Ask a knowledge worker to define an AI agent, and odds are they can't. Section, an AI transformation firm, put that question to more than 5,000 U.S. workers for its AI Proficiency Report on July 7, and fewer than 10% got it right. But 69% of them say their company has already taken action on AI agents — the very thing most of them can't define.
Only 3.8% can write instructions to build one. Just 16% have actually used an agentic tool at work. And of the people whose companies are deploying agents, only one in three got any agent-specific training at all.
"The AI value gap isn't closing because most CEOs keep skipping the hardest part of this work."
— Greg Shove, CEO, Section, in comments accompanying the report
That's the whole problem in one line, and it's not a training-budget problem. It's a distribution problem. Section's data shows C-suite executives are five times more likely than individual contributors to receive agentic training. The capability is being deployed top-down and the literacy is being deployed top-down too, which means the people actually running the workflows the agents are supposed to touch are the last ones taught how they work — if they're taught at all.
Look at where this sits relative to everything else happening in AI staffing this year. Microsoft, OpenAI, Anthropic, and AWS, the four companies with the deepest AI talent benches on the planet, have collectively committed more than $9 billion to forward-deployed engineering units this year, betting that enterprise AI deployment is a headcount problem, not a software problem, and that the fix is embedding specialists directly into the customer's workflow. Section's report is the same diagnosis running one level down, inside the company that just bought the agent license. The vendor sends in specialists to make the tool work.
Palantir pioneered the model the rest of the industry is now scaling. Its CTO has been blunt about why the fix was headcount, not software:
"Palantir was ridiculed endlessly for our Forward Deployed Engineering model. Investors would impugn Palantir as a services company simply because we didn't believe that throwing our software over the wall for consultants to implement was the correct thing for our customers."
— Shyam Sankar, CTO, Palantir Technologies, Technology Is the Problem
Nobody's doing the equivalent for the floor.
And the instinct most companies reach for — a company-wide training rollout — doesn't produce what they think it produces. Section's proficiency tiers make that obvious: 73.5% of workers land in Experimenter, meaning basic, one-off AI use. Only 5.5% reach Practitioner or Expert, the tier where someone uses AI regularly in ways that drive real business value. A one-time course, an LMS module, a lunch-and-learn move people from never having touched it to Experimenter. It doesn't move them to Practitioner. Proficiency at that level comes from doing the work with someone who already knows how, not from watching a slide deck about it.
Docebo's own research on corporate AI training backs this up. Its CEO put it bluntly:
"Spending on AI training isn't the same as building capability. If it doesn't change what people can do in their work, you've spent the budget and built nothing."
— Alessio Artuffo, CEO, Docebo, Forbes
Which is the staffing decision hiding inside this data. You don't close a proficiency gap by training everyone a little. You close it by putting someone who's already a Practitioner inside the team that needs to become one — not a vendor's forward-deployed engineer parachuting in for a short-term engagement, but a person embedded in that function long enough to turn Experimenter into Practitioner through repetition, not a webinar. That's a smaller, cheaper hire than most companies are budgeting for, because most companies are budgeting for licenses and calling the rollout done.
If your company bought agent seats this year, ask where the training dollars actually went. If the answer is "a course everyone was assigned" or "the leadership team got briefed," you've replicated Section's finding inside your own building: capability at the top, a gap everywhere the work actually happens. The fix isn't more licenses. It's one person per function.
VC5 Consulting helps technology companies staff the roles that make AI tools actually usable inside real workflows — embedded AI operators and implementation leads, not just the executive briefing. If your agent rollout has a proficiency gap nobody's staffed for, let's talk.