The AI skills gap in American workplaces may soon have less to do with who can write the smartest chatbot prompt and more to do with who can safely delegate work to software that can research, make critical decisions, and act.
That is the distinction Nate Herk, a 24-year-old Chicago-based AI automation creator and educator, sees emerging as businesses move from generative AI assistants toward autonomous agents.
"Someone who’s only learned to prompt a chatbot takes what the model gives them and passes it along," Herk said in an interview with Benzinga. An AI-ready worker, by contrast, can break a job into tasks, decide which should go to AI and "use their own expertise to check what comes back."
The gap matters as employers raise their expectations. A 2026 Harris Poll survey for Express Employment Professionals, a staffing and workforce-solutions company, found 93% of U.S. job seekers believe they possess the AI skills employers want. Still, only 35% think companies expect advanced or expert proficiency. Among hiring managers, 41% say candidates need skills at that level, while 90% look for at least basic AI proficiency.
AI Literacy Moves Beyond The Prompt
According to data from LinkedIn, the world’s largest online professional networking platform, demand is already headed in that direction. U.S. jobs requiring AI-literacy skills grew 70% year over year, LinkedIn said, while Gallup, an analytics and advisory company known for its workplace and public-opinion research, found in July that 52% of American employees used AI at work by mid-2026. Only 30%, however, used it a few times a week or more and 15% used it daily.
For Herk, a former Goldman Sachs Group, Inc.‘s (NYSE:GS) business-intelligence analyst, fluency means more than frequency.
"You should be able to explain why you used AI for a task and how you know the result is good," he said. "You can outsource the thinking, but you can’t outsource the understanding."
Research from Microsoft Corp. (NASDAQ:MSFT), the technology giant behind the Copilot AI platform, points in much the same direction. Its 2026 Work Trend Index found 50% of surveyed AI users said quality control of AI output had become more important and 46% pointed to critical thinking. Some 86% said they treated AI output as a starting point rather than a final answer. Microsoft said workers are increasingly shifting from generating answers themselves toward "evaluating, refining, and owning" AI-generated work.
The emphasis on judgment has also surfaced in warnings from former Microsoft Business Division President Jeff Raikes, who has argued that AI literacy without critical thinking could leave companies with a workplace skills problem if automation removes the entry-level experiences through which employees learn judgment.
That reinforces Herk’s argument that traditional workplace skills are changing rather than disappearing.
"If two people have those qualities and only one knows how to use AI well, I’m taking that person ten times out of ten," he said.
Agents Raise Stakes When They Act
The bigger shift comes when AI stops drafting and starts doing.
"The risk goes up exponentially," Herk said of agents that can send emails, update databases or make purchases. He recommends narrowly scoped permissions and human approval before actions that are difficult to reverse. "If you don’t want it sending emails, don’t give it permission to send emails."
That closely matches guidance from the Open Worldwide Application Security Project (OWASP) Foundation, a global nonprofit that develops open-source guidance to improve software security. OWASP warns against giving AI agents excessive functionality, permissions or autonomy and recommends human approval before high-impact actions.
But Yoshua Bengio, a pioneering artificial intelligence researcher and founder of nonprofit AI-safety research organization LawZero, takes the concern further. Bengio said at the United Nations Security Council in September that increasingly capable agents had already displayed dangerous behavior against instructions, including attempts to evade controls. His research argues that unchecked AI agency could create risks ranging from malicious misuse to loss of human control.
Herk sees a more immediate workplace failure in which an agent can follow instructions correctly and still produce a bad result because its underlying information was stale, contradictory or wrong.
"If you use a hundred agents to build something and it does well, you deserve the credit," he said. "If it goes wrong, you don’t get to say, ‘Well, the AI did it.’"
Experts Split On How Far Automation Goes
Herk also pushes back on promises that one autonomous system can replace an entire department.
"The claim I think is overhyped is, ‘One agent runs my whole marketing team,’" he said. "I automate tasks, not jobs."
That view has powerful supporters and equally prominent skeptics.
Anthropic CEO Dario Amodei, the artificial intelligence company behind the Claude family of models, has offered a much more disruptive forecast. He warned in 2025 that AI could eliminate as many as half of entry-level white-collar jobs within several years, particularly across technology, finance, law and consulting. Amodei argued that AI use would increasingly shift from helping employees work toward actually automating jobs.
Jensen Huang has publicly challenged that outlook. Huang said technological advances tend to create new work as productivity expands and argued companies will need people to manage growing fleets of AI agents. "I don’t know why AI companies are trying to scare us," Huang said while offering what he called a counter-view to predictions of widespread job losses.
The evidence so far remains mixed. Anthropic’s Economic Index initially found AI usage leaned toward augmenting workers rather than fully automating their tasks, while its 2026 research emphasizes substantial uncertainty around how quickly AI will ultimately affect employment and wages.
Herk advises job seekers to prove they can navigate that uncertainty instead of simply advertising familiarity with the newest tool. "Learn to think like a consultant," he said. "Figure out what’s slowing [a business] down, and solve that problem with as little AI as you need."
Photo courtesy: Golden Dayz / Shutterstock
© 2026 Benzinga.com. Benzinga does not provide investment advice. All rights reserved.
