Goldman Sachs put hundreds of autonomous coding agents onto real production work in 2025. The bank deployed Devin, built by AI startup Cognition, straight into its technology division where the bots now operate alongside roughly 12,000 human engineers. This is not a chatbot suggesting code snippets. Devin scopes requirements, writes, tests, debugs and submits work for human review. It works inside Goldman's actual infrastructure.
What makes this deployment stand out is the speed gain. One team inside the bank saw vulnerability fixes drop from 30 minutes per issue to just 1.5 minutes. CIO Marco Argenti said Devin outperforms previous AI tools the bank tried by three to four times. The numbers matter because they suggest this experiment is already moving past proof-of-concept territory into something that reshapes how a major financial institution handles engineering work.
What comes next
Goldman plans to expand into Anthropic's Claude for trading systems, transaction processing and client onboarding in 2026. The bank sees agentic AI as a scaling lever for internal capacity rather than a replacement play. Human engineers still review outputs. The difference is the ratio of what gets handled without asking a person to start from scratch.
The broader market is watching closely. Analysts estimate up to 200,000 US banking jobs could be at risk, many held by junior developers who traditionally do the grunt work that Devin now handles. Whether Goldman's success story becomes the blueprint for other institutions or stays locked inside the bank depends partly on how quickly other startups can build competing agents. The timeline just compressed from "maybe in five years" to "happening now."
This article is informational only and does not constitute financial or investment advice.
