Articles & write-ups
Notes on agentic AI architecture, AI security and production LLM systems, written from what actually breaks when these systems leave the demo and go live.
MCP Security Risks: Why AI Tool Integrations Need a CISO's Review, and how to control authority, authorization and audit across the tools your AI can reach.
Beyond Vector Search: Engineering Reliable Enterprise RAG with GraphRAG and Contextual Retrieval, and how to decide which techniques a use case has earned.
Why Traditional Security Tools Fail Against Autonomous AI Agents, and what a zero-trust, behavior-aware security architecture looks like in a finance example.
Beyond the Perimeter: Securing Agentic AI Workflows and Autonomous Multi-Agent Swarms, and why the security boundary now sits wherever an agent can exercise authority.
The Economics of Autonomous AI: RLVR, test-time compute, secure enclaves and the trade-off between intelligence, cost and risk.
The Agent Card Problem — why trusting another company's AI agent based on its Agent Card is riskier than it looks.
The Demo-to-Production Gap — why your AI agent works in the room and fails once real users, real data and real failures show up.
The topic pages cover the broader ground: