Here is the typical AI rollout story: the CEO or founder attends a conference, comes back convinced that AI is going to transform the business, purchases a set of tools, and announces a new initiative. Employees are given access. A few adopt it enthusiastically. Most use it occasionally when reminded. Six months later, the efficiency gains are underwhelming, the initiative quietly loses momentum, and the conclusion — usually wrong — is that AI was not the right fit for this business.
The failure was not the tools. It was the layer between the strategic decision and the operational execution — the midlevel leaders, the team leads, the managers who were supposed to carry the transformation into day-to-day reality. Nobody built their capability. Nobody gave them the knowledge, the mandate, or the support to drive change within their teams. And without that, no AI rollout survives contact with the actual complexity of a real business.
This is not speculation. Harvard Business Publishing Corporate Learning's 2025 research paper, Succeeding in the Digital Age: Why AI-First Leadership Is Essential, identifies midlevel leaders as "essential drivers of AI transformation" — and reveals a significant gap between what organizations expect from them and what they are actually equipped to do.
The Expectation Gap
In a 2024 survey of senior leaders conducted by Harvard Business Publishing Corporate Learning, 81% reported having significantly greater expectations of midlevel leaders to lead the adoption of digital tools and technologies this year compared with last year. (references below)
Read that again. Eight in ten senior leaders are now expecting meaningfully more from the middle of their organizations on AI — and that expectation increased year over year. The pressure on midlevel leaders to drive AI adoption is real, growing, and coming from the top.
Now read this: the same survey found that only 48% of midlevel leaders feel their organizations effectively utilize their creativity and ingenuity to drive transformation efforts. Among senior leaders, only 60% felt the same about how their organizations use midlevel talent for transformation. (references below)
This is the expectation gap in one number: expectations went up 81 percentage points in urgency, but less than half of the people those expectations are directed at feel they are being effectively deployed. You cannot close an AI capability gap by raising expectations without raising support.
Why the Middle Is Where It Actually Happens
The reason midlevel leaders matter so much for AI transformation is structural. They sit at the intersection of strategy and execution — they understand the organization's goals well enough to connect AI initiatives to business objectives, and they are close enough to daily operations to identify where AI can actually create value.
As the Harvard paper puts it: midlevel leaders "are close enough to the business that they can spot opportunities for AI-driven efficiencies that may not be visible to others — if they have the knowledge, mindset, and skills to know what to look for."
That conditional matters. The visibility is structural. The capability is not automatic. A midlevel leader who does not understand what AI agents can do, who has never run an AI experiment, who does not have permission to pilot something without a committee sign-off — that person cannot be the change agent you need them to be, regardless of how much you expect of them.
The organizations that get AI transformation right do not rely on the enthusiasm of a few early adopters on the frontline. They build the capability of the people in the middle to actively drive adoption — to identify opportunities, to design experiments, to connect their teams' day-to-day work with the strategic direction from above.
What the Data Says About Transformation Success
This is not just organizational theory. Harvard Business Review published research in 2024 analyzing more than 300 large companies that attempted major transformations. (references below) The finding was clear: among the organizations that succeeded, "driving change from the middle out" was identified as a critical factor contributing to enduring results.
Top-down transformation — announce a strategy, expect execution, measure results — fails at scale because it treats the organization as a machine that executes instructions. But the people who actually implement AI are making hundreds of small decisions every day about when to use it, how to use it, what to flag as a problem, what to try next. Those decisions require judgment. Judgment requires capability. Capability requires investment.
The organizations that built AI capability in their middle layers — giving midlevel leaders the knowledge, the mandate, and the runway to run experiments — produced transformations that compounded. The ones that treated AI adoption as a policy rather than a capability-building exercise produced the familiar pattern: early enthusiasm, uneven adoption, stalled momentum.
The Three Things Middle Managers Need to Drive AI Transformation
Based on the Harvard framework, there are three distinct requirements for midlevel leaders to be effective AI transformation drivers. Organizations typically underfund all three.
1. Foundational AI knowledge. This means more than watching a demo. It means understanding how AI systems work well enough to have an informed opinion about what they can and cannot do in your specific operational context. What data does it need? What decisions is it actually making? Where does it fail, and under what conditions? A leader who cannot answer these questions cannot responsibly oversee AI in their team — and certainly cannot identify new opportunities for it.
2. An AI-first mindset. The Harvard paper identifies the shift from cautious observer to active experimenter as a distinct developmental stage — not something that happens automatically when tools become available. It requires letting go of limiting beliefs about AI, building familiarity through use, and developing the ability to tolerate the ambiguity of experimenting with something that does not always produce consistent outputs. Organizations that create safe conditions for experimentation get this shift faster. Organizations that only approve AI use after it has proven itself elsewhere never get it at all.
3. Practical AI skills and a mandate to use them. Midlevel leaders need to be able to pilot projects, troubleshoot when AI outputs are wrong, guide their teams through AI-enabled processes, and scale what works. This is a skillset, not a posture. It develops through doing, through feedback, and through having enough organizational authority to actually make changes rather than just recommend them. The 48% utilization figure is partly a capability gap and partly a permission gap — leaders who do not feel empowered to act will not act, regardless of how much they know.
What This Looks Like for a Small Business
In a company with five to fifty employees, the "midlevel leader" dynamic looks different than in a large enterprise — but the problem is identical. The business owner sets the direction. The frontline employees handle the work. And whoever sits between those two layers — the operations manager, the team lead, the department head — is the person who either translates AI strategy into operational reality or watches it evaporate into good intentions.
If that person does not understand what AI can do, they will not champion it. If they do not have the authority to change a process without approval, they will not experiment. If they are evaluated on throughput and punished for disruption, they will not prioritize transformation work even if they want to.
The practical steps are not complicated:
- Identify the one or two people in your organization who sit between strategy and execution. Invest specifically in their AI knowledge — not a generic AI overview, but a grounded understanding of the specific tools and workflows relevant to your business.
- Give them an explicit mandate. "I want you to find one process in your area that we can improve with AI and build a working version in the next 90 days" is a concrete directive. "I want us to use more AI" is not.
- Protect experimentation time. AI capability-building competes with operational throughput. If you do not carve out time for it explicitly, throughput will win every time, and the transformation never starts.
- Evaluate and reward the outcome, not just the process. If the experiment works, make it visible. If it does not, treat it as useful information rather than a failure. The signal you send about experimentation determines whether you get more of it.
The Cost of the Expectation Gap
The 81%/48% gap is not just a management problem. It is a competitive problem. Organizations where senior leaders expect AI transformation but do not invest in the people who have to execute it are in the worst possible position: they have the liability of being a laggard (falling behind on AI capability) and the vulnerability of false confidence (believing they are AI-first because they bought the tools).
The organizations pulling ahead are not the ones where the CEO read the most about AI. They are the ones where the people responsible for day-to-day execution — the team leads, the operations managers, the department heads — have the knowledge, the mindset, the skills, and the mandate to drive transformation from the inside out.
That is what the Harvard research describes as "driving change from the middle out." And the data from 300+ companies is clear: it is the difference between transformations that endure and transformations that stall.