Anthropic Claude Autonomous Agents for Trade Accounting and Client Onboarding

Goldman Sachs deployed autonomous AI agents built on Anthropic Claude model for two high-stakes back-office workflows: trade and transaction accounting, and client due diligence and onboarding. Embedded Anthropic engineers spent six months co-developing the agents with Goldman technology teams and domain experts. For trade accounting, the agents review millions of transactions annually, match records across internal ledgers, counterparty confirmations and external bank statements, and flag discrepancies — replacing slow, resource-intensive reconciliation processes. For client onboarding, the agents parse KYC documents (passports, corporate registrations, beneficial ownership documents), apply AML compliance rules, and cross-reference against regulatory requirements — accelerating client setup while maintaining regulatory accountability. Claude pairs with rules-based systems and human oversight to handle the exception-heavy workflows that pure automation cannot manage. Anthropic engineers were embedded within Goldman technology teams to co-develop the agents, leveraging Claude large context windows and strong instruction-following capabilities for document-heavy workflows. The agents are deployed in the workflow layer augmenting human analysts, who focus on exceptions and final decisions rather than repetitive extraction and comparison work. Goldman plans to extend agents to vendor management, procurement, and lending workflows next.