Harnessing AI Agent Teams: How Squad Helps Developers Tackle the Rising Tide of Vulnerabilities

The Crisis of Developer Productivity

At KubeCon Europe, Linux kernel maintainer Greg Kroah-Hartman shared a startling observation. For over a year, AI-generated pull requests and security reports were largely useless—dismissed as "slop". But recently, those reports suddenly became valuable. Kroah-Hartman suspected improved tools and better usage methods were behind the shift.

Shortly after, Anthropic's Claude Mythos emerged, triggering a scramble across open-source and closed-source projects to patch critical vulnerabilities it uncovered. Large projects with substantial teams could handle the fixes. But smaller projects—often run by one or two volunteers in their spare time—face an overwhelming challenge. The rise in AI-discovered bugs demands code fixes immediately, yet skilled developers are scarce. This is a crisis of developer productivity.

Can AI Agents Solve the Problem?

Agent harnesses have become powerful tools for orchestrating teams of AI agents. General-purpose solutions like OpenClaw are popular but expensive, consuming large numbers of tokens across models and services. Moreover, like many LLM-based tools, they suffer from inaccuracies and hallucinations.

However, an approach grounded in a defined methodology and a substantial corpus of data could help meet the sudden demand for developer productivity. Code and APIs provide structured grounding, while the diverse skills needed in modern software development can be replicated by a team of agents. What's needed is a way to combine these elements—using spec-driven development and agent harnesses—to give every developer their own AI team. Soon, agents could serve as a force multiplier, outpacing AI red teams and reducing technical debt.

Introducing Squad: An Open-Source Agent Harness

One promising example is Squad, an open-source project by Brady Gaster, Principal PM Architect in the CoreAI Apps and Agents team at Microsoft. Squad builds an agent harness around GitHub Copilot, orchestrating a team of agents to work collaboratively with you on your code. Designed for a single CLI installation, Squad creates agents that mimic a full development team: a developer lead, a front-end developer, a back-end developer, a test engineer, and a documenter. Each agent has a specific role and can handle tasks like writing code, running tests, and generating documentation.

How Squad Works

After installation, Squad uses prompts and context to define tasks and assign them to appropriate agents. The agents communicate via a shared context, allowing them to coordinate like human team members. For example, the backend developer might create an API endpoint, the front-end developer builds the UI, and the test engineer validates the integration. The developer lead oversees progress and can intervene when needed. This structured collaboration reduces errors and speeds up development.

Key Features of Squad

• Role-based agents: Each agent specializes in a specific area (front-end, back-end, testing, etc.).
• GitHub Copilot integration: Leverages Copilot's code generation capabilities as a core component.
• Spec-driven development: Agents work from defined specifications, ensuring alignment with requirements.
• Open-source and extensible: Developers can modify or add custom agents to fit their workflow.

Why Small Teams Need Squad

For small projects, Squad offers a democratic approach to AI-powered development. Instead of relying on a single overworked maintainer, the agent team handles routine tasks, freeing humans for complex problem-solving. The force multiplier effect is critical: with limited developers, Squad helps triage bugs, write patches, and maintain documentation. This allows small teams to respond to vulnerability waves without burning out.

The Future of Development with Agent Teams

As AI red teams uncover more vulnerabilities, the need for efficient developer workflows grows. Projects like Squad demonstrate that agent harnesses can provide the necessary boost. By combining structured code, spec-driven methods, and collaborative agents, we can keep software secure and reduce technical debt. Squad is still evolving, but its approach points toward a future where every developer has a personal AI squad—ready to code, test, and deploy alongside them.

Learn more about Squad's architecture or explore agent harnesses for your next project.