Key Takeaways
- Goldman Sachs is deploying Devin, an autonomous AI coder from Cognition Labs, to work alongside its 12,000 human developers, handling tasks like legacy code updates and app development .
- Productivity gains of 3–4× are anticipated, far exceeding earlier AI tools, with initial deployments scaling to "hundreds of Devins" and potentially thousands .
- Job impacts loom large: Industry forecasts predict up to 200,000 finance roles could be cut in 3–5 years due to AI automation, shifting human roles toward supervision and prompt engineering .
- Security and specialization define Wall Street’s AI push: Goldman’s tools operate within a firewalled ecosystem, while rivals like JPMorgan and Morgan Stanley accelerate their own AI agent rollouts .
- Hybrid workforce models emerge as the norm—developers now articulate problems for AI execution, blending technical and managerial skills .
Goldman’s Non-Human Hire: Meet Devin, the AI Software Engineer
So get this: Goldman Sachs just hired something that doesn’t take lunch breaks. Or vacations. It’s name’s Devin, and it’s not a person. It’s what they call an “autonomous software engineer” built by this startup Cognition Labs. Marco Argenti—he’s Goldman’s tech chief—told CNBC Devin’s gonna join their army of 12,000 human coders soon. And like, it ain’t just a helper; it does multi-step projects start-to-finish. Building apps. Migrating old codebases. Stuff that used to need teams of people .
Cognition’s kinda wild—founded just late 2023, already valued near $4 billion. Peter Thiel’s involved. They showed demo videos last year of Devin doing full-stack engineering work with barely any hand-holding. Argenti’s vision? Start with hundreds of these AI agents. Then maybe thousands. “It’s going to be like our new employee,” he said. Not replacing humans tomorrow, but redefining what they do .
What Makes Devin Different? It Doesn’t Just Suggest—It Builds
You know those chatbots that summarize documents or draft emails? Yeah, Devin ain’t that. Earlier bank AIs were glorified assistants. This thing executes. Like, give it a task—“update this internal system to Python 3.11” or “build a compliance tool”—and it does it. Minimal oversight. Argenti specifically mentioned legacy updates because, well, that’s the work junior devs hate. Or consulting firms charge millions for. Now? Goldman just points Devin at it .
Here’s how it compares to prior tools:
Tech giants like Microsoft already get 30% of code from AI. Salesforce says 50%. But Wall Street? This is its first taste of true autonomy. Not a copilot—a pilot .
Productivity or Layoffs? The 200,000 Job Question
Argenti talks about 3–4× productivity leaps. That’s huge—previous tools maybe gave 20–30%. But here’s the tension: if one AI does the work of three people, what happens to the people? Bloomberg’s researchers predicted in January that banks could shed 200,000 jobs globally in 3–5 years thanks to AI. Not just coders, either. Operations. Analysis. Even management’s getting automated; one banker told Gizmodo AI now writes staff reviews .
Goldman says it’s about “augmentation.” Freeing humans for “higher-value work.” But real data’s messy: another bank’s AI handles 85% of margin call replies automatically. That alone saved hiring 30 ops staff. So yeah—the hybrid workforce is coming. But it’ll be smaller. Fiercer. And if you’re not training to manage AI? Gonna be rough .
Why Banks Are All-In on Agentic AI Now (It’s Not Hype)
Timing’s everything. Last year, JPMorgan and Morgan Stanley rolled out chatbots. Helpful, sure. But not transformative. Devin’s different because it acts. And banks are ready: stricter data controls exist now (Goldman’s AI runs in a firewalled space). Costs plummeted—training models got cheaper. And pressure? Brutal. Margins shrink. Clients demand faster everything .
Goldman’s not alone. Citi uses AI to scan legal docs. Bank of America automates contract analysis. But Cognition picked Goldman as Devin’s first banking partner because they’ve got skin in the game—12,000 developers is a massive test lab. Succeed here? Every rival follows. Already, Argenti calls this a “proof point” for expanding AI into other roles. Finance. Risk. You name it .
