Every organization that has rolled out enterprise AI has a few people who have rebuilt half their workflow around it. They’re the ones colleagues ask when a prompt isn’t working, and their output looks different from everyone else’s. Leaders know these people exist, but there’s no reliable way to see who they are, what they’re doing differently, or how to make the rest of the organization work the same way.

Power users are hard to manufacture. The usual tools for building skills in an organization, such as a training program, a prompt library, or a lunch-and-learn, tend to produce a short spike in usage and then a plateau.

There’s a better path, and we have the data to prove it. 

‍

The biggest AI gap is between desks

Most AI benchmarking compares your company to other companies, or one department to another. New Workhelix research suggests that’s the wrong comparison.

We analyzed 18 months of complete AI usage logs at a global financial-services and technology company of roughly 11,000 employees, matched to its HR system. Usage was extremely concentrated. The top 10% of employees generated about two-thirds of all prompts. The median active user sent 47 prompts over the whole period, while the 99th percentile sent 2,340.

What surprised us more was where that variation lives. Knowing someone’s division, seniority or even exact job title explained less than a third of how much they used AI. About 69% of the variation was between people holding the same job.

The practical implication is that your best example of great AI usage for a given role is probably someone doing the same job, on the same team, often at the next desk.

‍

Why one-time training doesn’t close the gap

If the playbook already exists inside your teams, why doesn’t it spread on its own? Because the things that separate power users from everyone else are habits, not facts. A training session is good at transferring facts.

The data makes this concrete. The heaviest users in our research showed up on a median of 124 distinct days over the period, roughly every other working day. Casual users showed up on about 10, and they packed two-thirds of their total usage into their five busiest days. Power use is AI becoming part of an ordinary workday.

Generic training also misses the role-specific detail that makes a habit stick. A marketing analyst doesn’t need a tour of what AI can do in general. They need to see what the best marketing analyst down the hall is doing with it this week, in work that looks like their own.

‍

Power users rotate, so a one-time census goes stale

There’s a second reason static approaches fall short: power users are not a fixed group.

Of the employees in the top usage decile in the first half of 2025, only 28% were still there a year later. Measured against a fixed bar rather than a rising one, about half held their place. That’s still far from a stable elite. Almost none of the others left the company, and most were still using AI, just less, and nearly one in five later climbed back to the top. Their usage simply tracked what their work demanded at the time.

The capability pool is also much bigger than any snapshot suggests. About 18% of active users had at least one week with 50 or more prompts, but only about one in twenty sustained top-tier usage week after week.

Roughly one in five of your people has shown they can work this way, but far fewer do it consistently. Closing that gap requires coaching.

‍

What coaching does that training can’t

Effective AI coaching has three ingredients, and each maps onto what the research shows:

• It’s continuous: Power users rotate and habits form over weeks, so feedback has to arrive regularly in bite-size, actionable chunks.
• It’s personal: Each person needs to see their own usage, what they’re doing well, and one concrete thing to try next. A company-wide average doesn’t give them that.
• It’s peer-based: The most credible example is someone in the same role. People adopt a method faster when it comes from someone whose job looks like theirs.

This is also the most cost-effective lever available. The expertise is already on the payroll, coaching simply makes it visible and portable.

‍

What spreading a skill looks like in practice

At one global software company, a 1,000-person marketing organization had broad but uneven AI adoption. When usage was measured, a handful of analysts stood out. They were using AI to write SQL queries against internal data to produce better demand forecasts for product launches. Nobody expected that from a non-technical role, and most of their colleagues didn’t know SQL at all.

The organization captured the common prompt pattern behind the work. It was reviewed for quality and redistributed it as a skill to the whole marketing team. A capability held by a few people became standard practice across the role.

That’s the loop coaching runs continuously: find what’s working, confirm it’s worth spreading, and put it in front of the people who can use it. Power users also tend to multiply themselves. In our research, top-decile users were 2.4 times more likely to have built an AI agent that other employees use.

‍

Spread what works, not just what’s popular

One caution is that heavy usage is not automatically good usage. Experimental research has found that people often over-rely on AI exactly where it performs worst, so coaching needs a quality filter. The goal is to spread methods that produce better work on tasks that matter, not to make everyone copy whoever sends the most prompts.

That’s why good coaching focuses on how someone uses AI as well as how much. Do they give better context? Do they break work into stages? Are they taking on harder problems over time?

‍

Introducing AI coaching

Workhelix brings the coaching approach to everyone in your organization, every week. It reads summaries of work conversations to understand how people use AI. Then it gives each person personalized insights in three parts:

• By the numbers: number of conversations and number of tools used, tracked week over week.
• What’s working well: the clearest ways the person is already using AI effectively, with feedback that helps them give better context, break work into stages and take on harder problems.
• One from your peers: a spotlight on what the best people in their role do differently, so power users’ methods can be documented and shared across the organization.

Leaders get a continuous view of where capability is growing. Employees get specific, role-relevant coaching that they can act on the same week.

‍

The takeaway

Your AI power users aren’t a rare breed you need to recruit. They’re already on your teams, they change over time, and the pool of people who’ve shown they can work at that level is larger than any snapshot reveals. The fastest way to scale them is to stop treating their methods as private know-how. Coach everyone toward them, every week.

Be one of the first to see AI coaching in action. Get a pre-launch demo.

‍