Goldman Sachs Onboards Devin, an AI Engineer Built to Code Like a Human

Goldman Sachs’ newest software engineer won’t need a desk, won’t take a sick day, and won’t ever step into the company cafeteria. That’s because it isn’t human. The global investment bank has begun deploying Devin , an autonomous AI software engineer developed by San Francisco-based startup Cognition Labs. In a move that could signal the start of a new phase in enterprise AI adoption, Goldman Sachs is integrating the tool into its developer workflows, with plans to scale usage from “hundreds” to potentially “thousands,” according to Marco Argenti , the firm’s Chief Information Officer. “We’re going to start augmenting our workforce with Devin, which is going to be like our new employee who’s going to start doing stuff on behalf of our developers,” Argenti told CNBC in a recent interview. This deployment marks a significant leap for Wall Street’s experimentation with artificial intelligence. While JPMorgan Chase and Morgan Stanley have introduced internal tools built on large language models to help employees draft emails or summarize documents, Goldman is taking it a step further: using AI agents not just to assist but to execute software engineering tasks from end to end. Devin is designed to function like a full-stack engineer, capable of writing code, building apps, debugging, and updating legacy systems all with minimal supervision. While Goldman is still in the early stages of implementation, the tool has already been tested internally and is being positioned as a teammate for the bank’s 12,000 human developers. “A Hybrid Workforce” Goldman Sachs isn’t trying to replace its engineers, at least not yet. Instead, the firm is introducing Devin as part of a broader strategy to evolve toward a “hybrid workforce” , where AI agents operate under human supervision to enhance productivity, particularly in repetitive or low-value tasks. “It’s really about people and AIs working side by side,” said Argenti. “Engineers are going to be expected to have the ability to really describe problems in a coherent way and turn it into prompts…and then be able to supervise the work of those agents.” The move aligns with Goldman’s broader AI strategy. Last month, the firm rolled out GS AI Assistant , an internal large language model-based tool that allows employees to interact with AI within the bank’s firewall. Over 10,000 staff have used it, suggesting a growing cultural shift inside one of the most traditional financial institutions. But Devin represents a new tier of capability. Unlike LLMs used to write text or summarize research, Devin belongs to a category of agentic AI systems that take initiative, make decisions, and complete multistep workflows without constant human input. This form of AI is about executing tasks, much like a junior engineer would. Not Just Hype, But Not Flawless Either Devin gained wide attention in 2024 when Cognition Labs released demo videos showing the agent building apps and fixing bugs with little intervention. At the time, the company declared it had created the world’s first AI software engineer, triggering both enthusiasm and anxiety across the tech industry. Cognition described Devin as “a tireless, skilled teammate, equally ready to build alongside you or independently complete tasks for you to review.” The tool’s anthropomorphic branding, giving Devin a name, a persona, and a job title has helped it break into boardrooms and engineering teams alike. And for Goldman Sachs, the value proposition is clear. In Argenti’s words, Devin has the potential to boost productivity by three to four times over traditional AI tools. Still, there are limits to the technology. Independent tests conducted by researchers at Answer.AI earlier this year found that Devin successfully completed only three of twenty assigned tasks. The tool reportedly performed well on simple requests but failed to handle complex workflows, occasionally hallucinating features or getting stuck in loops. “Polished interface, unreliable results,” the researchers concluded. Cognition has since updated Devin to version 2.1, claiming that it now performs best when working on large codebases that provide enough contextual information to guide its reasoning. While Goldman has not disclosed whether it is using this version, it has made clear that Devin will work under the supervision of human developers at least for now. Goldman’s AI Bet Goldman Sachs has a track record of early AI adoption, particularly under Argenti, who joined the firm from Amazon in 2019. The bank began experimenting with developer copilots well before the current wave of AI hype, and now sees agentic tools like Devin as the next evolution. Interestingly, Goldman does not hold an equity stake in Cognition Labs, according to a source familiar with the matter. Cognition, meanwhile, has seen a meteoric rise. Founded in late 2023 by a team of engineers and competitive coders, the startup raised hundreds of millions in early funding from 8VC, Peter Thiel, and Joe Lonsda

Goldman Sachs Onboards Devin, an AI Engineer Built to Code Like a Human

“A Hybrid Workforce”

Keep reading. It's free.

Related stories