When One Person's AI Habit Stays One Person's Habit

McKinsey surveyed the workforce on generative AI in 2024 and found 91 percent of employees reporting they use it and expressing high enthusiasm. The same study found only 13 percent of their organizations had implemented multiple AI use cases at scale. Those two numbers sitting next to each other are the whole problem.

The tool isn't the bottleneck

Gallup measured weekly AI use among US employees in 2024 and found one in ten. That figure was identical to 2023, a year when organizational AI exposure grew substantially. Employees weren't blocked by access. They weren't waiting for a better tool. Something else was keeping usage flat.

The Technology Acceptance Model literature points to two variables that predict whether someone actually uses a system over time: perceived usefulness and perceived ease of use. Not enthusiasm. Not awareness. Usefulness — specifically, whether the tool connects to tasks the person already does — and ease, meaning friction low enough that using it beats not using it. When one employee on your team uses an AI assistant every day and their colleague in the same role doesn't, the most likely explanation isn't that one person is more progressive. It's that one person figured out where the tool fits their specific work and the other person never did.

That's the problem a manager is positioned to solve. Not the organizational transformation problem. Not the governance problem. The narrower, more tractable problem of moving a team from one person's habit to eight people's habit.

Why sharing the link doesn't work

Most managers try to spread AI use by mentioning it in a meeting or forwarding a link. This produces a small spike in sign-ups and no change in weekly usage. Diffusion research distinguishes between first-time adoption and sustained implementation, and treats them as different phenomena requiring different interventions. First-time adoption needs exposure and motivation. Sustained implementation needs the tool to alter how work actually flows.

A McKinsey employee survey found workers cite workflow integration as their primary unmet need, ahead of training. That's a specific complaint: not "teach me more about AI" but "show me where it fits in the work I already do." A shared link answers neither.

The plan that follows is not a transformation program. It's three actions a manager takes over roughly six weeks to move a team from one person's successful habit to a shared one.

Step one: one session, one tool, real tasks only

Run a single 60-to-90-minute session with your team. Not a demo of AI in general. Not a survey of ten tools. Pick the one tool your early adopter already uses successfully and build the session entirely around tasks your team performs this week.

The session has one job: give everyone a moment of genuine usefulness before they leave the room. If your team writes client summaries, spend 40 minutes writing one together using the tool. If they consolidate meeting notes, consolidate actual meeting notes. The session fails if it ends with people thinking "interesting" rather than "I could use that tomorrow." Perceived usefulness isn't a feeling you can lecture someone into. It comes from doing a real task and noticing the result.

Step two: prompts tied to actual job titles, not AI in general

After the session, build a prompt library. Not a generic one. A list of 8-to-12 prompts written for the specific recurring tasks your team members perform by role.

The distinction matters. A generic prompt ("summarize this document") sits in a shared folder and gets used occasionally. A role-specific prompt ("you are reviewing a client account update; flag any items where the client's stated priority conflicts with the action items from last quarter's review") gets used every time that task comes up. The second prompt raises perceived usefulness because it meets the employee inside a task they already own. That's the mechanism the Technology Acceptance Model research identifies as the driver of sustained behavior change.

Write these prompts with your early adopter. They already know where the tool fits. Your job is to translate their intuitions into reusable language for the rest of the team.

The counterargument worth taking seriously

McKinsey's 2025 survey found 71 percent of organizations report regular generative AI use, yet only 21 percent have fundamentally redesigned any workflows to account for it. A reasonable reading of that number is that most AI "adoption" is a layer on top of existing processes, and a three-step manager-led plan adds another layer without fixing the underlying structure. The critique is that employees return to jobs whose rhythms were built before the tool existed, and no prompt library changes that.

This critique is correct about organization-level implementation. It's less correct about team-level behavior change. The Technology Acceptance Model literature doesn't require workflow redesign as a precondition for sustained individual use. It requires perceived usefulness and low friction. A manager who builds role-specific prompts tied to tasks employees already perform is doing task-level integration within the authority they hold. That's not the same as redesigning the job. It's enough to change whether someone reaches for the tool on Tuesday morning.

The 21 percent figure is a problem for the CTO. The prompt library is a problem for you.

Step three: a monthly loop with actual numbers

Once a month, spend 20 minutes with your team on two questions: which prompts got used, and which ones didn't. Pull whatever usage data the tool surfaces. Ask the people who used a prompt what changed in their output. Ask the people who didn't use one what stopped them.

This loop does two things. It surfaces social proof — when one team member describes a specific result, colleagues who haven't tried that prompt yet get a concrete reason to. And it keeps the prompt library from going stale. Tasks shift. New recurring work appears. A library built in week two becomes irrelevant by month four if nobody updates it.

Gallup's flat usage numbers from 2023 to 2024 track a period of organizational exposure without structured team-level support. A monthly feedback loop is the mechanism that keeps individual use from reverting to the same plateau.

Six weeks in, you'll know whether this worked by one observable fact: whether team members are updating the prompt library themselves, without being asked.