A few weeks ago, one of our teams used AI to turn around a technical proposal to a client in a matter of days, work that used to take weeks. That's fluency: operating the tool comfortably enough to produce accurate, well-organized content at speed, without friction. Proficiency showed up in what came next.
Someone on the team caught that AI had answered one of the evaluation criteria with solid, generic language, technically correct, but missing the specific certification and track record the client was screening for. They rewrote that one section and left the other pages alone. Knowing which section still needed a human hand, and which didn't, is proficiency.
Speed matters, and AI makes most things faster. Getting something done and getting it done well are two different achievements. Fluency gets you the first: comfortable, frictionless use of the tool. Proficiency gets you the second, and it's worth being precise about both words, because they get used interchangeably far more than they should.
Three different things wearing the same name
Use is the easiest thing to measure and the least meaningful: seats activated, queries logged and a license being opened. It tells you adoption is happening. It doesn't tell you much else.
Fluency is a step up from that: operating AI without friction. The prompts, the features, the shortcuts stop being something you think about and start being something you just do. It's real, it matters, and most organizations, including ours, have gotten a lot better at building it.
Proficiency, the kind that changes outcomes for clients, is something else again. We think of it on a scale, beginner to intermediate to expert, and it's the layer where judgment lives: knowing how much of the work to still shape yourself, even when AI could hand you a finished answer. It's the instinct to question an answer that appeared in 10 seconds rather than accept it just because it appeared at all. Expert proficiency is that judgment applied at scale: catching the shallow answer yourself, and teaching someone else to catch it too.
When I use the word proficiency in this piece, that's what I mean: judgment, distinct from simply using the tool or being comfortable operating it. Use, fluency and proficiency get lumped together in most conversations, but they develop differently, get measured differently, and require a different kind of investment from leaders.
Why the checkbox falls short
Most of that investment goes toward the two you can put a number on.
A lot of well-intentioned AI enablement work, ours included, has leaned on adoption dashboards, completions, certifications and badges as the proxy for readiness. Those are useful signals of use and fluency. Proficiency, the judgment layer, is a different question entirely.
Someone can pass every module in a program and still hand a client a flawed AI-drafted recommendation, because the training taught them how to generate an answer and never asked them to defend one. The miss sits in what we chose to measure. The person did exactly what the training asked.
It shows up as a trust problem, too. Associates who are fluent with a tool's outputs but haven't built judgment about its reasoning tend to fall into one of two camps: leaning on it without question or avoiding it out of caution. Both are expensive, and neither is proficiency.
What building real proficiency requires
Proficiency develops the way most critical thinking skills do: through repeated use, real feedback and progressively harder problems, not a single course with an end date.
That's also why the pace of change in the tools themselves matters less than people assume. Someone who has built proficiency, the judgment underneath the tool, picks up the next interface in days. Someone who has only built fluency with the current one has to start over every time the tool changes.
The practical implication is that proficiency-building needs structure without needing uniformity. Not everyone has to become an expert. Staged paths, beginner to intermediate to expert, where people can go as far as their role requires, work better than a single training track applied to everyone. That's the structure behind our own learning pathways and role-based standards for AI tools.
The operating model that scales it
Individual proficiency doesn't translate into organizational capability on its own. Three things must be in place for it to compound.
Trust must be built as infrastructure, not added as a compliance step. It's what keeps people out of the two failure modes already mentioned, over-relying on AI or avoiding it, and lets them move with confidence rather than hesitating at every decision. Done well, governance is what makes speed possible, not what slows it down.
The skill must be visible, though visible doesn't mean scored the way a certification is. An organization cannot develop a capability it cannot see, and that includes a manager's read on where someone sits on the proficiency scale. That read has to inform staffing decisions, next roles, and opportunities. That's what turns individual learning into something the business can use.
And culture has to carry the part that no platform can. Reward curiosity and the willingness to keep learning as highly as tenure or the ability to describe a process well. Clients feel the difference in the judgment calls in addition to the deliverables.
What this requires of leaders
Three questions worth putting on your own leadership team:
- When you look at your AI training data, are you measuring use and fluency, or proficiency?
- Where in your organization would someone feel safe telling you an AI-generated answer was wrong?
- Is there a next step, role, project, or stretch assignment tied to a skill someone just built, or does the badge sit alone?
The answers will tell you whether you're building fluency or proficiency.
Proficiency, especially at the expert end, is harder to put a number on than fluency is, which is part of why usage and completions end up on most dashboards. That judgement often comes from experience, but not always, and the exceptions are where trust, visibility and culture come into play.
Judgment is built through doing, not a training course. It gets passed down by working alongside someone who already has it. But that knowledge transfer travels in both directions. Someone earlier in their career often moves faster with new tools. Someone further along is more likely to catch when the answer is wrong. Pairing them on real AI-assisted work, regardless of tenure, builds fluency and proficiency faster than another training module would. Our own approach has been to invest in AI to increase everyone's capacity.
Ready to build a workforce with that kind of proficiency? Contact us to learn how Unisys helps organizations turn AI fluency into expert proficiency, at scale.
