Citi's Arc: the largest measured agent deployment

Citi is running the largest measured enterprise agentic AI deployment in financial services. Since the Arc platform launched in April 2026, the bank has shifted from a race over generative capability to governing and scaling agentic workflows — treating AI agents as a centralized operating system rather than a collection of isolated chatbots.

Scale: running like an operating system

Citi has a workforce of 180,000 staff across 85 countries, all using AI tools. Central to the effort is Cognition's Devin: 40,000 developers now use the platform for agentic coding tasks. According to data shared at the May 7, 2026 Investor Day, the ecosystem generates over 100,000 agentic AI development hours per week — equivalent to saving 50 developer-years of effort every single week.

For enterprise technology leaders, the most striking metric is on legacy infrastructure: the time to migrate legacy systems has fallen from 12 months to just four weeks, underpinned by a 30-40% boost in developer productivity. Service operations show similar gains: service agents handle over 3 million inquiries annually, contributing to a 25% reduction in overall servicing effort.

The financial model: $5 billion, self-funded by the agents

The financial model behind this transition is as notable as the technology. Citi has committed $5 billion to the AI transformation, but the investment is largely self-funded through the structural efficiency savings the agents themselves generate. The savings from deployment provide the capital to expand the agentic infrastructure — AI expansion no longer depends on continuous out-of-budget funding.

This contrasts with the more fragmented strategies elsewhere. JPMorgan reports 240,000 employees using generative AI, and Goldman maintains a developer base of 12,000. Arc's distinction is the unified risk framework it enforces: every agent built on Arc is monitored, auditable and governed, ensuring deployment speed never outpaces the bank's regulatory and risk management requirements.

Breadth: from wealth management to the service middle office

The platform's portability shows on the business side. Beyond developer productivity, Citi has deployed Citi Sky, an AI-powered wealth management assistant built with Google Cloud and Google DeepMind. The services division has identified over 50 use cases, ranging from research and synthesis to complex task preparation and execution.

As Citi CTO David Griffiths put it: \"For the first time, we can deploy embedded AI agents at enterprise scale across every business line, every geography, every function.\" The weight of that sentence is not in the wording — it is in the measured numbers behind it.

The boundary: 20x faster, but humans stay in the loop

The transition has operational boundaries. Devin can perform specific tasks two to 20 times faster than traditional methods, but it remains firmly under human supervision. The current deployment model explicitly requires human review of all outputs, and there is no autonomous deployment of code or financial decisions.

This caveat matters for investors and regulators alike: the platform's \"agentic\" nature refers to its ability to execute complex, multi-step workflows — not to the removal of human accountability from the decision-making loop. The line between speed and responsibility is drawn explicitly.

The bottom line

Citi has proven something beyond \"agents can write code\" — that was never news. It has proven that when agents are governed as an operating system, bound by a unified risk framework and funded by their own efficiency savings, scale can satisfy speed and regulation at the same time. For enterprises scaling agents, this is a financial and operational benchmark to measure against: unified governance, human review, self-funded efficiency — miss any one, and the scaling story does not close.

References

• Forkast: Citigroup's Arc Platform Is the Largest Measured Enterprise Agent Deployment Yet — https://forkast.news/citigroups-arc-platform-is-the-largest-measured-enterprise-agent-deployment-yet/
• CNBC: JPMorgan Chase AI agents (2026-06-09) — https://www.cnbc.com/2026/06/09/jpmorgan-chase-ai-agents.html
• CNBC: Goldman Sachs autonomous coder pilot (2025-07-11) — https://www.cnbc.com/2025/07/11/goldman-sachs-autonomous-coder-pilot-marks-major-ai-milestone.html