How does generative AI affect real workplace productivity? This column reports results from an experiment at the National Bank of Slovakia, where staff were randomly assigned GPT-4o access while completing routine and non-routine tasks. The model improved output quality by up to 44% and cut completion time by 21%, with the largest gains on non-routine, domain-specific work. Lower-skill staff saw the biggest quality boosts, while higher-skill colleagues saved the most time. The findings