Goldman Sachs partners with Anthropic to develop AI agents automating internal tasks

Goldman Sachs has been working with AI startup Anthropic to develop autonomous agents that would automate a widening set of internal, paperwork-heavy functions, the bank said. The effort involves Anthropic engineers embedded inside Goldman teams for the past six months and relies on Anthropic’s Claude model and related products, including a tool described by Anthropic as Claude Cowork.

Marco Argenti, Goldman’s chief information officer, described the collaboration as “for the past six months collaborating with Anthropic engineers embedded within its teams to build autonomous agents for tasks including trade and transaction accounting as well as client due diligence and onboarding.” He said the work is intended to target trade accounting, trade reconciliation, client vetting and onboarding processes along with other compliance and document-processing workflows. Argenti added that “the technology is expected to significantly reduce the time required to complete core operational processes,” and that the bank plans to launch the agents “soon.”

Goldman framed the pilot as a productivity and client-experience play rather than an immediate cost-cutting move. Argenti said it was “premature” to expect the technology will lead to job losses in compliance and accounting functions. The bank’s CEO, David Solomon, has previously said the firm is reorganizing around generative AI and plans to “constrain headcount growth,” a comment that helps explain the strategic stakes for operations teams.

Analysts and industry commentators say Goldman’s approach - embedding engineers and co-developing models with extensive customization and control - reflects how large financial firms plan to adopt AI safely. Finimize highlighted the need for audit trails, controls and accountability when integrating models into legacy systems and said the project “could speed up compliance-heavy work and reduce reliance on outside providers over time.” Techbuzz AI characterized the move as “one of Wall Street’s most aggressive bets on autonomous AI agents yet,” noting the work targets back-office processes that have resisted automation for decades.

The announcement has wider market reverberations. Recent Anthropic model updates have prompted investor nervousness that contributed to a sell-off among software firms and their credit providers, underscoring how enterprise AI partnerships can reshape vendor economics if internal tools prove dependable enough to replace third-party platforms.

Key questions remain unanswered for workers and managers: which Claude version is being used, how models will be hosted and audited, how data privacy and regulatory rules for KYC and accounting auditability will be enforced, and how many roles might be affected if pilots scale. For employees in affected functions, the immediate takeaway is likely increased pressure to document processes, participate in pilots and reskill as Goldman pushes to accelerate manual, compliance-heavy workflows. Watch for pilot performance metrics, governance details and a clearer rollout timeline as the bank moves from co-development into production.