Agentic AI Skills Gap: Only 13% of Workers Are Ready

The headline number

Enterprises racing to deploy AI agents may be outrunning the people expected to work with them. Workera's 2026 AI Skills Enterprise Benchmark Report draws on more than 88,000 skills assessments. It found that only 13% of employees qualify as "Accomplished" in agentic AI skills before any upskilling, the lowest result across the 14 capabilities the company measured. 12

The report does not only deliver bad news. It also argues that targeted training works. Only 25% of employees start out Accomplished in Responsible AI, but 81% reach that level after focused upskilling. 12 Workera CEO Kian Katanforoosh framed the issue as a matter of talent economics. Frontier labs such as OpenAI and Anthropic are hiring aggressively, and most companies cannot compete for that talent. Their realistic option is to make the workforce they already have AI-ready, which he said takes more than standard training. 1

Where the gaps sit

The more revealing finding is the pattern behind the numbers. Across all 14 capabilities, workers scored best where familiar skills carry over to new tools. They scored worst where the technology itself is newest. 2

The two weakest benchmarks were Deep Learning Fundamentals, with a score of 163, and a category covering prompts, agents and retrieval-augmented generation (RAG), with a score of 185. The 13% figure comes from that second category. 2

The report treats this as a strategic risk rather than a training footnote. If the gap is left alone, advanced AI capability ends up concentrated in a small number of people. That creates bottlenecks, raises project risk and leaves organizations behind competitors. 2 The report also flags a related imbalance: confidence in AI runs high while security readiness lags. 2

Enthusiasm without infrastructure

An EY survey of more than 1,100 U.S. desk workers fills in the human side of the picture. Respondents worked across six industries at companies with at least $1 billion in revenue. 3

Enthusiasm is not the problem. Some 84% of employees said they are eager to use agentic AI and expect gains in productivity, efficiency and their day-to-day work experience. 3 Anxiety runs alongside that optimism. Fifty-six percent worry about job security when working next to AI agents, and 51% fear the technology could make their role obsolete. 3

The most telling EY data point concerns how workers are learning. According to the survey:

• 85% of desk workers are learning to work with AI agents outside of work.
• 83% say most of what they know about agentic AI is self-taught. 3

EY places the blame on organizations rather than employees. It describes companies as fumbling the basics of strategy, training and communication. 3

Read together, the two studies fit. Workera finds low baseline proficiency, and EY finds that workers are filling the gap on their own time. Self-directed learning can produce the shallow, uneven familiarity that would show up as weak benchmark scores in the newest technical areas.

The organizational layer

The gap does not stop at individual skills. A Microsoft-hosted analysis cites the company's September 2025 Agent Readiness Survey of 500 enterprises. Roughly 60% of those organizations landed in the lowest "Discoverers" tier. Only about 20% qualified as "Achievers" across strategy, process mapping, data foundations and governance. 4

The same analysis cites Gartner figures showing a 1,445% jump in inquiries about multi-agent systems between Q1 2024 and Q2 2025. It also notes that 40% of multi-agent pilots fail within six months. 4 The analysis argues that stabilizing tooling, which Microsoft aimed to do by shipping Agent Framework 1.0, addresses only half of the adoption problem. 4 These figures come from a vendor-affiliated source, which is worth keeping in mind. Even so, they point the same way as the workforce data.

What it adds up to

The sources differ in method and incentive. Workera sells skills assessment and upskilling, so a finding that "training works" fits its business. EY's survey measures sentiment rather than competence. Microsoft's readiness tiers rely on its own framework.

That is exactly why their agreement matters. Interest in agentic AI is surging at the level of the individual worker, the IT inquiry and the executive agenda. Competence and organizational readiness are trailing on every measure presented.

Our reading is that the bottleneck in agentic AI is shifting away from model capability and framework maturity. It is moving toward people and process. Companies that treat agent deployment as a software rollout risk adding to the failed-pilot statistics. A stronger approach would pair it with structured, role-specific training and clearer communication about job impact.

The Responsible AI result offers a hopeful counterpoint. Measured, targeted programs can move proficiency quickly. The open question is whether the same lift can be achieved in the more technical agent and RAG skills, where the starting line is lowest.