Between November 2024 and February 2025, over 1,000 technical staff across 50 UK government departments trialed AI coding assistants — notably Microsoft GitHub Copilot and Google Gemini Code Assist — as part of a Department for Science, Innovation and Technology (DSIT) initiative. Results show developers saved about 56 minutes per working day (roughly 28 working days per person per year) through using AI to draft code, review existing code, search for examples, and other repetitive tasks. Satisfaction was generally strong: 72% said the tools offered good value, 58% would prefer not returning to pre-AI workflows, 65% said they completed tasks faster, and 56% felt problem-solving was more efficient. That said, only about 15–16% of AI-generated code was used without edits, underlining the continued need for developer oversight, especially around security, quality, and alignment with existing code bases.
Key Takeaways
– Strong Efficiency Gains, But Not “Set-and-Forget”: AI assistants delivered real productivity boosts — nearly an hour saved each working day — largely by helping write first drafts, perform reviews, and reduce search / lookup time. Still, the low rate of use-without-edits (~15–16%) shows human judgment remains essential.
– High Value Perception & Adoption Appetite: Most participants believe the tools provide good value (≈72%), many prefer working with them than without (≈58%), and a majority report faster task completion and more efficient problem solving (≈65% and 56%, respectively). This suggests AI tools may become standard tools in government dev workflows.
– Risk & Oversight Still Matter: Given that much of AI-produced code needs editing, attention must be paid to software security, code quality, governance, and making sure that deployment pipelines, testing, and review processes are robust. Guidance from the Government Digital Service reflects that risk.
In-Depth
Over the past few months, the UK government has taken measurable steps in integrating artificial intelligence into its software development workflows, testing how AI coding assistants can drive efficiency without sacrificing accountability. From November 2024 to February 2025, the Department for Science, Innovation and Technology (DSIT) ran a trial across more than 50 central government departments involving more than 1,000 developers who used tools like GitHub Copilot and Google’s Gemini Code Assist. The goal was to see just how much time could be saved in tasks like drafting code, reviewing existing code, and searching for examples — things that often chew up chunks of a developer’s day. The result? On average, around 56 minutes were saved every working day per person — a boost equivalent to 28 working days annually.
But the story isn’t purely about shiny numbers. While participants generally reported that the AI tools offered “good value” (72% agreed), many emphasized the importance of editing and oversight: only about 15–16% of AI-suggested code was used exactly as generated, without revision. That implies that while AI can speed up the initial phases of code work, critical checks — for correctness, security, alignment with standards — remain in human hands. Developers also noted improvements in how quickly tasks are completed and how efficiently problems are solved; roughly two-thirds said they got work done faster, and more than half said they solved problems more efficiently. Moreover, a significant share said they would prefer not to go back to working without AI aids.
On the flip side, there are still challenges. Ensuring secure code, preserving code quality, preventing overreliance on suggestions, integrating these tools into existing development pipelines, and managing licensing and procurement are all areas that warrant careful policy, guidance, and investment. That’s why the Government Digital Service has already published guidance for AI-Coding Assistant use, encouraging departments to require human understanding of AI outputs, responsibility over generated code, secure-by-design thinking, and sufficient test coverage. As the UK government eyes up to £45 billion in potential savings via broader AI adoption in the public sector, scaling these tools will require attention to both the upside and the risk.
In short: the trial strongly suggests that AI coding assistants can add up to major efficiency gains when used responsibly — but they don’t replace skilled human developers; rather, they partner with them.