Breaking: AI Agent Teams Emerge as Lifeline for Small Dev Teams Facing Vulnerability Surge

As AI-powered red teams like Anthropic's Claude Mythos uncover critical bugs at an alarming rate, small open-source projects run by one or two developers are being overwhelmed. A new open-source tool called Squad, from Microsoft, aims to give these teams a force multiplier by orchestrating a team of AI agents to help fix code.

“Suddenly, in the last month or so, AI reports became useful,” said Greg Kroah-Hartman, Linux kernel maintainer, at Kubecon Europe. He noted that improved tools and deeper understanding had transformed previously worthless AI output into actionable security intelligence.

But for projects with limited manpower, the influx of vulnerability reports is a crisis. “We need code that’s fixed and we need it now, but we don’t have enough skilled developers,” said Brady Gaster, Principal PM Architect at Microsoft, creator of Squad.

Background

Anthropic’s Claude Mythos recently unveiled significant bugs across both closed- and open-source projects, triggering a scramble to patch them. Large projects have teams of developers and volunteers to respond, but smaller ones often lack resources.

Agent harnesses—frameworks that manage teams of AI agents—have grown more powerful, with tools like OpenClaw gaining popularity. However, they can be expensive and prone to hallucinations. Squad addresses this by grounding agents in structured code and APIs, and by focusing on the software development lifecycle.

What This Means

Squad, installed via a single CLI command, creates a developer lead, front-end developer, back-end developer, and test engineer—all agents working alongside the human developer. “It’s like having a small team that never sleeps,” Gaster explained. “We’re using spec-driven development and agent harnesses to clear technical debt faster.”

This approach could become a critical tool for small projects trying to keep pace with AI red teams. By automating routine fixes and testing, agents may provide the productivity boost needed to maintain security and stability in an increasingly AI-driven threat landscape.

“We’re seeing a pivot from general-purpose LLMs to specialized, grounded agent teams,” said Gaster. “Squad is designed to be both practical and affordable, using GitHub Copilot as the foundation.”