Somewhere in your business, a manager is trying to answer questions you should have answered before the AI subscription was approved. Can I put this customer information into the tool? How do I assess a draft when the employee used AI to create it? If a task now takes twenty minutes instead of an hour, do I give that time back, ask for more output or change the role?

If that manager is improvising, your business does not have an AI strategy. It has a collection of licences and a person in the middle absorbing the uncertainty.

That person matters because the manager is where grand announcements meet ordinary work. Founders can speak confidently about efficiency, growth and innovation. Employees experience the change as a new process on a Tuesday morning, a confusing expectation from a client, or an anxiety that the work they were hired to do is quietly being redesigned. The manager is expected to translate both languages. Leaving them unsupported is not delegation. It is abdication.

Curiosity is useful. It is not enough.

There is a reasonable instinct behind letting people explore. Teams often find practical uses that a founder would never have spotted. The danger is confusing early experimentation with a leadership model.

Gartner reported in 2026 that 46% of managers were experimenting with AI to improve their work, compared with 26% of employees. Yet only 14% of managers said they faced no challenges in driving effective AI use across their team. That is the gap in one set of numbers: the people nearest the work are already testing the tools, but most are not confident they can lead the change well.

For a small business, this is actually an advantage if you use it. You are close enough to the work to create a simple, usable system before a dozen unofficial workarounds harden into culture. Do not begin with a 40-page policy. Begin with a manager toolbox that answers the questions that recur.

Give every manager a few non-negotiables

First, define the boundaries. Which tools are approved? What types of data cannot be entered? When must a human review the output? What does good disclosure look like when AI contributes to customer-facing work? Your legal, privacy and industry obligations may differ, so get specialist advice where needed. But do not make the manager invent the baseline every time a new tool appears.

Second, define the goal for each team. “Use AI more” is not a goal. A good goal sounds like: reduce the time spent preparing routine project updates without lowering accuracy; free customer-support staff to resolve complex cases; improve the first draft of a proposal while keeping the account lead responsible for the final recommendation.

Third, say what will not be measured. If you want careful, responsible adoption, do not reward people simply for the largest number of prompts or the fastest first draft. You will get lots of activity and very little judgement.

The manager then has a stable message: these are the boundaries, this is the point of the experiment, and this is how we will know whether it helped.

Do not make managers sell a future you have not explained

Gartner’s research makes a sharper point: managers must balance maintaining trust with translating the benefits they observe into the business case leaders want. That is a demanding task. It becomes impossible when leaders have not been honest about the trade-offs themselves.

You cannot ask a manager to reassure a nervous employee that AI is there to augment their work if, privately, you are using the trial to identify who can be removed from payroll. You may eventually need to redesign a role. Businesses sometimes have to make decisions that hurt. But then lead that decision directly, with the appropriate employment and HR advice, rather than turning managers into cheerful messengers for an outcome they cannot explain.

Trust does not require pretending uncertainty has vanished. It requires saying what you know, what you are testing and when people will hear more. A manager who can say “we are piloting this for six weeks; no individual performance decisions will be based on the trial; here is what we are looking for” has something real to offer their team.

Decide where saved time goes before it becomes a fight

This is the conversation leaders most often postpone. Gartner found that only 7% of organisations provided guidance on how to use time saved by AI. HR leaders and managers also disagreed about where a hypothetical freed-up hour should go: 55% of HR leaders favoured special projects outside the core job, compared with 28% of managers.

Those figures describe the classic problem of every productivity tool. Leadership sees capacity. Employees see a chance to reduce the overtime that has become normal. Managers see the unglamorous work that has been starved of attention. None of them is necessarily wrong. But they will collide if the founder declares a gain and leaves the meaning of that gain unresolved.

Give managers permission to have a practical conversation role by role. What should the freed time buy us? Better client preparation? Faster turnaround? More training? A cleaner process? A reasonable workload? The answer may differ across functions. What should not differ is the principle that a worker should not be punished for becoming more efficient.

Create a weekly learning loop

AI adoption does not need a steering committee for every small decision. It does need a rhythm. In a 20-minute weekly manager check-in, ask four questions: What did the team try? What produced a measurable improvement? What created risk or poor quality? What needs a decision from the founder?

Capture the answers in a shared one-page log. Over a month, you will have something more valuable than impressive anecdotes: an internal map of use cases that work, the conditions they need and the places human judgement remains essential. Share the lessons across managers so one team does not repeatedly learn what another team has already solved.

This also makes it easier to stop a bad use case without anyone losing face. A trial that produces weak outputs or creates customer risk has done its job by showing you where not to invest further.

The manager is not a support ticket

Your managers should not be the people who clean up after a technology decision. They should be partners in shaping how it changes work. That means involving them early, giving them clear guardrails, and expecting them to bring back inconvenient information—not merely upbeat adoption stories.

The founder’s role is not to have every answer. It is to make sure the people leading others have enough clarity to lead honestly. Buy the tool if it earns its place. But do not call it strategy until your managers can explain what the team is doing, why it matters, and how people will be treated while they learn.

Sources

• Gartner: Managers and Effective AI Use