The UK government argued with itself about AI in public

Twenty-six minutes saved per civil servant. Or nineteen. Or no real productivity gain. The chair of Parliament now wants the math explained.

Hey Adopter,

The UK government published the most detailed civil service AI trial in history. Twenty thousand civil servants, twelve departments, three months. The headline number traveled the world: 26 minutes per day saved. Nearly two weeks per year per civil servant. Eighty-two percent of users said they would not go back to working without the tool.

Three months later, the same government quietly published a second study of the same trial period. That study found “no robust evidence” that the time savings translated into improved productivity. Excel work got worse with Copilot. Twenty-two percent of users identified hallucinations. The Department for Work and Pensions later ran a third study with a comparison group of 2,535 non-users and reported 19 minutes per day, not 26.

In March 2026, the chair of the Public Accounts Committee wrote to the Cabinet Office permanent secretary calling the original 26-minute figure “curiously specific” and asked her to explain how it was calculated. As of mid-May, no public response.

That is the most rigorous public-sector AI trial ever conducted. And it is still arguing with itself.

Which raises the question this edition is actually about. Why did the UK run the trial in the first place?