Half of managers now say that evaluating employee performance has become harder since adopting AI tools. That finding, from a recent survey, should stop every leader cold, not because AI is failing, but because the frameworks we use to judge work were designed for a world that no longer exists.
The core problem is straightforward: when an employee uses an AI assistant to draft a report, write code, or analyze a dataset, what exactly is being evaluated? The employee's strategic thinking and oversight? The quality of the AI's output? The skill with which the employee prompted and iterated? Most organizations haven't answered that question, so managers fall back on instinct, and instinct is unreliable when the work itself has been transformed. This echoes the knotty ownership questions we explored in When AI Writes Code, Do You Still Own the Work. If you can't confidently say who contributed what, you certainly can't evaluate the person's contribution fairly.
Blame is misplaced here. Managers aren't incompetent; they're operating with a legacy toolkit. Performance reviews, annual goals, and output metrics were built for a slower, more visible era of work, type a memo, attend the meeting, ship the feature. AI collapses that timeline and makes the input invisible. A developer might produce three times the code in half the time, but if your evaluation system counts only output volume, the manager sees a suspicious spike. Worse, if the manager suspects AI assistance but can't separate skill from tool, resentment and mistrust creep in. That's not a technology problem; it's a management design failure.
There are practical lessons here from past transitions. Legacy digital shifts, from paper to spreadsheets, from on-premise to cloud, required leaders to stop measuring old proxies and start measuring new outcomes. Our piece What Legacy Digital Shifts Can Teach Us About Smarter AI Adoption makes the case that the same principle applies today. You can't evaluate a person's use of AI with the same yardstick you used for manual work. Yet that's exactly what most organizations are trying to do.
The specific takeaway is this: managers need to shift from evaluating outputs to evaluating judgment. Did the employee know when to use the AI and when to override it? Did they check the AI's work for errors? Did they interpret the results thoughtfully? Those are human skills, and they're the ones that matter. Measuring them requires new habits, brief weekly check-ins focused on process, not just results; peer reviews that capture collaboration; and explicit training for managers on what good AI-augmented work looks like.
The open question is whether organizations will invest in that retooling, or whether they'll keep expecting managers to solve an unsolvable puzzle. The survey suggests we're at a fork. One path leads to better evaluation frameworks and more confident teams. The other leads to frustrated managers, demoralized employees, and a slow retreat from the very tools that could make everyone more productive. Which one we take is not a technology decision. It's a leadership decision, and it's overdue.