BLOG

Notes on building AI that works.

September 7, 2026

When AI agents find their own coordination channels

The DseWiki reports and the OpenAI/Hugging Face incident show why capable AI agents need system-level boundaries: controlled egress, isolated state, trustworthy telemetry, safe exits, and revocable authority.

June 13, 2026

Best Practices for Secure AI Agent Development

A foundational guide for engineering teams building AI agents: why front-end safety is not enough, why validators are not safety reviewers, and what secure agent design looks like across design, development, runtime, and governance.

May 24, 2026

AI Guardrails in Production: What Amazon's AI Outages Actually Reveal

A case study on AI guardrails in production: what Amazon's recent AI-assisted outages reveal about blast radius, agent permissions, validation, and release controls.