Citi says AI is helping it launch products 30% to 40% faster, while use of an agentic coding tool by 14,000 engineers has contributed to a 30% year-on-year reduction in technology-vendor spending. The bank’s account offers a rare look at AI adoption at enterprise scale, with the controls and process changes that make the numbers more than a software demo.
Watch Desk analysis
What happened
Citi CEO Jane Fraser says the bank has rolled AI out to more than 180,000 colleagues across 85 jurisdictions. She says around 100 of the company’s most manual processes have been or are being redesigned using AI and automation.
Fraser says 14,000 engineers use Devin, an agentic coding tool, and that this has made it more efficient for Citi to bring work in-house. The bank says technology-vendor spending fell 30% year on year as a consequence. These figures and outcomes are Citi’s claims, reported in an interview published by Diginomica.
Citi’s Global Head of Services Technology, Amit Rijhsinghani, says the bank buys models but builds its own control plane to monitor how AI operates and assess its outcomes. He also warns that adding AI to fragmented systems or poor data can produce a wrong result faster.
Why it matters
The useful signal is not simply that a large bank is using AI. Citi connects that use to specific operational changes: faster product launches, less reliance on external technology vendors and the redesign of manual processes. Its account also puts governance alongside adoption, rather than treating controls as an awkward extra once the pilot has finished.
The numbers come from Citi’s leadership, not an independent assessment of the savings or productivity gains. Still, they give other organisations concrete claims to test against their own AI programmes, particularly the idea that value comes from redesigning work rather than bolting a model onto it.
Our read
This is a more useful AI story than another announcement that a company has launched a chatbot. Citi is describing an operating model: buy the models, keep accountability and controls in-house, and use the tools to change how work gets done. The results are promising, but the bank’s own figures should be treated as a starting point for scrutiny, not a universal forecast.
What to watch
- Whether Citi provides further detail on how it measures the claimed speed and spending gains.
- How the bank evaluates the quality and risks of work completed with agentic coding tools.
- Whether the redesigned manual processes produce sustained benefits beyond the initial rollout.
- How Citi’s control plane works across different models and business processes.
Discussion spark: Should large organisations buy AI models but keep their controls in-house, as Citi says it does, or is that separation harder to manage than it sounds?
Sources and evidence
- Want to hear more about AI adoption? (26 October 2026, 10:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.