Goldman Sachs (GS.US) is collaborating with AI startup Anthropic to develop intelligent agents aimed at automating various roles within the bank, according to a senior technology executive. Marco Argenti, Chief Information Officer at Goldman Sachs, revealed that Anthropic's engineers have been working on-site at the bank for the past six months. Together, the teams are developing autonomous agents, with the initial focus on two key areas: trade accounting, and client due diligence and account opening processes.
Argenti stated that agents based on Anthropic's Claude model are in the early stages of development and are expected to significantly reduce processing times for these core functions. He confirmed the agents will be launched soon but did not provide a specific timeline. Argenti described the technology as acting like a "digital colleague" for many of the bank's standardized, highly complex, and process-driven roles.
Last October, Goldman Sachs CEO David Solomon announced the bank had initiated a multi-year plan to comprehensively restructure its business around generative AI. Since the launch of OpenAI's ChatGPT in late 2022, generative AI has created significant waves across industries. Solomon also noted that despite soaring revenues in trading and advisory businesses at Goldman Sachs and other investment banks, the institution plans to focus on controlling headcount growth during this transformation.
This partnership announcement coincides with Anthropic, founded by former OpenAI executives, completing a model upgrade. This development has contributed to significant stock price declines for software companies and their creditors, as investors place bets on the eventual winners and losers in the AI race.
Argenti disclosed that Goldman Sachs began testing an autonomous AI programming tool named Devin as early as 2025, which is now available to all of the bank's engineers. During testing, the bank quickly recognized that Anthropic's AI models could be applied to other business areas. "Claude's programming capabilities are outstanding," Argenti analyzed. "The question is whether this is due to the unique nature of programming itself, or because the model possesses the core ability to reason through complex problems step-by-step and apply logic flexibly."
He expressed that the Goldman team was pleasantly surprised by Claude's ability to handle tasks beyond programming, particularly in fields like accounting and compliance. These areas require both parsing vast amounts of data and documents, and accurately applying rules to make professional judgments in practical scenarios, areas where Claude has demonstrated strong performance.
Argenti stated that Goldman Sachs has reached a clear conclusion: other business areas within the bank have the potential to achieve automation levels and significant efficiency gains comparable to those in programming. He indicated that with the deployment of these developing agents, client onboarding processes will accelerate, and the efficiency of resolving trade reconciliation and other accounting-related issues will substantially improve.
Reportedly, Goldman Sachs plans to next expand the application of these agents to tasks including employee supervision and the creation of investment banking pitch materials. Although Goldman's accounting and compliance departments currently employ thousands of staff, and these AI agents are slated for deployment in such departments, Argenti emphasized that it is still too early to conclude that this technology will lead to job cuts in these areas. However, he acknowledged that as AI technology matures, Goldman Sachs may gradually reduce its reliance on some current third-party service providers.
"This is always a trade-off," Argenti commented. "Our current core approach is to use AI to inject additional capability into our business. In most cases, this makes our operations more efficient, leading to better client experiences and creating more business opportunities."