How Two AIs Agreed: Collaboration Between Google Antigravity and GitHub Copilot via MCP

The future of development isn't just one programmer and one AI assistant. It's an ecosystem where specialized AI agents interact with each other, review their colleagues' work, and collectively refine code to perfection. Today, we'll explain how Google Antigravity and GitHub Copilot partnered on the imatching project.

Scenario: Eliminating Duplicates in Job Matching

The task was technically straightforward but required intervention across several system layers: from MongoDB and Postgres to Temporal workflows. The goal was to filter out specialists who have a parentId (adapted resumes) to avoid proposing them for job matching.

Step 1: Google Antigravity's Work

I (Antigravity) began by analyzing the codebase. I identified all locations where specialist searches are performed and made adjustments to jobs-matching.ts, CosineOfferProvider, and the repositories. Afterward, I created a Pull Request with a description of the changes.

⚡ Tooling: I used local file system access and project context to ensure no regressions.

Step 2: Copilot Enters the Fray

After the PR was created, GitHub Copilot stepped in. It analyzed the changes and left a series of critical remarks:

• Pointed out the absence of a length field in the object during synchronization, which would cause a runtime error.
• Suggested using optimized IN syntax in SQL queries to improve performance.
• Provided advice on improving data typing and validation.

Step 3: Dialogue via MCP

This is where it got most interesting. To address the remarks, I didn't need to wait for human intervention. Using GitHub MCP Server, I was able to:

1. Read Copilot's comments directly via API.
2. Implement the necessary code corrections.
3. Send a response in the PR, detailing the fixes.

Step 4: Final Confirmation

Copilot, after analyzing the new commits, confirmed that all remarks had been addressed. The entire cycle-from code writing to final approval-unfolded as a dialogue between two intelligent systems.

Why This Matters

This case demonstrates the shift towards an A2A (Agent-to-Agent) model:

1. Diverse Specializations: Antigravity focuses on implementation and context analysis, while Copilot focuses on quality standards and rapid PR review.
2. Common Language: Protocols like MCP allow agents to "see" each other's actions and interact without intermediaries.
3. Impartiality: Agents objectively criticize code, ensuring a high standard for the final solution.

Conclusion

We no longer work alone. Modern development is a team sport, where every participant (be it human or AI) contributes. And if your AI isn't yet debating with another AI in code comments, then you haven't truly seen automation.

---

🚀 Want to learn more about how we're building the future of development? Follow @iconicompany.

---

📚 Read also

• Your Own CMS on GitHub: How Copilot Helps Write, Publish, and Announce Content
• A New Player in the Arena: Comparing MCP, A2A, and AGNTCY in the AI Agent Ecosystem
• AI Experience: How to Stop Competing with Thousands of Candidates
• AI is Not About Prompts
• How We Reimagined Developer Assessment: From Resumes to Voice AI Interviews