Goldman Sachs is accelerating its bet on artificial intelligence, partnering with AI startup Anthropic to build autonomous agents that could fundamentally reshape how core banking functions are performed, from trade accounting to client onboarding.
For the past six months, the Wall Street giant has been working closely with embedded Anthropic engineers to co-develop AI agents powered by Anthropic’s Claude model. The initial focus is on two operationally intensive areas: accounting for trades and transactions, and client vetting and onboarding, according to Goldman’s chief information officer, Marco Argenti.
The agents are still in development, but Argenti said the bank expects to deploy them soon. While he declined to give a specific launch date, the direction is clear: Goldman is no longer treating generative AI as an experimental add-on but as a core pillar of its operating model.
Register for Tekedia Mini-MBA edition 19 (Feb 9 – May 2, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab.
Argenti described the technology as a “digital co-worker” designed to operate alongside humans in roles that are complex, repetitive, and highly scaled. These are functions that sit at the heart of a modern investment bank and traditionally require large teams to manage regulatory requirements, documentation, reconciliations, and approvals.
The initiative fits into a broader transformation outlined by Goldman Sachs CEO David Solomon last year. In October, Solomon said the bank had embarked on a multi-year plan to reorganize itself around generative AI. Even as Goldman benefits from strong revenues in trading and advisory businesses, Solomon said the firm would seek to constrain headcount growth as AI-driven productivity gains take hold.
At Goldman, the move into autonomous agents builds on earlier experiments with AI-assisted coding. Last year, the bank began testing an autonomous coding tool known as Devin, which is now widely available to its engineers. That project served as a proving ground, demonstrating that advanced models could reliably handle complex tasks within a highly regulated environment.
What surprised Goldman’s technology leadership was how quickly those capabilities translated beyond software development. Argenti said Claude’s strength is not limited to writing code, but lies in its ability to reason through complex problems step by step, applying logic across large volumes of data and documents.
That capability is particularly valuable in areas like accounting and compliance, where staff must interpret rules, reconcile discrepancies, and make judgments based on incomplete information. In Argenti’s words, technology teams realized that “there are these other areas of the firm where we could expect the same level of automation and the same level of results that we’re seeing on the coding side.”
The potential operational impact is significant as client onboarding, often slowed by manual checks, document reviews, and regulatory approvals, could be completed much faster. Trade reconciliation issues, which can take days to resolve, could be identified and fixed more quickly, reducing operational risk and improving client experience.
Goldman is also exploring the use of AI agents in other parts of the business. Argenti pointed to possibilities such as employee surveillance and the creation of investment banking pitchbooks, both of which require processing large amounts of information under tight timelines. While these ideas are still exploratory, they underscore how broadly the bank is thinking about automation.
The announcement comes at a sensitive moment for the AI sector. Recent model updates from Anthropic have triggered sharp reactions in financial markets, with investors selling off shares of software companies and reassessing which firms are best positioned to benefit from the AI boom. The volatility reflects growing recognition that rapid improvements in foundation models could disrupt existing business models, including those built around legacy enterprise software.
Goldman’s willingness to work closely with Anthropic also highlights a shift in how large financial institutions engage with AI vendors. Rather than simply buying off-the-shelf tools, Goldman is embedding engineers and co-developing systems tailored to its specific needs, regulatory obligations, and risk controls.
Despite the scale of automation being discussed, Goldman has been cautious in addressing concerns about job losses. The bank employs thousands of people in compliance, accounting, and operations, and Argenti said it would be premature to assume the technology will directly eliminate those roles. Instead, Goldman’s stated aim is to “inject capacity,” allowing teams to do more work faster and improve service quality.
Still, the longer-term implications are difficult to ignore. As AI agents mature, Goldman could reduce its reliance on third-party service providers that currently handle parts of its operational workload. That could shift costs away from external vendors and further concentrate expertise and control inside the bank.
More broadly, Goldman’s strategy signals a bigger change in how Wall Street views technology. Generative AI is no longer framed as a productivity tool for individual workers, but as an organizing principle for the firm itself. By embedding autonomous agents into core processes, Goldman is testing whether a global investment bank can be redesigned around machines that reason, decide, and act with limited human intervention.
If successful, the effort could set a precedent for the industry, forcing rivals to accelerate their own AI adoption or risk falling behind.