The Chief of Staff Boom Is Really About the Orchestration Layer

When Richard Gelfond hired his first chief of staff at IMAX, he wanted a shadow: someone to take notes in meetings and tell him what was next. He was not thinking in terms of an "orchestration layer." He was merely thinking about managing his time.

That was sixteen years ago. Since then, IMAX has grown from 55 theaters to nearly 2,000 across more than 90 countries. And the shadow grew with it: Gelfond's chiefs of staff stopped taking notes and started operating as the metaphorical glue between departments. Gelfond now describes the chief of staff role as helping him think through what he wants to accomplish and then how to execute it. Without meaning to, he had created a coordination function, beginning with only one person.

Gelfond's note-taker became IMAX's nervous system. That infrastructure, repeated across thousands of companies, is a bigger story than it looks. The role nobody could quite define has become indispensable at the exact moment AI was supposed to make coordination cheap and easy.

If AI makes everyone faster, why does the job that connects everyone keep growing? 

The answer is that AI does not just change the work; it changes how that work must be coordinated. When individuals and functions accelerate, orchestrating the work becomes an in-demand capability: deciding what should happen, who or what does it, how information moves, and where decisions get made. 

That orchestration layer is now the element companies compete on. The chief of staff is simply the role where it has become visible: a role already absorbing the coordination work. Companies that will succeed with AI integration are building a well-designed orchestration layer, and they are designing a real role to lead it. In the future, the chief of staff role will evolve into a role with the mandate to route work, own guardrails, and make decisions across functions.

Why AI Chief of Staffs Are in High Demand

The structural change exacerbating the challenge shows up in the org chart. Korn Ferry's 2025 Workforce survey of 15,000 professionals found that 41 percent of employees say their company stripped out management layers. Gallup data puts the average manager's span of control at 12.1 direct reports in 2025, up from 10.9 the year before and roughly 50 percent higher than in 2013. Gartner predicts that through 2026, one in five organizations will use AI to flatten structure and cut more than half of their middle-management roles. Companies are betting that AI can absorb the coordination work those managers used to do.

‍

‍

But coordination work doesn't vanish just because it's erased from an org chart: it moves. Often it moves up, into the executive office, where it gets a new name: chief of staff. The number of people holding that title in North America has more than tripled since 2021, according to Live Data Technologies, with senior roles reaching $400,000 in tech and finance.

Need for Speed Increases Need for Orchestration

Ask most enterprise leaders what AI is for, and they will tell you it makes people faster. Faster drafts, faster code, faster analysis. So they measure success the way they measure any tool: adoption rates, licenses bought, headcount saved. But that framing hides the real problem. The hard part of deploying AI isn't building the agents; it's coordinating them: deciding what should happen, who or what does it, how information moves between systems, and where decisions get made. Plus, these agents are often deployed into existing, imperfect human systems that usually need retrofitting to work with AI.

Most large enterprises are highly matrixed. Finance, HR, engineering, product, support, and marketing each have their own leaders, systems, and priorities. That structure was built for control and stability, not for a market that changes monthly. When a competitor compresses a decision cycle from weeks to days, a slower company falls behind as speed compounds.  

This is why middle management is being cut: companies are trying to move faster by removing the layers that slow them down. But the coordination work the flattening displaced just reappeared one level up, in a role whose growth tracks the problem almost exactly. And the title keeps shifting (some call it Chief of AI, some Chief of Agents) because the need is arriving faster than the language.

But removing layers without building something to replace their coordinating function creates a new kind of drag. Korn Ferry found that 43 percent of employees say their leaders are not aligned, and 37 percent feel directionless after the cuts. In the US, 72 percent of senior executives said they felt stretched past their abilities as they absorbed the work their managers used to handle.

The chief of staff sits closest to this coordination problem, which is why the role has grown. But sitting close is not the same as being built for it. A survey by the Chief of Staff Network of more than 250 senior operators found that 86 percent use AI every day, yet only 7.3 percent qualified as what the report called AI-Native. Only 5.2 percent had AI agents actively running any part of their work. The rest use AI like a faster keyboard. The growth in Chief of Staff roles signals that orchestration has become scarce. But it is not yet the finished answer. It needs real authority and a dedicated title, not a borrowed seat.

‍

‍

Who Routes the Work Between People and Machines?

The second problem is routing: once you have decided what should happen, who or what actually does it. This is the part of the job a chief of staff has always done by hand: pulling the right people into the room, deciding what goes to which team, and holding the handoffs between finance, HR, engineering, product, support, and marketing. Each has its own systems and its own leaders, so the routing is as much art as science. AI does not remove that work. It adds a second set of options: some of those handoffs can now go to agents instead of people. The routing question gets harder, not easier, and someone still has to own it.

Walmart ran into this directly. Through 2024 and early 2025, the company built AI agents rapidly. Agents to help price items, stock shelves, request time off, and analyze sales trends. It worked, until it did not. The company ended up with so many separate agents that using them became confusing. Each was useful on its own, but together they were a maze.

