Goldman Sachs Tests AI Agent Devin as a ‘New Employee’ | Blockchain Council

Goldman Sachs is actively testing Devin, an autonomous AI software engineer developed by Cognition Labs. The bank has deployed hundreds of Devin instances in its internal systems, with plans to scale up. This marks one of the first real-world cases where an AI agent is evaluated like a human employee.

If you’re asking what this means – Goldman is seeing how far AI can go in performing real coding tasks usually handled by junior developers.

What Is Devin?

Devin is an AI-powered coding agent trained to perform full software engineering tasks. It operates inside its own environment, complete with a shell, browser, and code editor. Unlike traditional AI coding assistants, Devin can execute multi-step programming tasks from start to finish.

Cognition Labs, the company behind Devin, claims it is the first AI capable of functioning as a full-stack software engineer. It has already shown that it can:

Write and debug code
Test software in realistic dev environments
Deploy code to live systems

This hands-free, agentic behavior is what makes Devin different from tools like GitHub Copilot.

Why Goldman Sachs Is Testing Devin

Goldman Sachs sees Devin as a potential productivity multiplier. According to CIO Marco Argenti, the bank’s goal is to evaluate Devin “as if it were a new employee.”

Here’s why Goldman is testing this approach:

To automate repetitive tasks in coding
To support its 12,000+ human developers with AI agents
To increase speed and accuracy in delivery
To explore cost-effective ways to scale engineering operations

The bank has already deployed hundreds of Devin instances and may roll out thousands more if tests go well.

What Devin Can Do at Goldman

Devin is not just a demo or a prototype. Goldman Sachs is already using it to perform specific software development tasks under human supervision. These include:

Editing and updating legacy codebases
Writing unit and integration tests
Refactoring sections of internal tools
Preparing code for production environments

Supervising engineers review all output, but Devin does most of the heavy lifting in these assigned tasks.

Task Breakdown in Live Deployment

Here’s how Devin fits into Goldman Sachs’ coding workflow:

Engineers assign Devin a ticket through a project management system
Devin reads the codebase, identifies dependencies, and proposes a solution
It writes and tests the code in its own sandboxed environment
Human engineers review and merge the pull request

Use Cases of Devin in Production

Task Type	Fully AI-driven	Human-supervised	Manual by Engineers
Code refactoring	Yes	Yes	Occasionally
Test writing	Yes	Yes	Rarely
Codebase navigation	Yes	No	No
Deploy-ready output	No	Yes	Yes
Debugging complex logic	No	Yes	Yes

This setup creates a hybrid workflow that blends AI automation with human expertise.

What Makes Devin Different

Devin isn’t just a chatbot for code. Its key differentiator is agency – it can carry out multi-step technical tasks independently. Unlike other AI tools that stop at suggesting code snippets, Devin:

Executes shell commands
Tests code in runtime environments
Uses context across multiple files
Follows task instructions with reasoning steps

This makes it more than just a helpful assistant. It becomes a co-worker – one that never sleeps, doesn’t need a paycheck, and can work in parallel with thousands of other instances.

Goldman’s Expected Productivity Gains

Goldman Sachs has already seen gains with AI-powered coding copilots. Some teams report a 20 percent boost in developer speed for small tasks. With Devin, internal estimates point to a 3x to 4x productivity increase in more complex workflows.

The firm believes this type of agentic AI could eventually take on 20 to 40 percent of all software engineering workload – especially in code-heavy departments like risk, compliance, and data engineering.

Devin vs Other AI Coding Tools

Tool Name	Autonomy Level	Deployment Type	Ideal Use Case	Agentic Behavior
Devin	High	Private instances	Multi-step coding tasks	Yes
GitHub Copilot	Low	Editor plugin	Code suggestions in real time	No
ChatGPT (Code)	Medium	Chat interface	Q&A and logic generation	No

Devin’s main edge lies in its ability to plan, execute, and revise – without needing repeated prompts.

Is Devin Replacing Human Engineers?

No. Goldman Sachs emphasizes that Devin supports developers, not replaces them. Humans remain responsible for:

Final code reviews
Deciding architecture
Approving deployments
Integrating outputs into larger systems

Devin is not ready to handle creative, abstract engineering work or deal with novel problems that require deep judgment. It is a tool – not a peer.

Career Implications for Developers

Devin’s arrival signals a shift in how engineering teams may function. As AI agents take on more of the repetitive work, developers will be expected to:

Work alongside AI agents efficiently
Review and refine machine-generated code
Spend more time on architecture, security, and strategy

For those looking to stay competitive, investing in AI skills is critical. The AI Certification helps professionals understand how agentic tools are built and deployed. If you’re more data-focused, the Data Science Certification is ideal. And if your goals lean toward strategy and growth, the Marketing and Business Certification explores how AI shapes commercial outcomes.

Strategic Impact for Enterprise AI

Devin’s deployment is not just a tech milestone – it’s a business case. It shows that:

AI agents can be embedded in live workflows
Large institutions are ready to experiment at scale
Hybrid AI-human teams are not just the future – they’re already here

Goldman Sachs is using Devin to push boundaries and create a blueprint that others in finance, insurance, and tech may follow.

Why This Pilot Matters

This is one of the first times an AI tool has been treated as an employee equivalent. From HR to deployment structure, Goldman is setting a precedent:

AI onboarding as part of team operations
Tracking AI performance like human contributors
Iterating workflows to better integrate agents

It’s not just a trial – it’s a restructuring of how enterprise tech teams function.

Devin’s Enterprise Benefits

Benefit Area	How Devin Helps	Resulting Impact
Developer productivity	Handles low-level code tasks	Saves time and effort
Cost efficiency	Runs at scale with fewer overheads	Reduces operational cost
Quality assurance	Generates test cases and checks	Improves code quality
Scalability	Can deploy thousands of agents	Meets demand flexibly

Devin is not a threat to human talent – but a tool to amplify it.

Final Thoughts

Devin’s pilot at Goldman Sachs marks a turning point in enterprise AI adoption. It’s more than a coding assistant. It’s a software agent with real responsibilities and performance expectations.

As more organizations explore agentic AI, developers, managers, and business leaders will need to adapt. Goldman is showing what this transition looks like – not in theory, but in action.