Hybrid Workforce Reality: Engineers Become “AI Whisperers”
So what’s left for humans? A lot, actually—but different. Argenti put it bluntly: engineers now need to “describe problems in a coherent way and turn it into prompts.” Then supervise Devin’s output. It’s less about writing loops, more about:
- Articulating tasks precisely
- Validating AI-generated code
- Pivoting when results miss the mark
- Architecting systems where agents collaborate
Think of it like tech management meets linguistics. If you can’t communicate clearly with AI? You’re stuck. Goldman’s already training devs on this—their year-long internal GS AI Assistant pilot (with 10k staff) proved prompt engineering isn’t fluff. It’s core .
Beyond Code: How AI Reshapes Finance Culture and Talent
Wall Street’s identity was hustle. 100-hour weeks. Grinding through pitch books. Now? Goldman’s Banker Copilot drafts those. Translation tools handle cross-border research. The work doesn’t vanish—but who does it changes. Junior bankers feared being replaced. Truth is, they’re being redirected. Analysts validate AI output instead of compiling data. Associates design AI workflows instead of managing teams .
But skepticism’s real. One insider noted the GS AI Assistant “avoided hiring 30 people” for margin calls. That’s one team at one bank. Scale that industry-wide? Hence the 200,000 job cut forecasts. Talent strategies now prioritize:
- AI literacy across all roles
- Prompt engineering skills
- Ethical oversight capabilities
- Cross-functional AI leadership
You don’t need fewer humans. You need different humans .
The Future According to Goldman: Thousands of Devins, New Roles
Argenti’s roadmap is clear: start with hundreds of Devins. Scale to thousands if it works. Then expand the model to other domains. Risk modeling. Client reporting. Trading analytics. Why? Because once you crack agentic AI for coding—a highly structured domain—applying it elsewhere gets easier .
Risks? Oh yeah. Hallucinations in code could crash systems. Security gaps might leak data. But Goldman’s betting its strict compliance firewall (tested for a year firm-wide) mitigates that. Competitors are watching. If productivity jumps 3×? Morgan Stanley won’t wait. JPMorgan can’t afford to. The first-mover advantage here isn’t tech—it’s organizational adaptation. Banks that retrain fastest win .
Conclusion: Augmentation, Not Replacement—For Now
Let’s be real: No one’s firing 12,000 developers tomorrow. But attrition? Hiring freezes? Task consolidation? Absolutely. Devin handles the “drudgery,” says Argenti. Humans handle the nuance. That’s the theory. Reality might be messier—some jobs will vanish. But new ones emerge: AI supervisors. Hybrid workflow designers. Ethics auditors .
Goldman’s move matters because it’s practical, not philosophical. They tested for a year. Scaled cautiously. If this works, Wall Street’s engineer shortage eases. Costs fall. Innovation? Accelerates. But the human-AI partnership must be intentional. Train people. Redefine roles. Measure outcomes beyond productivity. Because if banks miss that? Even the smartest AI won’t save them .
Frequently Asked Questions
What exactly is Devin at Goldman Sachs?
Devin is an autonomous AI software engineer created by Cognition Labs. It’s not a chatbot—it builds, tests, and deploys code independently. Goldman uses it for tasks like updating legacy systems or creating apps, supervised by human developers .
Will Goldman Sachs fire developers because of AI?
Not immediately. But Marco Argenti (CTO) sees Devin boosting productivity 3–4×, implying fewer humans needed per task long-term. Bloomberg predicts up to 200,000 finance jobs cut industry-wide by 2030 due to AI .
How is Devin different from GitHub Copilot?
GitHub Copilot suggests code snippets inside your editor. Devin writes entire projects—it’s a standalone “employee” handling multi-step tasks (e.g., migrating codebases). Think collaborator vs. executor .
What banks besides Goldman use AI like this?
JPMorgan and Morgan Stanley use AI for summaries/emails. But Goldman’s the first with agentic AI (Devin) that replaces human actions. Citi and Bank of America automate document analysis, though .
Are there risks with autonomous AI coders?
Major ones. Hallucinations (flawed code), security gaps, and over-reliance. Goldman claims its firewalled AI environment and human supervision mitigate this—but it’s untested at scale .