Software Engineer | Applied AI Researcher | AI Workflows & Automation
Engineer. Researcher. Builder. I'm a Software Engineer and Applied AI Researcher interested in building reliable software, experimenting with AI systems, and understanding the engineering decisions behind modern technology. I write about software engineering, AI, developer tools, automation, and technical research. My work combines practical engineering experience with curiosity-driven experimentation and evidence-based analysis. Publication: https://letinix.com Portfolio: https://mdazlaanzubair.com
Markdown supported · ⌘↵ to post
Framing eval as a security boundary is exactly right: a weak eval is an open door for a tool-calling agent. I treat the eval suite like an allowlist, testing what the agent must refuse as rigorously as what it should do. I went deeper on vetting that boundary for MCP servers here: kartiknvjk.hashnode.dev/how-i-evaluate-mcp-server… — how are you testing the refuse-path without a real exploit to replay?
Julian Neagu
This is the part people miss with agent evals. A sandbox is only useful if every path around it is locked down too.