AI Is Generating Work, Not Eliminating It — And Leaders Are Missing the Signal

The productivity promise of AI rests on a simple assumption: that outputs from these tools flow cleanly into value. That assumption is already cracking. According to a Glean survey, for every hour an employee spends getting a useful output from AI, they spend another hour making it usable. That's a 2:1 effort ratio — not the efficiency multiplier your board approved in the AI budget line.

This isn't a tooling problem. It's an organizational design problem dressed up as a technology problem. And most leadership teams are solving for the wrong layer.

The Gap Isn't in the Models — It's in the Operating Model

Leaders at the 2026 MIT Sloan CIO Symposium were candid about the distance between AI's promise and its operational reality. As MIT Sloan Management Review framed it: the question is no longer whether the agents are ready, but whether the humans are. That framing matters. It shifts accountability from IT procurement to organizational readiness — which means HR and operations leaders are now squarely in the hot seat, whether they've accepted that or not.

The companies that have figured this out share a common trait: they stopped treating AI as a tool attached to existing roles and started redesigning roles around what AI can actually own. McKinsey's analysis of 15 AI-native companies found that most organizations get this transition fundamentally wrong — not because they lack AI access, but because they lack the operating architecture to absorb it. Almost every company now has AI tools; almost none have restructured decision rights, workflow ownership, or accountability systems to match.

The implication is uncomfortable: your AI investment may be generating activity without generating value, and your current org chart obscures that entirely.

The Leadership Gap Is the Load-Bearing Wall

New data from Marsh and Mercer makes the structural problem explicit. Their research links leadership gaps directly to AI underperformance — not as a soft observation about culture, but as a measurable driver of workforce risk alongside disengagement and escalating operational costs. Leaders who cannot articulate what AI should not do — what judgment, context, and relationship work must stay human — are the single biggest bottleneck in AI adoption.

This is where the Glean data becomes a strategic signal rather than a complaint. When workers spend as much time managing AI as producing work, it means no one upstream has defined the handoff points. That's a leadership design failure, not a user error. The question worth asking before your next AI rollout: who in your organization actually owns the human-AI boundary, and what authority do they have to enforce it?

What AI Still Can't Do — And Why That's the Real Agenda

MIT Sloan professors Deborah Ancona and Katherine W. Isaacs have articulated the clearest version of this problem: leaders are giving away too much agency to AI in precisely the domains where human judgment is irreplaceable — sense-making in ambiguous situations, building trust across teams, and holding accountability through organizational change. The risk isn't that AI replaces these leaders. It's that leaders outsource the hard parts and hollow out their own authority in the process.

There's an additional blind spot that compounds this. Research flagged by Personnel Today reveals low awareness among employers about which workers are most exposed to AI displacement — particularly those in entry-level roles. If your AI strategy is reducing headcount at the bottom of the pipeline, you are simultaneously destroying the talent development infrastructure that produces your next generation of mid-level managers. That tradeoff rarely surfaces in AI business cases.

The Reframe Senior Leaders Actually Need

The board conversation about AI ROI is happening in the wrong register. It's being measured in cost reduction and output volume, when the real variables are decision quality, organizational learning capacity, and whether you're building a workforce that can adapt or one that's simply compliant with the current toolset.

The companies winning on AI aren't the ones who deployed fastest. They're the ones who treated the human operating model as the primary design constraint — and had leaders willing to make explicit choices about where AI stops and judgment begins. The rest are measuring hours saved while quietly accumulating a very different kind of debt.