Organizations are tying AI proficiency to promotion and pay decisions while the people responsible for developing that proficiency say they are not ready to do it. That is not a training gap you close with another course. It is a structural problem, and most of the writing on manager readiness stops well short of naming it.
Two numbers tell the story. From our own session with HiBob, only 36 percent of managers feel prepared to upskill their teams on AI, even as two-thirds of organizations now link AI skills to promotion decisions. Zoom out further and it gets worse: Gartner's 2025 Leadership Perspective Survey found that only 28 percent of CHROs believe their organization's leaders are adequately prepared to lead change of any kind, AI included.
Read those together and the pattern is not a coaching skills gap. It is that the entire manager layer, the one every AI-proficiency initiative depends on to actually happen, was not built or resourced for this job.
Because the tasks junior employees used to learn on are the same tasks AI now absorbs, and almost nobody is tracking the loss in real time. Research from McKinsey partners Bryan Hancock and Charlotte Seiler describes it plainly: research, documentation, data cleanup, and preliminary analysis, the traditional building blocks of a junior career, are being streamlined or absorbed into AI systems. That is the exact grind that used to build instinct, one corrected mistake at a time.
The data backs this up. Stanford's Digital Economy Lab, using ADP payroll records, found employment among 22- to 25-year-olds in the most AI-exposed occupations sitting roughly 19 percent below where it would be if it had kept pace with less-exposed peers. Critically, the researchers found no comparable gap among more experienced workers in those same roles. The damage is concentrated at the bottom rung, precisely where the apprenticeship used to happen.
This is a manager problem before it is anything else, because managers are the ones who assigned that grind work in the first place, coached people through it, and watched judgment get built one correction at a time. Remove the grind without redesigning the role, and the manager loses the mechanism they used to develop people, often without realizing it happened.
Accenture's Hadi Skalli put the shift directly: AI is transforming middle managers from coordinators of work and hierarchy into leaders of outcomes and human-AI collaboration. Instead of producing status reports, managers increasingly need to interpret the signals AI produces and judge what to do next. Monitoring individual tasks gives way to managing outcomes, exceptions, and risk.
Dr. Kenneth Matos raised a related point in that same HiBob session: organizations may need to rebuild a middle management layer oriented around results rather than task supervision. If a manager's value used to come from knowing whether the work got done, and AI now does much of that tracking on its own, the manager's value has to come from somewhere else, specifically from judgment about outcomes and coaching people through ambiguity.
Three things, and none of them are a single training session.
A shared vocabulary for proficiency levels. Managers cannot coach what they cannot name. If your organization is building out the kind of layered skills structure we covered in our piece on skills taxonomies, give managers that same language: specific observable behaviors, not vague labels like “good with AI.” A manager who can say precisely what “proficient” looks like for a given behavior can coach toward it. One who can't is guessing.
Permission for managers to admit they are also learning. This matters more than it sounds like it should. The same psychological safety dynamic that drives employees to hide their own AI use applies to managers pretending to have answers they don't have. A manager who has to perform expertise instead of modeling learning teaches their team to hide struggle instead of surfacing it, which is precisely backward.
A method for evaluating skills they do not personally hold. This is the hardest piece and the one most enablement programs skip entirely. Some organizations are solving it structurally rather than through manager judgment alone: one real estate firm, per McKinsey's research, now has junior staff build market assessments by hand first, walking neighborhoods and studying traffic patterns, before comparing that work against an AI-generated version to expose the gaps. The manager is not evaluating AI fluency directly. They are evaluating whether the junior employee can explain where their own analysis and the AI's diverged, and why. That is a proxy any manager can assess, regardless of their own AI depth.
Treat junior training as a deliberate design choice again, not something that used to happen on the job by accident. Bank of America kept intern numbers close to 4,000 in 2026 specifically so it could build AI simulations designed to compress years of on-the-job learning into a shorter runway, rather than letting the pipeline erode quietly. The lesson generalizes: if AI has removed the task that used to teach the lesson, someone has to design a replacement task on purpose. That design work sits with HR and with managers together, not with either alone.
Coaching AI skills is not a standalone competency bolted onto a manager's existing job. It draws on the same underlying capabilities as every other kind of leadership development: emotional intelligence to hold the psychological safety piece, and digital fluency to actually understand what they are coaching toward. That is the combination behind our HQ Leader framework, and it is worth building manager AI-readiness on the same foundation rather than as a separate initiative competing for the same hours.
If you are rethinking how performance conversations, development plans, and manager training need to change to actually support this, that is exactly where our Performance, Learning + Development practice starts.
Because the manager layer was not built or resourced for it. Only 36 percent of managers feel prepared to upskill their teams on AI, and more broadly, only 28 percent of CHROs believe their organization's leaders are adequately prepared to lead change of any kind.
AI is absorbing the routine research, documentation, and analysis tasks junior employees used to learn on. Stanford's Digital Economy Lab found employment among 22 to 25 year olds in AI-exposed roles running about 19 percent below expected levels, with no comparable gap among experienced workers.
A shared vocabulary for proficiency levels, explicit permission for managers to admit they are still learning too, and a method for evaluating skills a manager may not personally hold, such as comparing a junior employee's own analysis against an AI-generated version.
Structural. The manager's job itself needs to shift from supervising tasks to managing outcomes and coaching judgment, which changes what the role requires day to day, not just what training it receives.
Design a deliberate replacement task rather than letting the gap go unaddressed. Some organizations now have junior staff complete an assessment manually first, then compare it against an AI-generated version to expose and discuss the gaps.