Listen to enough vendor spiels, and one can be forgiven for thinking that plugging in the right artificial intelligence into their operations will deliver overnight miracles. However, all the measurements and metrics that seem to point to smashing AI potential keep missing a crucial point, and potentially an imposing roadblock: human acceptance of AI and its adaptation into workflows.

A perhaps fitting term one survey assigns to this factor is “ AI-whelmed workers ,” who understand AI is supposed to deliver tremendous benefits, but aren’t given guidance on how to do this. 42% are not confident integrating AI into their workflow. At least 44% say they do not have a clear path for where to start when building AI skills, and 60% say AI is either not expected at their job or has come up without well-defined expectations, the survey of 1,000 U.S.-employed workers by Resume Now finds.

A more rigorous academic analysis dove even deeper into the factors that are missed by AI proponents. Most AI studies have measured automation capabilities and AI adoption patterns, according to a paper that summarizes discussions and findings of the CIVIC-AI 2026 workshop held in July in Singapore.

Indeed, many surveys out there – both corporate and academic – show X percentage of companies have adopted AI, and X percentage haven’t seen the value, or whatever the case may be. “Such metrics ignore the greater impacts of human-agent collaboration in transforming the nature of work," the paper’s authors state. What is needed is a more holistic picture of how AI is augmenting workflows – via “task allocation, decision rights, hidden verification work, recovery burdens, and human capability evolve together over time.”

“AI doesn’t just vaporize work,” said Andy Thurai , founder of The Field CTO and former chief strategist with IBM, commenting on the paper. “The friction moves downstream into messy exception handling and endless review cycles. The paper’s insistence on tracking the whole workflow instead of the single task is what lets you actually find where the bottleneck landed.”

AI may be simply speeding up one process, while creating more work for someone else. “While an AI system may reduce the time required to produce an initial output, it may increase the effort required for verification, exception handling or recovery,” the paper contends.

“Such verification costs must be included when accounting for the value of the redesigned workflow," they assert. "Structural changes to workflows can also erode human capability, such as when workers’ skills atrophy. The integrity of human–AI collaboration drops if humans can no longer reliably oversee AI outputs.”

The paper “skips straight past how much AI can automate and asks whether the redesigned work is still holding up five years out,” said Thurai. "Most enterprise AI scorecards can’t answer that yet."

Adoption rates and productivity surveys may measure activity, but “people are leaning on a pretty dumb proxy for value here, mostly just tokens used,” Thurai continued. "If a team saves ten hours drafting and spends eight checking and fixing what the AI produced, the dashboard logs a win. The business barely feels it. The paper calls this ‘durable net value,’ and every CFO should be asking for it.”

While Thurai considers the paper to be a “well-reasoned checklist,” examining the efficacy of AI-enabled workflows may still be missing the essential ingredient to promote successful AI implementations. "The paper assumes enterprise workflows get designed centrally and rolled out top-down in some orderly way,” he said. “That’s not really how it’s happening. AI tools are spreading bottom-up, with developers, sales teams, and marketers wiring autonomous agents into their daily work to route around friction, well outside any centralized design process. Good luck governing a workflow that’s mutating organically at the edge of the network.”

Before reporting AI productivity numbers, managers need to "report what it cost to verify and fix,” Thurai urged. “And for every role you automate, ask who’s going to be qualified to supervise that system ten years from now. If you can’t answer that, you are simply borrowing against the workforce that was supposed to replace itself.”