Summary: Anthropic and PwC announced an expanded alliance on May 14, 2026 to deploy Claude Code and Claude Cowork across PwC, including an Office of the CFO practice and regulated-industry workflows. Reported production wins include underwriting falling from ten weeks to ten days, but the figures come from the announcement and clients, while human oversight details remain only partially disclosed.
From Pilot Projects to Live Production
May 14, 2026 is when Anthropic and PwC announced they were widening their strategic alliance, a move that pushes Claude out of the experimentation phase and into the center of how PwC operates worldwide. According to the announcement, Claude Code and Claude Cowork are being rolled out across PwC’s workforce, beginning with teams in the U.S. and fanning out toward hundreds of thousands of professionals around the world. Nothing about this is tentative: the press release names production systems already running in insurance underwriting, mainframe modernization, HR transformation, and cybersecurity, and clients themselves report delivery gains as high as 70%.
What product builders should take from this is not that one more large firm has put AI to work. What matters is the demonstration of how agentic systems can be woven into sectors that treat accuracy and auditability as absolute requirements. The marquee figures—underwriting cycles cut from 10 weeks to 10 days, security incident response down from hours to minutes—describe more than faster execution. Behind them sits a redesign of the workflows themselves, of the compliance machinery around them, and of which business activities make economic sense at all.
The Three Areas Doing the Heavy Lifting
Three areas anchor the collaboration, and each one maps to a different source of enterprise pain:
- Agentic technology build: With Claude Code, engineering teams are getting production-grade software into the hands of major companies in weeks instead of quarters. A growing set of these agentic builds spans financial services, pharma and life sciences, healthcare, and consumer markets.
- AI-native deal-making: Agents now sit beside the deal teams across the whole life of a transaction—diligence, value creation, integration—as PwC reworks how deals get executed. Private equity sponsors and corporate acquirers get a shorter road from investment thesis to captured value, and the arithmetic of which pursuits are worth the effort shifts with it.
- Reinvention of the enterprise function: Instead of parking efforts in pilots the way most organizations do, PwC is running live operations—including scalable AI-native operating models for finance, supply chain, HR, and the engineering function itself.
In the announcement, Anthropic CEO Dario Amodei made the point that the largest returns will come from AI systems that carry tasks through end-to-end, keep running without interruption, and give the most seasoned professionals reach at a scale that was never available before. That idea sits at the heart of PwC’s approach: the goal is not to automate tasks one at a time, but to rebuild whole processes.
Office of the CFO: A Standalone Unit Running on Anthropic Technology
The launch drawing the most attention is a new finance business group at PwC called the Office of the CFO—presented as the first standalone business unit to have Anthropic’s technology as its foundation. The practice combines PwC’s finance expertise with the complete Anthropic product lineup: Claude, Claude Cowork, and Claude Code. Work begins in regulated industries—banking, insurance, healthcare—because those fields hold accuracy and auditability to the strictest standard. Engagements run from focused support on individual finance tasks all the way up to redesigning the entire function from the ground up.
Before any client saw the offering, PwC signed on as “Customer Zero,” running Claude inside its own operations first. The firm’s teams handled journal entries, variance analysis, and RFPs with it, and leaned on Claude Code to sharpen annual planning. In parallel, PwC has been supporting Anthropic’s own CFO office as it scales operations, controls, and international payroll. Proving the technology internally before taking it to market is about as practical a route as enterprise AI deployment offers.
Deployments in Production: The Numbers Behind the Claims
Specific production deployments, each with quantified outcomes, appear in the press release:
- Professional sports operations: Drawing on Anthropic’s portfolio of assets, PwC has reworked digital fan engagement and an agentic-first sports management operation.
- Insurance underwriting: Underwriting now takes ten days where it once took ten weeks, and lines of business that never made economic sense before have opened up.
- Mainframe modernization: The COBOL codebase turned out to be four times larger than scoped, yet the project is tracking on time and under budget.
- HR transformation: A program that had stalled was revived—one week produced a working prototype, the full application shipped in under two months, and it now handles thousands of transactions every day.
