Barclays is extending Claude across the bank and expects adoption of Claude Code to reach half its developer population by the end of 2026, rising to a majority of its software engineers during 2027. Anthropic published the expanded agreement on Thursday.
What makes it worth reading is that it comes with figures. Most bank AI announcements do not.
What is already running
Barclays UK has had a Colleague Knowledge Assistant in production since 2025, built on Claude with a retrieval-augmented generation architecture. Anthropic says more than 16,000 colleagues have adopted it and it has handled over a million searches, helping staff answer questions for the bank’s more than 20 million UK retail customers.
In the Global Markets business, Claude models classify incoming client emails, enrich them and decide how each should be routed. The platform processes approximately 120,000 emails a day. That is a narrow, high-volume task with a measurable error rate, which is the shape of deployment that tends to survive contact with a regulated environment.
What the bank says it is for
“Software engineering and cyber security are both being reshaped by increasingly capable AI systems,” said Craig Bright, group co-chief operating officer at Barclays. “We’re moving towards AI as an increasingly agentic capability embedded within how we build, test, secure, and operate technology.” He said the bank is using Claude to modernise legacy platforms and improve software quality.
Anne Marie Darling, the other group co-chief operating officer, framed it around outcomes rather than adoption: “The true measure of any technology is the impact it has on customers, clients, and colleagues, and the outcomes it enables.”
The timing
Anthropic is signing large regulated customers while preparing a listing that Bloomberg reported on Thursday could be marketed from the week of 9 November. A universal bank putting a coding agent in front of most of its engineers is the kind of reference an issuer wants in the room during a roadshow.
What to watch
Whether Barclays publishes anything about results rather than adoption. Developer headcount using a tool is an input. Defect rates, delivery times and the cost of running the legacy estate are the numbers that would show whether it worked.
