Goldman Sachs Deploys Virtual Software Engineers Using Agentic AI
Goldman Sachs has become the first major bank to deploy virtual software engineers on real production tasks, using agentic AI to modernize its legacy infrastructure and accelerate software development.
The financial services industry relies heavily on data, speed, and accuracy, but much of its digital infrastructure was built decades ago using outdated programming languages. Modernizing this infrastructure is a painstaking process that involves rewriting old code from scratch, shifting data, and testing for errors.
Goldman Sachs turned to AI agents to tackle this task, deploying Devin, developed by Cognition, across its technology division in 2025. Devin is fully agentic, meaning it can work autonomously within the bank's existing IT infrastructure and alongside human teams.
The results of Goldman's deployment were impressive: one large organization saved five to 10 percent of development time using Devin for fixing security issues in code, while another customer saw the time taken to fix vulnerabilities reduce from 30 minutes per issue to 1.5 minutes.
Goldman CIO Marco Argenti described Devin as 'like a new employee' and expected it to be three or four times more productive than existing AI tools. The bank has since expanded its agentic engineering project, adopting another platform, Anthropic's Claude, for work involving trades and transactions, client vetting, and onboarding.
However, the deployment of agentic AI raises important questions about the impact on human workforces. Researchers warn that redundancy and replacement are likely, with up to 200,000 jobs potentially being lost in the US banking sector alone. Goldman's CEO David Solomon has talked about restricting headcount, while Argenti expects AI-driven job cuts to extend beyond IT engineering.