American Banker reports that Goldman Sachs has laid out plans to use generative AI, specifically Anthropic's Claude model, in trade accounting and client onboarding, and that engineers at the bank have built agents for both. Chief information officer Marco Argenti said that with AI agents and humans working together, "now you can do it in a way that scales so the same group of people can handle five to ten times the cases, or in less time, because a lot of this machinery of work is actually automated."
Trade accounting and know-your-customer compliance are rules-based processes, but exceptions remain. Argenti said the best rules-based systems can automate a high percentage of this work, but if a bank is processing hundreds of millions of transactions per day and even 1% doesn't fit neatly within the rules, that creates thousands of exceptions. In one know-your-customer step, an applicant's government-issued ID might be close to expiring or show some other discrepancy. "The beauty of neural networks is the fact that they can reason like a human in those micro use cases, and therefore they complement a rules-based system into something that actually can get you really close to 100%," Argenti said.
Two years ago, Goldman's developers began using a version of Claude with Cognition's Devin agent. "That gave us the confidence that the reasoning capabilities of Claude were extremely effective in not only suggesting how to code to our developers, but also in executing autonomous tasks," Argenti said. To create the new agents for trade accounting and client onboarding, engineers at Goldman and Anthropic spent time with domain experts in the bank to rethink the processes. They observed how people use their computers, how the processes work in practice and what the bottlenecks are.
According to Argenti, the agents review documents, extract entities, determine whether another document needs to be requested, and recognise cases where an applicant and a spouse both own a company, which requires a separate KYC check because they are co-owners. He described these as micro decisions within boundaries that are based on reasoning rather than rules. Goldman keeps a human in the loop, as the article notes.
Argenti plans to continue to bring AI agents into the back office, including vendor management, procurement and lending, where manual processes include analyzing agreements that are sometimes hundreds of pages long and turning them into formats for an investment committee or a credit committee. "But the biggest unlock is going to be on sales enablement," he said. He added that many proofs of concept are in the works throughout Goldman.