The AI capability gap has a mandate problem

Most organizations say AI capability matters. Fewer have decided who is accountable for building it in the work. The edX Enterprise and NewtonX State of Enterprise Learning 2026 survey says 83 percent prioritize AI capability, 29 percent report mature capability, and 78 percent lack a strong AI upskilling mandate. That supports a narrower reading than broad enthusiasm suggests: priority is present, but authority and role clarity may not be.

The survey describes a mandate in terms of executive support, priority capabilities, and money to deliver them. That is necessary, but it is not yet an operating decision. Start with one role group. Identify a task AI has changed, the judgment the employee still owns, the handoff to a manager, and the evidence that the task is performed to standard. Then assign a business owner who can change that workflow. An L&D team cannot be accountable for a process its managers cannot authorize.

Perceptyx reports that 75 percent of executives say their organization has a clear learning strategy for the next one to two years, compared with 43 percent of individual contributors. That is a 32 point perception gap. It may be a communication problem, but it can also point to an operating problem: the strategy may not yet be translated into role expectations, manager guidance, workflow changes, or local accountability in the parts of the business where people are meant to act on it. The survey alone cannot tell us which of those is true. Those are the places I would inspect next.

Other surveys point to the same strain in the operating model. HRD reported an LHH leader citing Adecco Group benchmarking that perceived future readiness fell from 10 percent to 6 percent. Thoughtworks, in a survey of 3,200 CIOs, found 89 percent increasingly responsible for workforce workflow redesign and 88 percent saying AI adoption is outpacing governance. That does not prove a single ownership model is right for every company. It does suggest that tool adoption, workflow redesign, governance, and capability building are moving on different tracks, and someone has to connect them.

For the next AI training request, ask the sponsor to select three work tasks that employees will perform differently. Have the role owner show a current example, the new expected output, the escalation boundary, and how the team will verify quality in actual work. If those cannot be described, the missing input is work design, not a course. Once the business sets the performance standard, learning can choose the minimum practice and support needed.

A named executive mandate should settle who changes the work, who can approve safe AI use, and who is responsible for evidence that performance improved. Test that on one role group before spreading a company-wide training program. Governance may remain shared; accountability for the work cannot disappear between IT, HR, and line management. These vendor surveys report perceptions rather than directly measured capability, so they justify investigating the gap, not assuming its cause. The practical result to track is work that meets the new standard, including the exceptions people detect and escalate, rather than course completion.