More companies now need their people to be fluent with AI. In BCG's AI Radar 2026, companies planned to double their AI spending this year, to about 1.7% of revenues. And since February 2025, the EU AI Act has required organizations that build or use AI systems to take measures to ensure their staff are AI literate.
The people expected to lead that shift are often not ready. In our Q3 2026 AI Transformation Pulse, six in ten of the people who lead AI adoption for others are not yet fluent themselves. The data comes from 410 people at 136 organizations, a third of them executives, who described how they use AI before starting a Lead with AI program.
So what does fluency look like? Most people use AI for quick, one-off tasks, such as a question, a summary, a brainstorm or an occasional first draft. They give it little context, rarely iterate and check the output unevenly, so AI saves them some time without reliably improving their work.
That is the level of fluency we call Chatting.
Fluency starts when AI improves the work itself. People at that level use AI routinely, direct it with clear context, iterate deliberately and check the output against the standard they hold for their own work. They stay accountable for the result, and their everyday work gets noticeably better and faster.
That is the level we call Collaborating, and it is the foundation of practical, safe AI fluency.
Two levels sit beyond Collaborating. At Integrating, people build AI into recurring workflows with human review designed in. At Transforming, they design and supervise agents that carry multi-step work forward. At the other end, AI is largely absent from the work of people who are Avoiding.
Together, these five levels make up our AI Fluency Matrix, from Level 0 (Avoiding) to Level 4 (Transforming). Each level rests on observable capability, meaning what someone can reliably do with AI.
In our data, three in four people sit at Level 1, Chatting, and only 18% have reached Level 2 or higher. (This group chose, or was chosen for, AI training, so it starts ahead of the wider workforce, where we estimate 11% of knowledge workers are at Level 2 or higher.)
Titles and Confidence Are Weak Guides to Fluency
Of the 73 people who lead AI adoption for others, 44 (60%) sit at Level 0 or 1. They either avoid AI or use it one prompt at a time.
Adoption leads are still ahead of everyone else. Two in five of them are at Level 2 or higher, against 13% of the rest of the group.
Only 29 of the 73 people at Level 2 or higher lead adoption, which leaves 44 fluent people outside any adoption role.
Confidence is a weak guide too. Median AI confidence is 5 out of 10, and of the 120 people who rate themselves 7 or higher, 70 (58%) sit at Level 0 or 1. In employer-run programs, that share is 75%.
Pick AI champions on observed Level 2 work. Ask each candidate to show something they made with AI, including the context they gave it, how they iterated and how they checked the output. Leads teach the habits they have, so check the work before the title.