- Cybersecurity: Incident response now runs in minutes rather than hours, while agentic vulnerability operations—code review, automated containment—shut exposure windows before adversaries get the chance to exploit them.
Claude Code’s lead in enterprise coding underpins these outcomes, and PwC’s Claude-native modernization practice keeps investing in it for its frontier performance and compliance readiness. The results reach across healthcare, life sciences, financial services, and consumer markets, with PwC helping clients shape the operating models they will run in the future.
Training and Ecosystem: Carrying Claude to Hundreds of Thousands
Running agentic operating models at this scale takes people equipped to engineer, run, and govern them. Client demand is accelerating, and PwC’s answer is to extend Claude across a professional workforce numbering in the hundreds of thousands. Two structures anchor the work: a joint Center of Excellence, plus an effort that will train and certify 30,000 US professionals. Leadership is included as well—earlier this year, PwC’s Advisory Leadership Exchange gave more than 5,000 leaders a close look at the alliance, and hands-on training turned a wave of early adopters loose on client workflows with Claude Code.
ChatPwC, the firm’s internal AI assistant, already offers Claude, and the technology is running in production on client engagements through three active AI incubation pods covering Finance, Supply Chain, and Deal Making. Claude Cowork is set to push that reach out to the broader workforce by living inside tools such as spreadsheet, word processing, and presentation programs, and by linking into enterprise data through Anthropic’s Model Context Protocol (MCP). The strategy amounts to penetration at many points at once: from coding tools down to everyday office software, AI ends up everywhere in the organization.
Of everything under Anthropic’s Claude Partner Network—$100 million committed this year to back services firms that get enterprises deploying AI rather than merely piloting it—this alliance stands as the deepest. What lets PwC operate at the scale these clients require is its depth in financial services, healthcare, life sciences, and cybersecurity.
Caveats and Trade-offs
The numbers look strong, but the announcement stays quiet about the difficulties encountered along the way. The mainframe modernization effort, for one, ran into a COBOL codebase four times the size originally scoped—a vivid illustration of how much more complicated legacy systems can be than anyone expects. Cutting underwriting from 10 weeks to 10 days is a dramatic result, yet reaching it almost certainly demanded heavy process redesign and compliance validation rather than simply deploying a model. And the 70% delivery improvement is a cross-deployment average; results for any single engagement may differ.
Reaching hundreds of thousands of users also demands an enormous training and certification apparatus—preparing 30,000 US professionals is a substantial investment in its own right. Whether the effort succeeds rests on PwC’s capacity to govern agentic systems inside regulated industries, where accuracy and auditability come before everything else. Risks such as model errors, data privacy, and the human oversight that high-stakes decisions demand go unaddressed in the press release. Product builders should remember that these are production systems that operate with human-in-the-loop safeguards, though exactly what those safeguards involve is not disclosed in full.
The Takeaway: A Playbook for Getting AI into Production
For AI practitioners and product builders alike, PwC’s method amounts to a playbook that can be copied:
- Pick scenarios with high leverage: Target areas such as underwriting, cybersecurity, or HR, where the pain runs deep and the impact can be measured.
- Put a number on the baseline: PwC knew its 10-week underwriting figure and aimed at 10 days. Without that starting number, success cannot be measured.
- Act as Customer Zero: Run the technology inside your own operations before taking it to market—PwC ran its own finance on Claude before anything went to a client.
- Fund training and governance: The joint Center of Excellence and the certification program carry as much weight as the technology does.
- Plug into the tools people already use: Claude Cowork showing up inside documents and spreadsheets, together with MCP connectivity, lowers the barrier to adoption.
The binding constraint for most enterprises is not how capable the models are; it is the work of fitting agentic systems into the workflows and compliance structures already in place. The alliance between PwC and Anthropic shows that, given a sound strategy, AI can make the jump from pilot to production and produce results you can point to. Anyone planning an enterprise AI rollout can start the same way: name one pain point, define explicit KPIs, and take PwC’s lead—let your own team be the first customer.