A new survey finds that more than half of managers say it has become harder, not easier, to evaluate their employees' work since AI tools entered the workflow. That finding cuts against the standard pitch for workplace AI: that it frees up time and sharpens output. For many managers, it has instead muddied a basic question they used to answer without much thought — is this person doing good work?
The survey, conducted among managers across a range of industries, points to a specific problem rather than a vague discomfort with new technology. When an employee submits a report, a piece of code, or a marketing draft, managers increasingly can't tell how much of it came from the person versus from a chatbot or writing assistant. That ambiguity complicates everything from performance reviews to deciding who deserves a raise or a promotion.
This isn't a new tension, but it has been building quietly for about three years. Early AI adoption in workplaces focused almost entirely on output: did the tool help people finish tasks faster. Less attention went to a second-order effect — once the tool is doing part of the work, the old yardsticks for measuring individual skill and effort stop working cleanly. A junior employee using AI well can look identical, on paper, to a skilled senior employee doing the work manually.
Managers in the survey described a related issue: it's harder to tell who on a team is actually developing skills versus who is becoming dependent on AI to produce passable work. That distinction matters for staffing decisions, training budgets, and succession planning, but there's currently no standard method for making it.
This fits a pattern that has shown up repeatedly as generative AI tools have spread through workplaces over the past two years. Tools get adopted fast because the upfront benefit — speed — is obvious and easy to sell internally. The harder, second-order costs, like how to manage, evaluate, and train people once the tool is in daily use, tend to surface six to eighteen months later, after adoption is already widespread. The same lag played out with remote work tools and productivity software in prior tech cycles, just on a slower timeline.
What's different this time is the speed of the gap. AI writing and coding tools reached broad workplace use within roughly a year of ChatGPT's late-2022 launch, far faster than previous enterprise technology shifts. Management practices, performance review systems, and HR policy simply haven't had time to catch up.
For a small business owner, this shows up in concrete ways. If you have even a handful of employees using AI tools for writing, customer service, scheduling, or analysis, your existing review process — annual evaluations, task checklists, output quotas — may no longer measure what you think it measures. An employee hitting every metric might be doing so almost entirely through AI assistance, which isn't inherently a problem, but it changes what you're actually paying for.
The practical fix doesn't require new software. It requires updating what you ask employees during reviews: not just what did you produce, but how did you produce it, and what would you do without the tool. Some small businesses have started asking employees to log which tasks involved AI assistance, not to police usage but to separate skill development from tool dependence when making decisions about raises or added responsibility.
Watch for whether larger employers start building AI-usage disclosure into formal review templates over the next year — that's usually a sign a practice is about to trickle down to smaller operations through HR software vendors. Also worth tracking: whether performance review platforms like Lattice or 15Five add AI-usage tracking as a built-in feature, which would signal this has become a mainstream concern rather than a niche one.
The bottom line: AI tools are changing how work gets done faster than most businesses have updated how they measure who's doing it well, and the gap is now showing up as a measurable management problem rather than a theoretical one.