Goldman Sachs Deploys AI Agents as Virtual Software Engineers

Goldman Sachs turns to autonomous AI for code modernization

Goldman Sachs has deployed autonomous AI agents as virtual software engineers, working alongside its 12,000-person development team to tackle one of banking's most persistent challenges: modernizing decades-old legacy systems.

The investment bank is using AI agents like Devin—tools capable of scoping projects, writing code, testing, and debugging autonomously—to accelerate the painstaking work of updating infrastructure built with outdated programming languages by engineers who have long since retired. According to details first reported by Forbes contributor Bernard Marr, Goldman's CIO Marco Argenti described Devin as functioning like a new employee on the team.

The results have been substantial. Argenti reported that these AI agents deliver three to four times the productivity of previous AI-assisted development tools, while drastically reducing both development timelines and the time required to fix security vulnerabilities. Goldman has also expanded its AI engineering capabilities by incorporating Anthropic's Claude into its development workflow.

Why it matters

Goldman's implementation represents a significant test case for agentic AI in enterprise software development. Unlike code completion tools that assist human programmers, these agents can execute entire development tasks independently—a capability that promises to reshape how financial institutions maintain and modernize their technology stacks. The approach directly addresses a critical industry pain point: legacy systems that are expensive to maintain and risky to replace, yet essential to daily operations.

The automation question

While Goldman emphasizes that its AI agents augment rather than replace human engineers, the deployment raises pointed questions about the future of software development roles, particularly for junior developers who traditionally handle routine coding tasks.

Predictions suggest the US banking sector could see 200,000 job losses as AI adoption accelerates. This creates a potential paradox: if entry-level positions disappear, how will the industry cultivate the senior engineers it will still need? That talent pipeline challenge extends beyond finance to every sector adopting agentic AI for software development.

The technology excels at exactly the kind of work human engineers find most tedious—repetitive code translation, systematic testing, and incremental updates to stable systems. AI agents, as Marr notes, have no opinion about mundane tasks and can work continuously on the unglamorous but essential work of keeping legacy infrastructure functional.

For financial services firms sitting on decades of technical debt, the value proposition is clear: faster modernization, reduced costs, and the ability to redirect human engineers toward higher-value architectural and strategic work. The open question is whether the industry can manage the transition without losing the foundational skills that create those senior engineers in the first place.

These details were first reported by Bernard Marr writing for Forbes.