Companies are deploying agents. Fewer than one in ten employees can define one
A 5,000-worker study finds the agent rollout running far ahead of the workforce meant to use it. The study comes from a training vendor, which is worth knowing and does not change the numbers' shape.
- 69% of workers say their company has taken action on AI agents, but only 16% have actually used agentic tools at work, per a July 7 study of more than 5,000 US knowledge workers.
- Fewer than 10% could correctly define an AI agent. Just 3.8% could write instructions to build an automation.
- Organizations that deployed agents and trained people scored 47.5 out of 100 on proficiency, versus 33.1 for those that deployed without training.
- The study comes from Section, a company that sells AI training. Weigh accordingly.
The agent economy has a user problem. A study of more than 5,000 US knowledge workers, published July 7 by AI training firm Section, finds that 69% of workers say their company has taken some action on AI agents. Only 16% have actually used an agentic tool at work. Fewer than one in ten could correctly define what an agent is, and 3.8% could write the instructions to build an automation.
Experimenters, not operators
The distribution is the striking part. Section's assessments, which included hands-on prompting exercises rather than self-ratings alone, classify 73.5% of workers as "Experimenters" and 20.9% as "Novices," leaving 5.5% rated practitioner or expert. General AI use keeps climbing, 67% weekly now against 55% last October, but casual chatbot use and operating an agent are different skills, and the second one is rare.
Section CEO Greg Shove's read is aimed at his buyers, but it lands: "The AI value gap isn't closing because most CEOs keep skipping the hardest part of this work. It's easier to buy licenses than to rebuild how a team operates."
The training gap has a number
Organizations that deployed agents and trained their people scored 47.5 out of 100 on Section's proficiency measure. Deployment without training scored 33.1. Meanwhile 37.8% of workers report receiving no AI training at all, and of those who did get some, only 17% were trained on agents or automation specifically. The disclosure matters here: Section sells the training it just measured demand for, and no independent replication of these figures exists. But the direction matches what every enterprise deployment story this year has shown.
Who profits from the gap
If the workforce cannot operate agents, the tools-and-seats model stalls, and the advantage shifts to vendors who sell the finished outcome with the operator built in. That is the quiet labor story inside the GaaS thesis: the skill of running agents becomes something you rent along with the agent, until your own people catch up. The companies writing checks for agent licenses this quarter should probably read this study before renewal season.
Sources: Business Wire, Section AI Proficiency Report.