Goldman Sachs builds AI agents with Anthropic to automate core banking work

Goldman Sachs is deepening its push into generative AI by building autonomous agents with artificial intelligence startup Anthropic to automate some of the bank’s most process-heavy work, reported CNBC.

The project focuses on back-office and compliance functions that sit at the heart of daily operations, as large banks look for ways to cut friction, speed up workflows, and manage complexity at scale.

The effort reflects a broader shift across global finance, where AI is moving beyond experimentation into live production systems that reshape how work gets done inside institutions.

Automating accounting and onboarding

For the past six months, Goldman Sachs has worked alongside embedded engineers from Anthropic to co-develop AI agents in two specific areas.

One is trade and transaction accounting, where reconciliation and error resolution can be slow and resource-intensive.

The other is client vetting and onboarding, a compliance-heavy process involving large volumes of documents and strict regulatory checks.

According to Goldman’s chief information officer, Marco Argenti, the agents are based on Anthropic’s Claude model and are still in early development.

The goal is to compress the time required to complete these essential tasks while maintaining accuracy and controls.

Goldman expects to launch the first agents soon, though no timeline has been disclosed.

A digital co-worker model

Inside the bank, the agents are viewed as digital co-workers designed for roles that are complex, scaled, and highly procedural.

Rather than replacing entire jobs, the tools are intended to handle repetitive steps that slow teams down.

As a result, clients should see faster onboarding and quicker resolution of trade breaks and accounting issues once the systems are deployed.

The initiative sits within a wider transformation announced last year by Goldman chief executive David Solomon, who said the bank is pursuing a multi-year reorganisation centred on generative AI.

Even as trading and advisory revenues rise, the strategy aims to constrain headcount growth by increasing productivity through technology.

From coding to compliance

Goldman’s work with Anthropic builds on earlier internal testing of an autonomous AI coding tool known as Devin, which is now broadly available to the bank’s engineers.

That trial revealed that Anthropic’s Claude model could extend beyond software development into other parts of the firm.

Claude proved effective in areas such as accounting and compliance, where systems must read and interpret large datasets and documents while applying rules and judgment.

That capability shifted internal thinking toward broader automation across functions that share similar characteristics.

Wider market implications

The timing is notable.

Recent model updates from Anthropic, co-founded by a former OpenAI executive, have contributed to sharp declines across software stocks and credit markets, as investors reassess which companies stand to benefit from rapid advances in AI.

Generative AI first surged into public awareness after OpenAI released ChatGPT in late 2022, setting off an arms race among model developers and enterprise users.

Looking ahead, Goldman sees potential to extend AI agents into areas such as employee surveillance and the creation of investment banking pitchbooks.

While thousands of staff work in functions likely to be affected, the bank says, it is too early to link the technology directly to job cuts.

Over time, however, reliance on some third-party service providers could decline as in-house AI capabilities mature.