Key Takeaways
- AI proficiency is demonstrated skill, measured in the work people produce: Knowing how AI works is literacy, and logging in is adoption; proficiency shows up in the output.
- Most organizations have adoption without proficiency: EY's 2025 Work Reimagined Survey found 88% of employees use AI at work, while only 5% use it in advanced ways that change how they work.
- Proficiency has a ceiling when everyone uses the same tools: As teams get better at prompting the same models, their work tends to converge on what common AI would produce anyway.
- The useful measure is what a person adds beyond common AI: That is the gap Original Intelligence measurement is built to show, at the level of people and of whole teams.
Companies have spent the last two years getting AI into people's hands. EY's 2025 Work Reimagined Survey, which covered 15,000 employees and 1,500 employers across 29 countries, found that 88% of employees now use AI at work. Most of that use is basic, with search and document summaries leading the list. EY estimates that companies are missing up to 40% of the productivity gains AI could deliver, and it ties that gap to talent strategy rather than to the technology.
That gap has a name in the market now. AI proficiency is the emerging term for how well people use AI on the work that matters, and it's quickly replacing adoption as the number leaders want to see.
What Is AI Proficiency?
AI proficiency is a person's demonstrated ability to use AI tools to produce better work than they could without them. It covers knowing which tool fits a task, how to direct it over several steps, when to check or discard what it returns, and when to leave AI out entirely. The word that matters in that definition is demonstrated. Proficiency is visible in output, so it cannot be inferred from a training certificate or a license count.
It also sits on a spectrum. At the low end, people use a chatbot as a faster search engine. Further along, they automate whole tasks, chain tools together, and hand multi-step work to agents. Most frameworks describe four or five levels between those points, and most workforces today cluster near the bottom.
Proficiency, Literacy and Adoption Measure Different Things
The three terms get used interchangeably, but they answer different questions. Separating them makes it easier to see which one a program is improving.
Our guide to what AI literacy means covers the understanding layer in depth. Proficiency picks up where that leaves off, at the point where knowledge has to show up in finished work.
Proficiency Plateaus When Everyone Uses the Same Models
Proficiency programs tend to assume the curve keeps rising. The research suggests it flattens in a specific way. In a 2024 study published in Science Advances, Anil Doshi and Oliver Hauser asked 293 writers to produce short stories, some with access to AI-generated ideas. The AI-assisted stories were rated more creative and better written, yet they were also more similar to one another than the stories written without AI.
That is the proficiency ceiling. Each person gets better, and the group's work converges.
Hupside’s co-founders have studied this effect directly, and their findings have shaped how the company measures originality. Dan Johnson, PhD, a cognitive scientist, and Adam Green, PhD, a psychology professor and expert on human creativity, studied more than 160,000 U.S. college admissions essays in research published in Information Systems Research. Across a sample of 2,200, essays written without AI produced up to eight times more novel ideas than AI-assisted ones. That gap between polished output and original thinking is the reason Hupside was founded to measure.
For an employer, this is where signal collapse sets in. When every proficient employee produces polished work from the same models, polish stops telling you who is adding value. Two people can score identically on any proficiency rubric and contribute very different things.
How to Measure AI Proficiency
Measurement works best in two layers. The first is behavioral, and it tracks how people use AI: the range of tasks they apply it to, how well they direct and check it, and whether the habit holds up over months rather than weeks. Self-reported surveys are a weak source here, because people tend to rate themselves higher than their work shows.
The second layer measures the output against common AI, which is the floor nearly everyone starts from now, because the same models that make a weak writer competent also hand every one of that writer's colleagues the same competent draft to start from. This layer asks how much a person's work adds beyond what the same models would produce for anyone who asked. It is the layer that separates two equally proficient people, and it is the one most AI proficiency frameworks leave out, because their benchmarks stop at tool skill. Hupside's Hupchecker measures this at the individual level through an OIQ score and OIQ Type, and Hupmapper shows Originality Intensity across a platform or organization.
Building a Program That Raises the Ceiling
A proficiency program works best when it treats the two layers as one system. Start by baselining where people sit on both, then set goals for each. A team that uses AI widely but produces converging work needs a different intervention than a team that has barely started, and a single training track cannot serve both.
From there, the practical steps are familiar. Find the people already working at the upper levels and let them teach the methods they've built for their own roles. Replace one-time training with short, recurring sessions tied to new tools and new use cases. Give people room to experiment without penalty, since that is how the upper levels get reached. Our guides to AI upskilling and change management for AI adoption cover the program design in more detail.
Then measure again on a regular cycle, because the tools move monthly and a benchmark set in January will not describe the work being done in June.
Measure What Matters With Hupside
Proficiency tells you that your people can use AI well. Original Intelligence tells you what they're adding with it, which is the part competitors can't copy by buying the same licenses. Hupside gives organizations a way to see that contribution across individuals and teams, so the AI investment shows up as better, more distinctive work.
To see how much your organization's work adds beyond common AI, explore Hupmapper for free.





