TL;DR
A Government Digital Service blog post published on 14 October reports that AI-assisted pilots let public bodies understand old IT systems far more quickly. At the Student Loans Company, analysis expected to take about five weeks was done in under a day. At the UK Hydrographic Office, regression testing that took roughly 15 days by hand dropped to around half an hour. GDS stresses the results will not transfer neatly to every estate.
The two case studies
The Student Loans Company used AI to examine 33 applications written in Clojure. The output gave the team a map of the applications and how they depend on each other, which fed into planning a move to supported versions of Java. GDS says the bigger win was not the time saved but the fact that the work, previously hard to justify, could now go ahead.
The UK Hydrographic Office pointed AI at a desktop application two decades old that had never had automated tests. The tools generated a test suite that took coverage from nothing to 96%. Engineers said they felt more confident changing the software, because they had more evidence to weigh each proposed change.
The Met Office also took part, and Kainos acted as delivery partner for the pilots, though the post gives no Met Office figures.
The caveats
GDS is careful about what the pilots show. AI sped up the discovery and assessment stage, where many modernisation projects stall, but it did not replace governance, engineering skill or delivery capacity. Results depended on clear ownership, access to the systems and experienced teams. The post warns that, used without proper controls, AI can magnify existing problems and speed up bad decisions.
GDS is now writing a playbook of good practice, currently in user testing and review, and invites organisations to get in touch through its Firmer Foundations programme.
Looking forward
These findings sit alongside the Public Accounts Committee chair’s warning that Whitehall is slow to replace legacy systems. In our view, the pilots point to a practical first use of AI in old estates: understanding and testing code before anyone rewrites it. For UK councils, NHS trusts and businesses running ageing systems, that is a lower-risk place to start than full AI-driven migration, provided experienced engineers check what the tools produce.