Claude inside the machine: how DXC built 95%-AI-coded enterprise software and then offered to do the same for the world's banks and airlines

On June 11, 2026, Anthropic and DXC Technology announced a multi-year global alliance that will put Claude directly inside the IT infrastructure DXC operates for the world's largest banks, airlines, insurers, manufacturers, and government agencies. 1 The headline is unusual: DXC isn't a startup adopting a new tool. It's one of the planet's largest IT services companies, with 115,000 employees across 70 countries and decades of operational history running mission-critical systems under strict compliance requirements. The announcement is less about a partnership and more about how regulated-industry AI adoption actually works when the stakes are high enough that a failed deployment makes national news.

Before taking Claude to a single client, DXC ran it through its own operations — same security requirements, same compliance constraints, same IT environment its enterprise customers face. The result was DXC OASIS, an AI-native orchestration platform for managed services that launched in April 2026 and now serves over 50 DXC customers. 1

Claude is both the default foundation model powering OASIS's agentic workflows and the tool DXC used to build the platform itself. DXC estimates Claude accelerated software development by 10x, with more than 95% of the code generated by Claude and reviewed by engineers. That figure lands close to the internal numbers Anthropic published earlier this month showing Claude now authors over 80% of Anthropic's own production code — but DXC's context is different. 2 Anthropic is a 500-person AI research company running on greenfield infrastructure; DXC is running 50-year-old legacy stacks for regulated industries where an outage or security breach triggers regulatory investigations. The fact that Claude cleared DXC's own compliance bar before the company bet its clients on it is the actual proof-of-concept.

The core mechanism of the alliance is what Anthropic calls a forward-deployed engineer (FDE) program. DXC will recruit engineers from its existing development teams, certify them through Anthropic Academy (the company's training program for partner engineers), then embed those engineers directly inside client organizations — the same model software infrastructure firms have used for decades to deploy complex systems, now augmented with Claude expertise baked in.

DXC adds its own curriculum on top of the Anthropic Academy baseline, covering the mission-critical system types its clients actually operate: legacy insurance platforms, banking transaction layers, airline operations stacks, government agency infrastructure. The FDE model matters because it solves a distribution problem that off-the-shelf enterprise AI licenses don't: the bottleneck isn't access to the model, it's the domain knowledge to deploy it safely inside environments where a misconfigured agent can process the wrong insurance claims or route aircraft maintenance records incorrectly.

This is also now formalized through DXC's membership in the Claude Partner Network, 1 Anthropic's emerging ecosystem of consulting and services firms. The alliance puts DXC in the same network as other Claude partners but at a scale — 115,000 staff, 70 countries, 50 enterprise customers already on OASIS — that makes it structurally different from a typical SI engagement.

The alliance launches in four areas where DXC already runs large client operations:

Insurance is the first vertical. DXC will use Claude to provide agentic solutions and modernize core insurance systems, working from each customer's specific business context and operating model. Insurance is a natural fit: the core workflow is document-heavy (claims, policies, underwriting decisions), highly regulated, and operationally expensive when handled manually. Claude's document understanding and structured output generation map well onto claims triage and policy review at scale.

Modernization as a Service (MaaS) addresses legacy codebase migration — one of the most expensive and error-prone activities in enterprise IT. DXC is already using Claude to help enterprise customers analyze, refactor, and modernize legacy codebases, with the claim that Claude enables higher speed and accuracy than traditional approaches. The 750,000-line Bun Zig-to-Rust migration Anthropic cited in its Opus 4.8 announcement is the kind of task that falls in this category — the computational work of understanding a large unfamiliar codebase and transforming it systematically is exactly what Claude's extended context and code generation capabilities make tractable.

Cybersecurity is where the deployment becomes most technically specific. For OASIS, DXC is developing an always-on security engineer subagent built on Claude Security, deployed across its security operations centers. The subagent model for security operations — monitoring environments, correlating signals, generating incident summaries — is one of the earlier validated agentic use cases, and DXC's SOC footprint across regulated industries gives it a large attack surface to defend and a large dataset of real security events to train against.

Application services rounds out the four verticals: OASIS agents embedded directly into the application maintenance and management environments DXC operates for enterprise clients. This is the broadest category — essentially any routine IT operations work that can be decomposed into discrete agentic tasks.

The DXC announcement is worth reading against the general pattern of enterprise AI adoption, which has been slower in regulated industries than in tech. The delay isn't skepticism — it's risk management. A bank that deploys a misconfigured AI agent into loan processing or sanctions screening faces regulatory sanctions that can dwarf the cost of the technology. An insurance company that uses an AI to make coverage decisions without compliant audit trails faces legal exposure.

DXC's model — run it on yourself first, certify the engineers, embed them in the client, stay within the compliance perimeter the client already operates inside — is a slower path than a cloud API subscription, but it's the only path that works at the scale of the clients DXC serves. Raul Fernandez, DXC's President and CEO, framed it plainly: "We know what it takes to deliver in these environments." 1

The 95% Claude-generated-code figure is the strongest signal in the announcement. It means DXC has already operationalized the recursive productivity loop Anthropic documented internally — not as a research finding but as the production basis for a platform its clients are actively using. The question that remains open is how much of that efficiency carries through to client deployments, where the model operates inside organizations with legacy constraints DXC's own infrastructure doesn't face. That's what the FDE program is designed to measure — one embedded engineer at a time.