AI Productivity Gap: Heavy Users Submit Work They Don't Understand

What happened

A report from enterprise AI search company Glean has found that heavy users of AI tools are increasingly submitting work they do not fully understand — raising significant questions about quality, accountability and the true productivity gains organisations are chasing. Rather than freeing up meaningful time for higher-value thinking, the research suggests that much of the time saved by AI is immediately consumed by correcting the tool's errors or, more troublingly, by shipping outputs workers cannot confidently stand behind.

The findings point to a growing gap between the appearance of productivity and its substance: employees are producing more, faster, but with diminishing personal comprehension of what they are producing.

Why it matters

For anyone responsible for customer experience or service design, this is a quiet crisis hiding inside a productivity story. When frontline staff, content teams or support agents submit AI-generated work they do not genuinely understand, the downstream risk lands squarely on the customer. Responses become harder to personalise, errors go unspotted before they reach the customer, and the human judgment that catches edge cases — the kind behavioral economics tells us customers remember disproportionately — gets systematically bypassed.

There is also a deeper behavioural dynamic at work. Automation bias, the well-documented tendency to over-trust algorithmic outputs, is being turbocharged by organisational pressure to demonstrate AI's return on investment. Employees face an implicit incentive to ship AI output quickly rather than interrogate it slowly. The result is a workforce that is nominally more productive but potentially less accountable — a combination that corrodes service quality in ways that are difficult to detect until a customer relationship has already been damaged.

The Renascence take

Most organisations are measuring AI adoption by volume — tasks completed, time saved, tools deployed. Almost none are measuring comprehension: whether the person submitting the work could explain, defend or improve it without the AI. That is the metric that actually predicts service quality.

The Glean findings reveal something the productivity dashboard will never show: AI can hollow out the expertise of the very people customers depend on, while all the headline numbers trend upward. The behavioral principle here is not laziness — it is rational adaptation to a system that rewards speed and penalises slowness. Customer-obsessed operators should respond not by restricting AI use, but by redesigning accountability structures: require staff to annotate AI outputs with their own reasoning, build spot-check loops into workflows, and treat unexplained AI submissions the same way a good editor treats an unchecked fact. Comprehension, not completion, is the new quality standard.