Barclays is expanding its use of Anthropic’s Claude, with Claude Code expected to reach 50 percent of its developers by the end of 2026 and a majority of its software engineers in 2027, according to Finextra, which reported the Barclays Claude plans on 2 October 2026. Fintechbits did not find a primary release from Barclays or Anthropic, so the figures here come from that report.

What the Barclays Claude plan includes

The headline target concerns Claude Code, Anthropic‘s coding tool. Finextra reports that Barclays expects half of its developers to use it by the end of 2026 and most of its software engineers to use it in 2027.

The bank has also described AI as a route to roughly 2 billion pounds of efficiency savings, a figure it gave earlier. The Barclays Claude programme is one of the ways it hopes to reach that number, though the report does not break the savings down by tool.

Two internal systems show how far the adoption has gone. A colleague knowledge assistant, built on retrieval-augmented generation and live since 2025, has more than 16,000 staff users and has handled more than 1 million searches. In global markets, Claude classifies, enriches and routes about 120,000 emails a day. Anne Marie Darling, group co-chief operating officer, is quoted in the report.

Fintechbits analysis

Why the Barclays Claude figures stand out

Many bank AI announcements describe pilots. These describe volumes. Sixteen thousand staff using an assistant and a trading division sending 120,000 emails a day through a model are production workloads, and they imply the bank has cleared the security and governance steps that slow most deployments.

The email example is a good test case for the Barclays Claude programme. Classifying and routing messages is repetitive and measurable, and a mistake is usually caught downstream. That makes it a safer starting point than a customer-facing decision, and it explains why markets operations went early.

The developer target is the larger bet in the Barclays Claude plan, and it is the part that changes how the bank works. Coding tools change how software gets built, and a bank with thousands of engineers has a lot of code to write and maintain. A target of half its developers within months is aggressive for a regulated institution, and the report does not say how the bank measures productivity or code quality.

What the Barclays Claude rollout asks of risk teams

Putting an AI coding tool in front of half the developers raises questions that the report does not answer. Who reviews generated code, how is it tested, and how does the bank keep proprietary data out of prompts? Banks answer these questions through controls such as code review, testing gates and restrictions on what data can enter a prompt, and the controls will determine whether the 2 billion pound target is reached.

Fintechbits has covered the other side of bank AI, where models are used to catch crime, in AI fraud detection in financial crime, and the risks that arise when software agents act on payments in the agentic payments verification gap. Barclays Claude adoption sits between them. It is internal use, with employees in the loop, which carries less risk than autonomous payments but more than a search tool.

What the Barclays Claude report does not say

The report does not give contract terms, the number of developers at the bank, or the cost of the licences. It does not say whether Barclays uses other model providers alongside Claude, which would be typical for a bank wary of depending on one supplier.

There is also no independent measurement of savings, and no breakdown of how many of the 16,000 knowledge assistant users are engineers rather than operations staff. The 2 billion pound figure is a bank estimate and the report does not tie any part of it to Claude.

What to Watch Next for Barclays Claude

Watch for results commentary from Barclays that links AI spending to cost lines, since the 2 billion pound target is only meaningful if it shows up in reported expenses. Watch also for whether the developer adoption target of 50 percent is met by the end of 2026, and whether regulators comment on the use of AI coding tools in banks. A primary statement from Barclays or Anthropic would settle the details this report leaves open.