Middle Managers Bear Hidden Cost of AI Adoption at Firms

Organizations rolling out AI tools typically frame adoption as a technology challenge—a question of software deployment and executive vision. But research into how two major consulting firms actually use artificial intelligence reveals the real friction point lies elsewhere: in the middle management layer now carrying unsustainable operational burdens.

Julia Shin and Sandra J. Sucher conducted 18 semi-structured interviews with partners, managers, and junior consultants to understand how AI adoption plays out in practice. Their findings, first reported in Harvard Business Review, show a consistent pattern across both firms studied.

Why it matters

Roughly 88% of organizations now deploy AI in at least one business function, yet only about a quarter generate tangible value beyond pilots. The gap between adoption and impact often traces to a structural problem: middle managers absorbing new AI-related responsibilities—validating outputs, coaching teams on prompting techniques, catching errors in polished but hollow work—without corresponding support or reduced delivery pressure. This dynamic threatens both immediate productivity gains and long-term leadership development.

The convergence problem

The research identified a clear pattern. Senior leaders pursue strategic opportunities with AI, expanding scope and reimagining services. Junior staff report dramatic productivity improvements—desktop research compressed from days to 30 minutes, analysis that once required weeks now completed in hours.

These gains converge on managers, who face what the researchers describe as a triple burden: managing AI experimentation, maintaining client delivery standards, and developing people. Managers spend mornings learning new prompting techniques, days validating AI-generated work for what they call "workslop"—content that appears professional but lacks substance—and evenings coaching analysts who've never built presentations manually.

This pressure compounds an existing crisis. Gallup data shows manager engagement dropped from 30% in 2023 to 22% in 2025, the steepest decline among employee groups. Gartner predicts 20% of organizations will use AI to eliminate more than half of current middle management positions in 2026.

Three structural breakdowns

The research identified specific failure modes. First, learning remains informal while delivery pressure stays constant. Teams that handled this better protected dedicated time for experimentation and built centralized hubs to capture and redistribute effective practices.

Second, incentive systems haven't adapted. Traditional metrics reward billable hours and individual output, while behaviors that drive AI adoption—sharing prompts, coaching others, contributing to internal tools—go unrecognized. Some employees avoided acknowledging AI use, reflecting systems that still equate personal effort with professional value.

Third, executives and managers operate in different realities. BCG survey data shows executives are roughly twice as likely as individual contributors to describe employees as enthusiastic about AI. Partners often remain removed from how AI changes operational work, leaving managers to make critical decisions in isolation: when AI output meets standards, what juniors should still learn manually, how to handle clients who assume all work is AI-generated.

The leadership pipeline risk

Beyond immediate operational strain, the research surfaces a deeper concern. Traditional consulting development relied on juniors watching managers structure workplans, pressure-test analysis, and navigate client conversations. If managers spend capacity validating AI outputs instead of coaching, firms risk hollowing out the path from contributor to leader.

The researchers recommend three interventions: temporarily lower utilization targets during AI transitions, tie performance reviews to knowledge-sharing and coaching rather than delivery alone, and provide manager-specific training on AI oversight including hallucination detection and prompt evaluation.

The difference between successful and struggling AI adoption, according to the research, isn't technology sophistication—it's whether leadership has built support structures around the people making AI work in practice.

The findings were reported by Julia Shin and Sandra J. Sucher in Harvard Business Review based on their research at two major consulting firms.