A leadership team has invested in AI.
Some employees are producing remarkable results.
Others have tried the same tools and quietly returned to their old ways of working.
Leadership begins asking familiar questions.
Do we need better AI training?
Should we standardize prompts?
Do we need different tools?
Would stricter policies improve consistency?
Those are all reasonable questions.
They just aren’t where I’d begin.
At this point, I probably wouldn’t have recommendations.
I’d have questions.
Quite a few of them.
I’d want to understand what work people are actually trying to accomplish with AI.
Where is it genuinely helping?
Where is it creating more work than it saves?
Where are people using it regularly?
Where have they tried it and stopped?
And where are people producing results that others haven’t been able to replicate?
Because uneven AI adoption may not simply be a technology or training problem.
It may be telling us something about the relationship between the work, the people doing it, the technology, and the organization around them.
Suppose several employees are consistently producing outstanding work with AI.
I’d want to understand what makes that possible.
Perhaps they understand the underlying work unusually well.
Perhaps they know which questions to ask before they ever write a prompt.
Perhaps they have enough expertise to recognize weak outputs and refine them.
Perhaps they know when AI is useful — and when it isn’t.
Perhaps they have access to organizational context or information that others don’t.
Perhaps they’ve developed workflows around AI that nobody has documented.
Perhaps their work simply lends itself particularly well to AI.
The difference may not be the software.
It may be the conditions surrounding its use.
That raises another question.
Can those conditions be understood?
Not copied mechanically, but understood well enough to see what the organization can learn from them.
What knowledge does someone need to use AI well in this kind of work?
What judgment are they exercising?
What information does the AI need?
Where does human expertise remain essential?
How are outputs checked?
What happens when the AI gets something wrong?
And which parts of this way of working are actually transferable to someone else?
Those are the things I’d want to understand.
Because some of what makes AI useful can be taught, shared, or designed into the way work happens.
At the same time, I’d become curious about the places where AI isn’t being used.
Not because everyone should be using it.
But because non-adoption is information too.
Are people struggling with the tools themselves?
Are they uncertain about when to trust the output?
Does using AI create more verification work than it eliminates?
Does their work depend on context the available systems don’t have?
Are there privacy, security, or accuracy concerns?
Have they experimented and discovered failure modes leadership hasn’t noticed?
Or does AI simply not improve this particular work?
What appears to be resistance may sometimes be uncertainty.
But sometimes it may be judgment.
And I’d want to understand the difference before trying to correct either one.
Only after developing a clearer picture would I begin suggesting experiments.
Perhaps some of the people getting strong results become mentors.
Perhaps teams begin sharing not only successful outputs, but how they arrived there.
Perhaps useful AI workflows are documented — not as rigid procedures, but as examples that make the underlying decisions visible.
Perhaps training shifts away from simply learning features or prompts toward developing judgment.
Perhaps a workflow is redesigned so AI handles a clearly defined part of the work while people retain responsibility for the decisions that require context or expertise.
Perhaps the organization tests where AI adds genuine value before expanding its use more broadly.
And as those experiments develop, I’d want to pay attention to what fails as carefully as to what succeeds.
What requires human review?
What information should the system be allowed to access?
What should it be allowed to do?
Where does responsibility remain?
What happens when something goes wrong?
The goal isn’t simply to increase AI adoption.
It’s to understand where AI genuinely improves the organization’s ability to do its work — and under what conditions.
Every organization is unique.
The situation you’ve just read isn’t a template.
It’s an example of how I approach complexity.
My goal isn’t to arrive with predetermined answers.
It’s to help your organization better understand itself before deciding what comes next.