Goldman Sachs taps Anthropic’s Claude to automate accounting

Goldman Sachs The firm is working with AI startup Anthropic to create AI agents to automate a growing number of roles within the bank, its technology chief told CNBC exclusively.

According to Marco Argenti, Goldman’s chief information officer, the bank has been working with resident Anthropic engineers for the past six months to develop autonomous brokers in at least two specific areas: accounting for trades and transactions, and client review and onboarding.

Argenti said the firm is in the “early stages” of developing agents based on Anthropic’s Claude model that will reduce the time taken by these core functions. He expects agents to be operational “soon,” although he declined to give a specific date.

“Think of it as a digital colleague for many professions within the firm that are scaled, complex and process-intensive,” he said.

Goldman Sachs CEO David Solomon said in October that his bank was launching a multi-year plan to reorganize itself around generative AI, the technology that has been making waves since the emergence of OpenAI’s ChatGPT in late 2022. Investment banks such as Goldman will try to “limit headcount growth” amid this overhaul, even as their revenues from trading and advisory activities increase, Solomon said.

Goldman’s news comes as model updates from Anthropic, co-founded by a former OpenAI executive, sparked a sharp selloff among software firms and lenders as investors bet on who the winners and losers from AI trading would be.

Goldman started last year by testing an autonomous AI coder called Devin; this encoder is now widely available to bank engineers. But Argenti said Anthropic quickly found that its AI model could work in other parts of the bank as well.

“Claude is really good at coding,” Argenti said. “Is this because the coding is a bit proprietary, or does it have to do with the model’s ability to reason through complex problems by applying logic step by step?”

Argenti said the firm was “surprised” at how skilled Claude was at coding, as well as tasks in areas such as accounting and compliance, which combine the need to parse large amounts of data and documents, especially when applying rules and judgments.

Now Goldman’s view is that “there are other areas of the firm where we can expect the same level of automation and the same level of results as we see on the coding side,” he said.

Ultimately, Argenti said, with the help of agencies in the development phase, customers will be onboarded more quickly and issues with trade reconciliation or other accounting issues will be resolved more quickly.

Goldman said it may next develop representatives for tasks such as employee supervision or preparing investment banking white papers.

While the bank employs thousands of people in compliance and accounting functions, where AI representatives will soon operate, Argenti said it was “premature” to expect the technology to lead to job losses for those employees.

Still, Goldman said it may opt out of the third-party providers it uses today as AI technology matures.

“It’s always a trade-off,” Argenti said. “Our philosophy now is that we inject capacity in many cases that will allow us to do things faster, which means a better customer experience and more business.”