More organizations are equipping employees with AI tools to boost productivity. But according to a survey of 1,700 senior and director-level managers in seven countries, the results have been messy—especially for those tasked with oversight. While polished work is flowing into their inboxes, managers are struggling to assess what’s genuinely the employee’s work versus what’s AI-assisted. The clean outputs hide contribution levels, making performance evaluation harder. As a result, nearly half of managers report spending more time reviewing and guiding team output than they did before AI tools were introduced.
Rewired Evaluation Criteria
Despite the evaluation challenges, managers aren’t losing faith in AI. In fact, 94% express excitement about AI’s potential to improve work and believe they have a solid sense of where it’s most valuable. But many concede current workplace structures can’t keep pace. Managers are changing how they assess employees: 78% now place greater emphasis on how well someone can craft prompts, decide what to assign to AI, and experiment with new workflows.
At the same time, AI itself is increasingly baked into performance review processes. Almost half of those surveyed use AI to aid decision-making, while 39% rely on it to draft review write-ups. That brings serious concerns about fairness. Large language models (LLMs) may recycle biases present in their training data—especially when assessments lack clear objective standards or oversight. Earlier research has shown LLMs can systematically disadvantage certain groups in performance reviews if not carefully managed.
Infrastructure, Oversight, and Invisible Labor
The survey suggests AI adoption has also spurred growth in software tools. More than half of managers say their organizations now use more tools than last year, and many report more complexity and interdependence among those tools. Only 29% feel their AI infrastructure is fully optimized or centralized. Shadow AI—when employees use tools outside sanctioned systems—is causing additional friction.
Managers are facing a new type of invisible labor: deciding what tasks to delegate to AI, reviewing its output, or stepping in when agents veer off course. Over half expect to manage more AI agents than human staff in the upcoming year. Yet nearly half also fear losing control over those agents. There’s growing discomfort with being held accountable for decisions made by AI systems—over a third of respondents voiced that concern. Many favor keeping human oversight at key decision points.
These findings make clear that adopting AI isn’t just about productivity. It’s fast becoming a governance challenge. Without aligned processes, ethical guardrails, and clear expectations, managers risk being bottlenecks in environments where tools evolve faster than structures can support them.
What this means: The survey underscores a pivotal inflection point. Organizations that want to seize AI’s productivity gains need to rethink performance metrics, invest in transparent tools, and update managerial training. How companies handle AI now will shape not just efficiency, but fairness and workplace trust.
