Citigroup is embarking on a high-stakes experiment to assess whether AI agents can enhance productivity across its workforce. Starting this month, the company will roll out a 5,000-person pilot of its newly upgraded internal AI platform, Stylus Workspaces. The initiative is among the first large-scale tests of “agentic” capabilities inside a major financial institution, and reflects growing enterprise interest in tools beyond simple AI chat interfaces.
The pilot will test whether AI agents can independently carry out sophisticated workflows that previously required constant human supervision, such as researching clients across various data sets, building detailed profiles, and translating documents.
David Griffiths, Citi’s Chief Technology Officer, noted that recent improvements in AI tool reliability and integration now make this level of automation viable. The trial is designed to gather insight into user behavior and the cost-benefit dynamics of running such agents at scale.
Why It Matters: As companies like Citi explore autonomous AI agents, they’re laying the groundwork for a new era of digital productivity. If successful, this could redefine workflows and raise important questions about staffing, efficiency, and the evolving role of human workers in knowledge-based industries, particularly financial institutions.
- Transforming Internal Workflows Through Agentic AI: Citi’s new capabilities allow employees to instruct AI agents to perform multi-stage processes like gathering client data and synthesizing it into a coherent profile, all within a single user prompt. Previously, these tasks would have required separate instructions at each stage, consuming time and employee attention. The system’s ability to string together actions autonomously signals a major leap in enterprise automation potential.
- Integration with Leading AI Models to Power Stylus Workspaces: The AI engine behind Stylus Workspaces integrates multiple advanced large language models, including Google’s Gemini and Anthropic’s Claude. By tapping into this blend of capabilities, Citi has built a more flexible and powerful toolset for internal users, capable of adapting to various task types and delivering nuanced results. This multi-model approach also allows for greater experimentation with strengths and weaknesses across different AI systems.
- Cost Controls and the Challenge of Measuring ROI at Scale: One of the key challenges Citi aims to explore during the pilot is balancing functionality with cost. AI agents running long or complex tasks can rapidly consume computational resources, leading to potentially high operating expenses. To manage this, Citi has imposed strict cost limits within the system. However, CTO Griffiths pointed out that as model pricing declines rapidly, traditional return-on-investment models may become outdated quickly, making long-term planning difficult.
- Scope, Duration, and Key Metrics: The pilot will run for four to six weeks and will include 5,000 Citi employees. The company aims to track how workers use the agentic features, how those features influence daily output, and whether the cost-to-value ratio justifies broader implementation. This structured trial format gives Citi a real-world lab to test the effectiveness of AI agents before considering further scaling.
- Potential Impact on Workforce Structure and Productivity: While Griffiths emphasized that it’s too early to predict the effect on jobs, he acknowledged that automating these workflows could bring a “massive boost of capacity.” Over time, this efficiency could change how teams are structured or even how many people are needed to handle specific categories of work.
Go Deeper -> AI Agents Arrive at Citi – WSJ
Trusted insights for technology leaders
Our readers are CIOs, CTOs, and senior IT executives who rely on The National CIO Review for smart, curated takes on the trends shaping the enterprise, from GenAI to cybersecurity and beyond.
Subscribe to our 4x a week newsletter to keep up with the insights that matter.
