Use of artificial intelligence by UK office workers is rising quickly, but a large share of the time saved is being used to review and rewrite AI outputs.

Tech site TechRadar reported on June 15 that UK office workers save an average of 12 hours a week through automation, but spend 6.3 hours of that on so-called "agent management".

A survey by AI startup Glean found 9 out of 10 UK office workers currently use AI for work. That compares with 84 percent in the United States. Still, only 42 percent described their workplace as an "AI-optimised environment". Use on the ground is increasing, but few organisations have shifted operating systems and performance management to be AI-centred.

The gap between perceived individual productivity gains and views of organisational performance was also large. Some 78 percent said AI improves their productivity, but only 18 percent said it has a meaningful impact on overall organisational performance. Employees may work faster, but whether that translates into company-wide efficiency gains is a separate issue.

The difference is seen as stemming from AI changing work toward supervision and verification rather than fully replacing people. Workers spent more time checking outputs than assigning tasks to AI. Agent management accounted for 38 percent of total related work time, while 36 percent was spent delegating tasks to AI.

Errors and rework also weighed on workers. More than one-third of AI use sessions failed outright, and 77 percent of UK workers said they had edited or redone AI-generated output in the past month. Some 26 percent said they did so in the past week.

Questions were also raised about how companies assess AI adoption. Rebecca Hinds (레베카 힌즈), head of the Work AI Institute, said many companies view AI adoption through usage rather than tangible performance. "There are limits to judging AI adoption based only on more accounts, more prompts and more usage," she said.

That has led to calls that IT organisations need more than simple time-saving figures to prove returns on AI investment. Companies should also consider the costs of error correction, prompt adjustments and output verification to gauge real performance. Hinds said productivity gains could be overstated, and Glean said remeasuring AI's actual impact is needed to identify where it delivers the best results.

In this environment, companies' challenge is shifting from expanding AI use itself to managing operational quality. Even with high AI usage, organisational performance gains may be limited if the burden of output review and rework does not fall. The metrics used to judge AI adoption are therefore increasingly likely to shift from usage to output accuracy and verification efficiency.