Practical AI engineering for real workflows
havyai is where I write about the engineering behind useful AI systems: agents, context, tools, memory, and the workflows that hold them together.
I share field notes from what I build, the failure modes I run into, and the tradeoffs that change the result. It is for builders who want practical detail without the hype.
Latest
November 2, 2026
How to Test an MCP Server: From Tools to Agent Evals
Test your MCP server with unit tests, tool-selection evals, and complete agent scenarios. A runnable Python example grounded in custom connector experience.
July 19, 2026
Prompt Guardrails Are Not Security Boundaries
A prompt-injection test passed at the text level and still failed at the system level. A practical way to prove an agent's trust boundaries hold.
June 29, 2026
Context Engineering 101: What to Keep Out of Your AI's Head
The hard half of context engineering is subtraction: deciding what to keep out of your AI's context, not just what to put in.