Most people use AI. Almost no one uses it well. A 2026 field report on the gap between adoption and proficiency, and how to close it.
About 76% of employees now use AI at work, yet only around 13% use it for real, value-generating work, and just 29% of organizations see meaningful ROI. The tools arrived; the skill did not. That gap, between using AI and using it well, is the whole opportunity, and it is still wide open.
Adoption looks healthy from the top. Underneath, the same pattern repeats: AI is everywhere, and almost nowhere is it being used to its potential.
Adoption is not the problem. Proficiency is. Sort the workforce by how well they actually use AI and it collapses into four tiers, with almost everyone stuck in the shallow end.
Roughly 85% of people have beginner-level use cases or none at all. Only about 15% are running the kind of use cases likely to drive real ROI. The tools are everywhere; the imagination for them is not.
The top use cases skew to search replacement, drafting, and editing, useful, but marginal. They make a task faster without changing what the task is worth. Value comes from redesigning the work, not speeding up the old version of it.
Ask people how much time AI saves them each week and the distribution is sobering. Nearly half save under two hours; only one in seven saves eight hours or more.
Organizations are spending: 63% have an AI policy, 52% have rolled out tools, 44% offer training. Yet even staff who have been trained score, on average, 40 out of 100 on proficiency. Inputs are not outcomes.
On almost every measure of AI maturity, the C-suite rates the organization 26 to 38 points higher than the individual contributors doing the work. Leadership is grading a company it does not actually operate.
The people closest to the work have the least access: fewer tools, less training, no reimbursement. The layer with the most use cases is the layer the investment reaches last.
Technology, financial services, and consulting lead on proficiency and ROI. Healthcare, education, and retail lag, held back by regulation, legacy systems, and thin margins for experimentation.
Engineering, data, and marketing are ahead. Most other functions trail, and even the leaders routinely skip their single most obvious use case. The gap is not knowledge of AI; it is the discipline to apply it where it counts.
All 15 pages, the way they are designed. Page through it here, or take the PDF with you.
AI proficiency is the ability to use AI for real, value-generating work, not just to draft an email or replace a search. It matters because adoption is already solved (about 76% of employees use AI), so the gap between people who use AI and people who use it well is now the entire competitive advantage.
About 76% of employees use AI at work in some form, and over a billion people use ChatGPT every week. But roughly 87% use it at a beginner level or not at all, and only about 13% use it for 30% or more of their daily work.
Only about 29% of organizations report meaningful ROI from AI because they invest in inputs (policies, tools, training) rather than outcomes. Most use cases speed up old tasks instead of redesigning the work, so the value never compounds.
Roughly two thirds of users save under four hours per week, and nearly a quarter save no time at all. The time savings are real but marginal, because most people apply AI to low-value tasks.
The Blank Collar framework reads as Vision plus Data, times Process over Human Experience, raised to the power of AI. It says: minimise the process you hand to machines, maximise the human experience only people can own, and let AI multiply the result.
The State of AI at Work is a 2026 field report by Kristian Kabashi, founder of The Blank Collar. He coined the term "the Blank Collar" in 2016 and founded the practice in 2018; he is based in Zürich, Switzerland.
Take the field report with you, then put the framework to work. The thinking continues in your inbox, link in the footer.
Download the PDF ↓