TL;DR
Barclays is extending Anthropic’s Claude models across the bank, Anthropic announced on Thursday 1 October. The bank expects half of its developers to be using Claude Code by the close of 2026 and most of its software engineers in 2027. Two deployments are already running: a knowledge tool for Barclays UK staff, and email sorting in its markets division.
What is already live
The clearest evidence of scale is Barclays UK’s Colleague Knowledge Assistant. Staff use it to find answers while helping retail customers, of whom the bank has more than 20 million in Britain. It has run on Claude since 2025 using retrieval-augmented generation, which looks up relevant material before the model answers. According to the announcement, over 16,000 employees have taken it up and it has answered upwards of one million queries.
In Global Markets, Claude models read incoming client emails, tag them, add missing details and decide where each should go. Around 120,000 messages a day pass through that system, with the aim of cutting manual handling for operations teams.
Engineering is the bigger bet
The developer targets carry more weight than either deployment. Barclays wants Claude to help update ageing legacy platforms and raise software quality, according to Craig Bright, one of its two group co-chief operating officers. “We’re moving towards AI as an increasingly agentic capability embedded within how we build, test, secure, and operate technology,” he said.
Anne Marie Darling, the other co-COO, framed the programme as reworking processes rather than simply buying a tool, all “within a secure and well-governed environment”. For Anthropic, chief commercial officer Paul Smith called the deal “a significant milestone for us in the UK”.
Reading the numbers
This is a vendor announcement, and every figure in it comes from Anthropic and Barclays. The announcement gives no contract value, no cost savings and no measure of whether the email triage or knowledge tool has changed response times or error rates. The 50% figure is an adoption target, not evidence of productivity.
The governance language still matters. In our view, for a regulated UK lender, a public commitment to “robust governance, security controls, and human oversight” is a statement supervisors can hold it to.
Looking forward
The test will come in 2027, when Barclays says most of its engineers should be using Claude Code. Watch for whether the bank reports measurable outcomes, such as faster delivery or fewer defects, rather than headcounts. For smaller UK firms, the knowledge-assistant pattern is the more transferable lesson: in our view, pointing a model at your own documents is a sensible first step towards useful internal AI.