Citigroup is currently running the largest measured enterprise agentic AI deployment in the financial sector, moving beyond isolated chatbots to treat AI agents as a centralized operating system. With the launch of Arc in April 2026, the bank has shifted its focus from generative capabilities to the governance and scaling of agentic workflows across a global institution. This transition marks a departure from experimental pilots toward a model where AI is a core, funded, and operational component of the bank’s infrastructure.
The scale of this deployment is substantial. Citi is leveraging a workforce of 180,000 staff across 85 countries, all utilizing AI tools. Central to this effort is the integration of Cognition’s Devin, with 40,000 developers now utilizing the platform for agentic coding tasks. According to data shared during the May 7, 2026 Investor Day, this ecosystem generates over 100,000 agentic AI development hours per week. This represents a massive shift in operational capacity, effectively saving 50 developer-years of effort every single week.
For enterprise technology leaders, the most compelling metric is the tangible impact on legacy infrastructure. Citi reports that the time required for legacy system migration has plummeted from 12 months to just four weeks. This acceleration is underpinned by a 30-40% boost in developer productivity. When applied to the firm’s broader service operations, the efficiency gains are equally stark: service agents now handle over 3 million inquiries annually, contributing to a 25% reduction in overall servicing effort. As David Griffiths, Citi’s CTO, noted, “For the first time, we can deploy embedded AI agents at enterprise scale across every business line, every geography, every function.” The financial model supporting this transition is as notable as the technology itself. Citi has committed $5 billion to this AI transformation, but the investment is largely self-funded through structural efficiency savings generated by the agents themselves. This creates a mechanism where the deployment of AI provides the capital necessary to further expand the agentic infrastructure. This approach contrasts with the more fragmented strategies seen elsewhere in the industry. While JPMorganChase has reported 240,000 employees using generative AI, and Goldman Sachs maintains a developer base of 12,000, Citi’s Arc platform distinguishes itself by enforcing a unified risk framework. Every agent built on Arc is monitored, auditable, and governed, ensuring that the speed of deployment does not outpace the bank’s regulatory and risk management requirements.
The platform’s versatility is evident in its application across the firm. Beyond developer productivity, Citi has deployed specialized tools like Citi Sky, an AI-powered wealth management assistant developed in collaboration with Google Cloud and Google DeepMind. Furthermore, the firm has already identified over 50 distinct use cases within its services division, ranging from research and synthesis to complex task preparation and execution.
However, the transition to an agentic enterprise is not without its operational boundaries. While tools like Cognition’s Devin can perform specific tasks two to 20 times faster than traditional methods, they remain firmly under human supervision. The current deployment model explicitly requires human review for all outputs, and there is no autonomous deployment of code or financial decisions. This caveat is essential for investors and regulators alike: the “agentic” nature of the platform refers to the ability of the software to execute complex, multi-step workflows, not to the removal of human accountability from the decision-making loop.
By standardizing the development and governance of these agents, Citi has moved the needle on how large-scale financial institutions integrate AI. The shift from treating AI as a collection of disparate tools to an integrated operating system is now a measurable reality, providing a clear financial and operational benchmark for the rest of the industry to evaluate.