Barclays is expanding its use of Anthropic's Claude across the bank, moving beyond small AI experiments into software development, employee support and operational workflows.
Anthropic says Barclays expects Claude Code adoption to reach 50% of its developer population by the end of 2026 and a majority of software engineers in 2027.
What Barclays is changing
The bank is expanding Claude across global operations with a focus on three broad areas: software development, knowledge assistance and operational processing.
The aim is not simply to give employees another chatbot. Barclays is using Claude inside specific workflows where the bank can measure the effect of AI on work.
Claude Code for software development
Barclays plans to expand Claude Code across its developer population.
The bank expects half of its developers to use Claude Code by the end of 2026, with a majority of software engineers expected to use it in 2027.
The bank says it is using Claude to modernize legacy platforms, improve software quality and help technical teams focus on harder engineering problems.
This is a useful example of how coding agents can be introduced into a large organization. Instead of replacing software teams, the stated goal is to reduce routine work and help engineers spend more time on complex tasks.
Barclays already uses Claude for employee knowledge
One of the existing deployments is the Barclays Colleague Knowledge Assistant.
The assistant uses Claude with a retrieval-augmented generation architecture. It helps Barclays employees find information when supporting customers.
Anthropic says more than 16,000 Barclays colleagues have adopted the system and that it has handled more than one million searches.
The bank supports more than 20 million UK retail customers, so faster access to internal information can have a direct operational effect.
How the knowledge assistant works
A retrieval-augmented system does not need to rely only on what the language model remembers from training.
Instead, the system can retrieve relevant information from approved internal sources and give that information to the model when answering a request.
For a bank, this approach can be useful because internal policies, procedures and product information change over time.
The retrieval layer can also help limit the model to information that the organization has approved for the specific workflow.
Claude processes around 120,000 emails each day
Barclays is also using Claude models in its Global Markets business to process incoming client emails.
Anthropic says the system processes approximately 120,000 emails each day.
The models help classify and enrich incoming messages and determine the appropriate processing route.
The goal is to reduce manual handling and make sure operational teams receive the information they need to act.
Why email processing is a useful AI use case
Email is often a good target for enterprise automation because large organizations receive high volumes of repetitive requests.
A model can classify messages, identify missing information and route requests before a human needs to handle the full process.
But this does not mean the model should make every decision without review.
For regulated industries, organizations need to define which decisions can be automated and which ones require human approval.
Governance is part of the deployment
Barclays and Anthropic emphasize security controls, governance and human oversight as part of the expansion.
This matters because financial institutions operate with strict requirements around customer information, operational risk and system access.
A useful enterprise AI system therefore needs more than a capable model.
It needs clear permissions, logging, testing, monitoring and defined responsibilities when something goes wrong.
What this means for enterprise AI teams
Barclays' deployment shows a pattern that other large companies can follow.
Start with workflows where:
- the input volume is high
- the process is repetitive
- the expected output can be checked
- the business impact can be measured
- human review can be added where necessary
Knowledge retrieval, email classification and coding assistance fit these conditions better than vague attempts to give an AI system control over an entire department.
Why this is different from a simple chatbot rollout
A chatbot answers questions when someone asks.
The Barclays examples go further. Claude is being connected to business workflows where information needs to be retrieved, classified, enriched or transformed.
That makes the surrounding system just as important as the model.
The retrieval layer, access controls, workflow logic and human review determine whether the deployment is useful and safe.
The bigger lesson
Barclays is treating AI as part of its operating systems rather than as a standalone productivity app.
The most important part of the rollout may not be the model itself. It is the process of connecting the model to real work while keeping governance around it.
For companies planning their own AI rollout, that is the part worth copying.