"Multiple agents, even if each one is useful, can quickly become overwhelming and confusing."

—Suresh Kumar, Global Chief Technology Officer and Chief Development Officer, Walmart

So in July 2025, Walmart did something that looked like subtraction. It consolidated its collection of roughly 200 agents into four "super agents," each built for a single constituency. "Sparky" serves customers. An associate agent serves the store employees, handling scheduling, payroll, and benefits questions. "Marty" serves suppliers and advertisers. A developer agent serves the engineers. Underneath each one sit the narrow task-agents, but the person or partner interacting with Walmart no longer has to know which agent does what. The super agent routes the request to the right place.

What Walmart built is an orchestration layer. It interprets what someone needs, finds the right specialized agent, and coordinates the handoffs across systems that used to be separate. The company connected these agents using an open standard so they could talk to internal apps and data systems through one interface. The results are already measurable in places: Walmart says AI helped its software developers save four million developer hours in a single year.

Walmart's answer routes machines to machines. But someone still has to decide where human judgment enters, and not everyone agrees a person should sit at that center at all. Spencer Rascoff, the chief executive of Match Group, has said he doesn't want a chief of staff, because he sees the role as a gatekeeper between him and his teams. He leans on AI instead to organize his day and prepare for meetings. But even Rascoff draws a line at what the tools can do.

"I think there's a difference between AI helping someone do their job better and AI replacing the judgment and relationship building that a good leader, or a good chief of staff, actually provides."

—Spencer Rascoff, Chief Executive Officer, Match Group

AI cannot yet absorb the judgment about which cross-functional trade-off is worth making, or the relationships that make a hard decision stick. Routing is the job, and the hardest routing decision is which lane, human or machine, each choice belongs in. The orchestration layer still needs an owner.

Can You Trust the Information the Decision Runs On?

The third problem is moving trustworthy information to the point of decision, and it is the one enterprises least want to admit. Much of the software AI is being layered onto is already bloated, and much of the data underneath it is already bad (or dark). Putting AI on top of a broken tool just makes the mess run faster.

This is the pattern of enshittification many are familiar with from the evolution of social media and now spreading through enterprise software, where products trade usefulness for features and pricing tiers that serve the vendor above the end user. Roughly 80 percent of features in the typical cloud software product are rarely or never used, representing about $29.5 billion in waste. Now add AI to that foundation, trained on data that can be incomplete or wrong, and the failures get even harder to see or remove.

Health care shows what happens when a company trusts an AI tool without accounting for the mess underneath. Hospitals adopted a transcription tool built on OpenAI's Whisper model to convert doctor-patient conversations into notes. It promised to free clinicians from paperwork. But researchers found that Whisper invents text that was never spoken. One University of Michigan researcher found fabricated text in eight of every ten public meeting transcripts he examined. The invented passages sometimes included imagined medications and racial commentary. The tool built on Whisper, made by the company Nabla, has been used to transcribe an estimated seven million medical visits at health systems including Children's Hospital Los Angeles. And Nabla's tool erases the original audio, so there is often no way to check the AI's transcript against what was actually said.

AI transcription can be a powerful tool when used correctly, but AI dropped into a workflow with no orchestration layer to check the output and manage the data underneath will produce confident errors at scale. Someone has to own the guardrails. The tool cannot be held responsible for that, and neither can the vendor who sold it. That ownership is crucial, and it is why the person who holds it needs a real seat, not a borrowed one.

Give the Orchestration Layer Real Authority

The three problems above share one answer. Coordination is now the binding constraint, and most companies have no one clearly responsible for it with the power to act. The fix is not to hire a chief of staff and walk away. Treat orchestration as a real function and give it the authority its responsibility already demands. Here is where to start.

1. Find where the orchestration layer already lives. Somewhere in your organization, a person is already absorbing cross-functional work with no line authority to show for it. That person is your coordination function, whether or not anyone has labeled it.
2. Score AI by workflows redesigned, not seats saved. Buying licenses is not adoption. Ask which processes actually run differently now, not how many people opened the app.
3. Decide the human-versus-agent routing on purpose. For each recurring decision, name what genuinely needs human judgment and what can be handed to an agent. That call is the orchestration owner's to make, deliberately, before the tools make it by default.
4. Match authority to responsibility. The chief of staff, and every role like it, has visibility across strategy, people, and information but almost no formal power. That mismatch is the constraint to fix. Responsibility without authority is how coordination fails.

The Shadow Knows

The seat that owns the work today goes by many names. Some call it Chief of AI. Some call it Chief of Agents. Some are handing the work to a chief of staff and hoping it holds. None of those names describes the whole job, which is to own the operating system by which humans and machines share work. We propose calling that person the Chief Operating System Officer.

So here is the question worth bringing to your next leadership meeting. Stop asking how many people are using AI. Ask instead who owns the orchestration layer in your company, and whether that person has the authority their responsibility already demands. 

That seat, whatever you end up calling it, will act as the engine for your entire company.