Banking titan Citi is rolling out a major test of so-called “agentic” artificial intelligence, betting the technology could eventually streamline work once thought to require human supervision.
The bank will run a pilot involving 5,000 employees over the next month, using its in-house AI platform to see how far agent models can perform certain tasks, reports The Wall Street Journal.
David Griffiths, Citi’s chief technology officer, says early agent models often struggled to work reliably, but improvements now make a broader test possible.
“A couple of years ago, you could do agentic things with the early versions of the models that were available then. But they weren’t always very reliable. They weren’t always very good at invoking tools. But they are now.”
The internal platform, branded Citi Stylus Workspaces, is powered by a mix of models, including Google’s Gemini and Anthropic’s Claude. Staff can prompt the system once and have it chain together multiple actions across company systems, from building a client profile out of internal and public data to translating it into another language.
“These things were possible before the latest update, but they would have required separate direction from the human at each point in the process. That’s just one example, and you can see how many thousands of examples of that kind of problem we would have in the company.”
The pilot will run for four to six weeks, with Griffiths looking for evidence on how staff use the tools, how impactful they are, and whether the economics of the technology justify broader rollout.
Citi is not the first banking giant to look at AI for productivity gains. Last month, reports emerged that Bank of America’s AI assistant Erica has topped three billion client interactions, underscoring the bank’s heavy reliance on automation while reducing demand for human service.