Barclays bets on Claude Code for half its developers by year-end, after RAG and email triage

Barclays is widening its use of Anthropic's Claude across the bank, the two companies said on October 1. The headline number is a target, not a result: Barclays expects Claude Code to be used by 50% of its developers by the end of 2026, and by a majority of its software engineers in 2027. The announcement (Anthropic) pairs that forecast with two deployments the bank says are already running.

What is already live

The first is the Colleague Knowledge Assistant, an internal tool that helps Barclays UK staff look up answers while serving the bank's more than 20 million UK retail customers. It has been live since 2025 and is built on Claude through a retrieval-augmented generation (RAG) architecture. According to the announcement, more than 16,000 colleagues have adopted it and it has handled over one million searches.

The second sits in Global Markets. There, teams use Claude models to classify incoming client emails, enrich them, and decide the processing route each one should take, checking that requests contain what operations staff need before they act. The companies say the platform processes about 120,000 emails a day and reduces manual handling. No error rates, accuracy figures or before-and-after timings were published for either system.

What the expansion targets

The new phase is aimed at engineering: accelerating software development, modernizing legacy systems and improving operational efficiency. Craig Bright, Barclays' Group Co-Chief Operating Officer, framed it as a shift toward "AI as an increasingly agentic capability embedded within how we build, test, secure, and operate technology," adding that the bank is using Claude to help modernize legacy platforms and improve software quality. He also tied the move to security: "Software engineering and cyber security are both being reshaped by increasingly capable AI systems."

Anne Marie Darling, the bank's other Group Co-COO, described the goal as "building a simpler, more efficient organization within a secure and well-governed environment." Anthropic's Chief Commercial Officer Paul Smith called the deal "a significant milestone for us in the UK."

What was not disclosed

• Financial terms, contract length or spend.
• How many developers Barclays employs, so the 50% target cannot be converted into a seat count.
• Which Claude models are used, or whether access runs through Anthropic directly or a cloud provider.
• Which legacy platforms are being modernized, or any measured productivity effect from Claude Code so far.

The announcement says Barclays applies governance, security controls and human oversight to its AI use cases, but it does not describe those controls in detail.

How to read it

Two kinds of numbers are mixed in this announcement. The 16,000 users, one million searches and 120,000 daily emails describe systems in production, as reported by the companies themselves; they have not been independently audited. The 50% and "majority" figures for Claude Code are adoption goals. Adoption is also not the same as impact: a developer counted as using a coding agent may use it daily or occasionally, and the announcement does not define the metric.

Even so, the shape of the rollout is informative for teams in regulated industries. Barclays started with bounded, read-heavy uses (internal search over a curated knowledge base, and triage of inbound email) before setting broad targets for a coding agent that touches production software. That sequencing, plus an explicit link to legacy modernization and cyber security, is a pattern other large banks are likely to be compared against.

Sourcing note: this article is based on the joint announcement published by Anthropic, which includes on-the-record statements from two Barclays executives. Independent reporting existed at the time of writing but could not be accessed for verification, so all figures here are attributed to the companies.