I was talking recently with my AZ Small Business Program mentor who also happens to be an R&D scientist.
Our conversation turned to AI and, inevitably, AI slop: the increasingly familiar flood of polished-looking, AI-generated content that may be faster to produce but isn’t necessarily making the work any better. Terrible in the research setting which relies on truth, accuracy, and reliability.
She described what many organizations are doing right now:
- Leadership has made a significant investment in AI.
- Employees are expected to complete AI certifications.
- Adoption is encouraged and usage is tracked. Even the number of hours employees spend using AI can become a measure of "progress."
Standard behavior reinforcement. On paper, it makes sense. If you've invested in the technology, you want people using it.
But she was noticing something far more fascinating (at least to me!).
The best scientists weren't just using AI more. They were using it differently.
I'd seen this before. When we studied global sellers and sales managers at Microsoft, we identified 11 distinct ways top performers were using Copilot.
The pattern was strikingly similar: same technology, different behavior, different results.
AI isn't eliminating the performance gap. It may be exposing it.
Imagine two employees who get Copilot, ChatGPT, Claude, or another AI tool on Monday.
Top performer:
“I know what good looks like. I know the outcome I'm trying to achieve. I know what matters, what doesn't, and where my judgment is required. Let me use AI to get there faster.”
So, when they open Copilot, ChatGPT, Claude or another AI tool, they're not asking AI to tell them what good looks like.
They already know. They might use it to pressure-test an idea, synthesize information, challenge their thinking, accelerate a first draft, explore alternatives or eliminate low-value work.
They're still driving. They use AI to get there faster. AI becomes leverage.
Average performer:
“Which AI should I be using? Maybe Claude is better. Let me try this prompt. Hmm. Maybe ChatGPT. I should build an agent. Wait, someone said NotebookLM/Gemini Notebook. Let me take another AI course…”
They don't necessarily have the same clarity about the desired outcome or the judgment required to produce it as the top performer. So, they start experimenting. Maybe ChatGPT would be better for this. Or Claude. Or Copilot. Maybe I need a better prompt. Maybe I should build an agent. Maybe I need another AI course.
AI becomes another place to work. Before long, they've become extraordinarily busy using AI without necessarily getting better at the work.
The technology is the same. The performance isn't. And suddenly you've recreated the performance gap inside the technology.
This is where I think organizations need to make an important distinction.
There are at least three levels of AI maturity that require leaders' attention:
Are people using it? Certifications. Licenses. Log-ins. Hours of usage. Useful information. But it primarily tells us if people have adopted the technology.
Are people doing things with it? Prompts. Drafts. Summaries. Analyses. Agents. More output, produced faster. Again, potentially valuable. But more activity isn't necessarily better performance.
Is AI helping people perform the work better? Better decisions. Stronger judgment. Higher-quality customer interactions. Less rework. Faster execution where speed matters. Better business outcomes. That's an entirely different (and more difficult) measurement. My point? An organization could conceivably be very successful at AI adoption and AI activity without ever reaching AI-enabled performance.
There is an enormous opportunity sitting inside organizations right now.
Most companies know who their strongest performers are. Instead of asking those employees simply whether they're using AI, I'd be much more interested in studying how they're using it.
Put real work in front of them and watch:
- What do they give to AI?
- What don't they give to AI?
- What context do they provide before asking for an output?
- How do they frame the problem?
- How do they recognize a mediocre answer?
- When do they challenge the AI?
- When do they ignore it altogether?
- Where does their own expertise and judgment enter the process?
And, most importantly: What are your best performers doing with AI that your average performers aren't?
Because the organization's next great AI training program may not need to begin with another course on prompting. It may need to begin by studying its own people.
I've spent much of my career studying a fairly simple question: What do the best people do differently?
Not what a competency model says they should do. Not what industry benchmarks say great performers typically do. What the people producing exceptional results inside a particular organization actually do differently when they're making decisions, navigating ambiguity, working with customers and getting results.
And as I consistently write, AI makes that question more important, not less. Because if we understand those distinctions, we have something incredibly powerful to build into AI.
- We can give AI examples of exemplary work.
- We can expose the decision rules top performers use.
- We can capture the questions they ask.
- We can identify what they notice that others overlook.
- We can understand where they use judgment instead of blindly following a process.
And then AI stops being simply a productivity tool and is a way to help scale excellence.
That's a very different ambition from getting everyone certified...and it's also a much harder metric to put on a dashboard. But it's the one at least I'd want to know.
AI doesn't know what excellence looks like in your organization unless you do.
Stop measuring AI adoption and start studying AI-enabled excellence.