noticing: "eliminate every silo" isn't always right. I've seen separation that was a deliberate, smart design choice, and I've seen separation that was just protecting someone's turf. The job was never elimination. It's telling the difference.
So before your next AI rollout, I'd skip the question of which silos to kill, what process to automate, or what department to make more productive. Instead, ask which parts of the business actually need to move together right now. Then ask who owns making sure they do.
Then ask one more question, because it's the one that actually gets skipped in the excitement of a rollout: what happens the first time two of those parts disagree about timing? Because they will. And whoever hasn't been assigned that decision yet is about to inherit it by accident, usually in the middle of the initiative everyone already thinks is working.
That answer tells you more about where your organization really stands than your tech stack does.
Which silo in your business is protecting something real, and which one is just costing you money?