Rolling out an AI tool is one milestone. Changing how a team works with it is another.
An organization may have completed pilots, provided employees with AI assistants, and introduced training, yet still struggle to see consistent changes in delivery. A tool can be available across the business without becoming part of the decisions, reviews, and handoffs that shape everyday work.
This is an important distinction in enterprise AI adoption. A Wharton review of research on organizational AI adoption distinguishes implementing AI from actually using it. The review examines influences at the individual, workflow, and organizational levels.
Middle managers work at the intersection of those levels. They are responsible for delivery today while being asked to change how delivery happens tomorrow.
The challenge is not simply whether employees have access to AI. It is whether teams have the conditions to use it responsibly in their work.
Executive leaders can set priorities, approve platforms, and establish policies. They cannot anticipate every decision involved in using AI during a claims review, an engineering release, or a customer-service interaction.
Those decisions happen closer to the work. A team manager may need to decide which tasks are suitable for AI assistance, what information can be shared with a tool, when a specialist should review an output, and how exceptions will be handled.
These are not minor implementation details. They determine whether AI becomes a useful part of a workflow or remains an optional tool employees use inconsistently.
There is evidence that the local environment matters. In a 2026 Gallup study of 23,717 US employees, among those whose organizations made AI available, 78% of employees who strongly agreed that their manager supported AI use reported using it a few times a week or more. Among employees who did not strongly agree, the figure was 44%.
That difference is an association, not proof that manager support alone causes greater use. The suitability of the tools and workflows may influence both. Even so, the findings draw attention to something a deployment dashboard cannot fully capture: what employees experience within their teams.
Middle managers are often expected to introduce new ways of working without stepping away from existing delivery commitments.
AI may reduce effort in some tasks, but its introduction can create additional work elsewhere. Outputs need to be checked. Exceptions need to be resolved. Employees need guidance about appropriate use. Managers remain accountable for the quality of work delivered, including work that AI has helped produce.
A 2026 McKinsey analysis of agentic AI adoption discusses the difficulty of translating leadership ambitions into operational practice while other organizational changes are underway. That challenge becomes especially relevant when managers are given responsibility for adoption but little room to experiment with the tools themselves.
Workforce concerns can shape the response as well. In an Institute for Fiscal Studies working paper, a survey experiment involving managers in the US and UK found that information emphasizing AI’s potential to displace workers reduced their stated intentions to adopt and advocate for AI. The study measured intentions in an experimental setting, not actual workplace deployment decisions.
These findings do not establish that managers are the main obstacle to adoption. They suggest a more useful question: Are organizations changing managerial responsibilities faster than they are changing the support around those responsibilities?
When adoption is added to an unchanged workload, the problem may be less about resistance and more about competing expectations.
A manager who uses an AI assistant effectively for personal tasks is not necessarily prepared to introduce AI into a team’s operating processes.
Individual proficiency might involve drafting, summarizing, or analyzing information. Managing AI-enabled work involves a different set of judgments: identifying suitable use cases, assessing operational risks, setting review standards, and understanding what should remain under human control.
Consider a software engineering team using AI to generate tests or explain unfamiliar code. The manager’s responsibility is not simply to encourage usage. It is to ensure the team has standards for reviewing generated work, protecting sensitive information, and identifying cases where the output may be incomplete or wrong.
The same principle applies outside engineering. In a customer-facing process, a faster draft may still require checks for accuracy, privacy, and policy compliance. In an internal operations team, automating one step may have little effect if approvals or downstream handoffs remain unchanged.
Managers therefore need AI fluency grounded in their own workflows, not only general awareness of tools. They also need the confidence to coach employees, respond to concerns, and explain how the definition of good work is changing.
AI adoption is frequently discussed in terms of licenses, active users, and frequency of use. These measures have value. They show whether tools are accessible and whether people are trying them.
They do not, on their own, show whether work is improving.
A team might use AI regularly without reducing turnaround time. Another might use it selectively in a high-friction process and see a meaningful improvement. The relevant measures depend on the workflow and the outcome the organization wants to achieve.
For some teams, that may mean examining rework, error rates, or the quality of completed outputs. For others, it may mean looking at time spent on specific activities, the reliability of handoffs, or the experience of employees and customers.
AI usage tells leaders whether a tool is being used. Workflow measures help them understand whether the work is changing.
Even then, care is needed. If delivery performance improves after AI is introduced, the improvement cannot automatically be attributed to AI. Changes in staffing, processes, project complexity, and other tools may also contribute. Baselines and contextual evidence make the assessment more useful.
A practical approach begins with a specific operational problem, rather than a broad instruction to increase AI usage.
Take a workflow where teams repeatedly spend time preparing information for review. Before introducing an AI tool, the organization can examine where delays occur, which tasks are repetitive, what quality standards apply, and where human judgment is essential.
Managers who own the workflow should be involved in those decisions. They understand the exceptions, dependencies, and informal practices that may not appear in a process diagram. Their involvement also helps identify where a proposed improvement could introduce new review work or shift a bottleneck elsewhere.
Preparation should then reflect the decisions managers will actually face. That may involve working through realistic scenarios, establishing clear approval criteria, practising how to assess AI-generated outputs, and agreeing on what evidence would justify broader use.
It is equally important to revisit workload and accountability. Managers need time to test new approaches and clarity about who owns decisions when AI is involved. Asking them to lead adoption while judging them only against unchanged short-term delivery targets can create conflicting incentives.
This workflow-oriented approach is also relevant to GenSpark’s AI Adoption and Enablement offering, which covers opportunity discovery, use-case design, applied enablement, and performance measurement. The underlying principle is to connect capability development to the work people are expected to perform.
Technology integration, security, governance, data quality, and executive sponsorship all remain important to enterprise AI adoption. Middle managers are not a substitute for any of them.
But managers often sit at the point where an enterprise strategy must become a repeatable way of working. They help determine what teams do with AI, how outputs are checked, and how new practices fit alongside existing responsibilities.
For CIOs, CHROs, and transformation leaders, that makes managerial readiness a design consideration, not simply a communication task. What are managers expected to do differently? Do they have the skills, authority, and time to do it? And how will the organization know whether the change is helping?
The answers will vary across teams and industries. Asking those questions early, however, gives organizations a better chance of moving from AI availability to responsible, useful adoption.