Overview
- Goldman Sachs has spent six months co-developing autonomous agents with embedded Anthropic staff for accounting of trades and transactions as well as client vetting and onboarding.
- The agents use Anthropic’s Claude model and are described as in the early stages, with Goldman expecting to launch them soon without providing a date.
- Chief information officer Marco Argenti says the systems are designed to cut the time needed for essential back-office tasks, accelerating client onboarding and resolving trade reconciliation issues faster.
- As part of a multi-year generative AI reorganization, Goldman aims to constrain headcount growth, though Argenti says it is premature to expect job losses and notes the bank may reduce reliance on some third-party providers.
- Goldman previously rolled out the Devin autonomous coding tool internally and found Claude unexpectedly strong in areas like accounting and compliance, with potential future applications including employee surveillance and pitchbook creation.