A grounded approach to agentic development and observability in the AI era

The blog post explores the topic of AI observability and agentic development, focusing on building an AI-powered SLO remediation workflow with tools like Groundcover, Claude Code, and MCP, while addressing the security risks associated with AI agent setups known as the "lethal trifecta." This framework identifies potential vulnerabilities when AI agents have access to private data, exposure to untrusted content, and the ability to externally communicate. The post discusses projects like Gas Town and Wasteland that exemplify these risks and presents a demo that illustrates how such risks can be managed in a controlled environment. The demo involves deploying a buggy microservice to EKS, using Groundcover for observability, and running Claude Code to autonomously detect, diagnose, and file incident tickets for SLO breaches. It emphasizes the importance of human oversight in AI workflows to mitigate potential security threats. Additionally, the post highlights the need for careful consideration of network boundaries and security measures when integrating observability solutions with external tools, especially in regulated sectors.

No tracked trend matches for this post yet.