TL;DR

Goldman Sachs has been working with Anthropic for six months to develop autonomous AI agents for trade accounting and client vetting. The bank’s CIO describes the agents as “digital co-workers” for complex, process-intensive tasks, with a launch expected soon.

From Code to Compliance

Goldman Sachs initially used Anthropic’s Claude model for coding tasks, but chief information officer Marco Argenti told CNBC the bank was “surprised” at how effective the model proved in other areas. That discovery led to a deeper partnership, with embedded Anthropic engineers working alongside Goldman staff to build agents for two specific use cases: accounting for trades and transactions, and client vetting and onboarding.

“Think of it as a digital co-worker for many of the professions within the firm that are scaled, are complex and very process intensive,” Argenti said. The agents are expected to launch soon.

The bank now believes it can achieve “the same level of automation and the same level of results” across accounting and compliance as it has seen on the coding side.

Broader AI Transformation

The AI agent work is part of a wider reorganisation. CEO David Solomon has described a multiyear plan to restructure the bank around generative AI. Argenti acknowledged this raises questions about human roles but framed the current approach as additive: “Our philosophy right now is that we’re injecting capacity, which in most cases will allow us to do things faster, which translates to a better client experience and more business.”

Last year, Argenti noted the challenge of retooling an entire organisation for AI while managing change responsibly. “We have the entire organisation that needs to somehow re-tune and re-tool itself for AI,” he said.

Looking Forward

Goldman’s approach — embedding AI vendor engineers directly into business teams — offers a model that other financial institutions may follow. The focus on accounting and compliance, rather than client-facing trading, suggests the bank is targeting back-office processes where accuracy and consistency matter most. If the agents perform as expected, expect rapid expansion into other operational areas.