Employees use AI, but many say they don't know why

You have probably sat in a quarterly review where a leader announces that AI adoption has doubled, then watched the same leader struggle to name a single workflow it improved. The gap between usage and understanding now defines enterprise AI programmes, and a recent Culture Amp analysis suggests that gap may be wider than most organisations admit.

Workload rates remain steady across employees who use artificial intelligence and those who do not, a vice president at Culture Amp observed. That steadiness is not a rounding error. It is a signal that companies need to reexamine their overall return on investment. When tools proliferate but outcomes stay flat, the problem is rarely the tool. It is the absence of a theory about why the tool exists in the first place.

What the steadiness reveals

Stable workload across user cohorts means one of three things is true. Either AI users are applying the technology to tasks that do not compress time, or they are filling the recovered hours with new work, or the tools themselves are too narrow to move the dial on effort. All three scenarios point to the same root cause: organisations rolled out capability before they articulated purpose.

The pattern is familiar to anyone who lived through the first wave of collaboration software. Slack channels multiplied, notification badges climbed and meeting calendars stayed full because no one paused to ask which conversations the tool should eliminate. AI risks the same fate when adoption becomes the metric and clarity becomes optional.

The literacy gap

Many employees report using AI without understanding why. That admission is more honest than most vendor case studies, but it also exposes a structural failure. Training programmes that focus on feature lists rather than decision frameworks leave people equipped to generate text but unable to judge whether the text is worth generating. Literacy is not familiarity with a prompt library. It is the ability to recognise which problems compress under automation and which problems require human judgement that no model can replicate.

Organisations that treat AI as a software rollout rather than a capability shift will continue to see usage climb and impact stall. The fix is not more tutorials. It is a return to first principles: what work do we want to stop doing, what work do we want to do differently and what work do we want to protect from optimisation altogether?

What this means for L&D

Learning and development teams now face a choice. They can continue to measure seat time in AI workshops, or they can measure the quality of the questions employees ask before they open a model. The latter is harder to instrument, but it is the only measure that correlates with the outcome Culture Amp identified: workload reduction that shows up in calendars, not just in survey responses.

Effective AI enablement starts with a use-case audit. Which tasks do employees currently perform that a model could handle with equal or better accuracy? Which tasks require context, negotiation or ethical judgement that a model cannot supply? The audit is not a one-time exercise. It is a standing practice, because the boundary between automatable and non-automatable work shifts as models improve and as employees learn to articulate what they actually need.

The steadiness in workload is not a failure of the technology. It is a failure of the conversation that should have preceded the technology. L&D leaders who close that gap will find that adoption and impact finally move in the same direction.

Sources:

• HR Dive, 23 September 2026