Barclays is expanding its use of Anthropic's Claude models into software development, legacy system modernization and operations, and expects half of its developers to be using the Claude Code agent by the end of 2026, Anthropic said Thursday. The bank aims to have a majority of its engineers on the tool in 2027.

What Happened

The announcement builds on deployments Barclays already runs. Its Colleague Knowledge Assistant, which has been live since 2025, is used by more than 16,000 staff in the U.K. who serve the bank's more than 20 million retail customers there. According to Anthropic, the assistant has handled more than 1 million searches, helping employees find internal information while dealing with customers.

In the Global Markets business, Claude classifies, enriches and routes about 120,000 emails a day, the company said. The next phase extends that footprint into engineering, where Claude Code will be used to speed up development work and to modernize older platforms.

RelatedAnthropic Prices Claude Haiku 5.5 at a Tenth of Its Predecessor for Most API Requests

Barclays' two group co-chief operating officers framed the rollout in operational terms. Anne Marie Darling said the goal is to strip out routine work so teams can concentrate their expertise where it counts. Craig Bright said the bank's work with Claude shows AI can help modernize legacy platforms and let technical specialists spend their time on harder problems. Paul Smith, Anthropic's chief commercial officer, described the deal as "a significant milestone" for the company in the U.K. and credited Barclays with applying strong security and oversight standards to its AI use.

Anthropic did not disclose financial terms, the number of developers involved or how Barclays will measure the results.

Why It Matters

The developer targets are the notable part of the announcement. Large banks have spent the past two years testing generative AI through chat assistants and narrow pilots, often in customer service or document search. Committing to put a coding agent in the hands of half of a bank's developers within roughly three months, and most of its engineers the following year, is a move from experimentation to standard tooling.

Legacy modernization is a large prize for lenders. Banks run core systems that have been extended over decades, and rewriting or documenting that code is slow, expensive and risky. Coding agents that can read and translate older codebases are being pitched as a way to cut that cost, and Bright's comments name legacy modernization as one aim of the expansion.

The numbers on existing tools also give a rare view of usage at scale inside a regulated bank. A knowledge assistant with more than 16,000 users and an email triage system handling six figures of messages daily are production workloads, not demonstrations.

Regulated Deployment

Both sides emphasized controls. Smith's praise for Barclays' security and oversight standards reflects the scrutiny that U.K. financial regulators apply to technology used in customer-facing work and trading operations. For a bank, letting an AI agent write and change code raises questions about review processes, access to production systems and accountability for errors. Anthropic's announcement set adoption targets in stages but did not describe how Barclays will review or approve code written by the agent.

What to Watch

The main markers are whether Barclays reaches the 50% developer target by Dec. 31 and a majority of engineers in 2027, and whether it reports measurable results from the legacy modernization work, such as systems retired or migration timelines shortened. Further disclosures on how the bank governs agent-written code would also be closely read by other lenders weighing similar rollouts.