context-architecture
6 articles with this tag.
Authority Is Granted in Four Places
When a team goes looking for how their agent got permission, they open the identity system. That is one surface of four — and the only one with an owner, a review cycle, and an auditor.
Trust Is a Checklist, Not a Feeling
A few weeks ago, my own system tried to hand me a fake — fluent, plausible, and wrong. A checklist question caught what inspection never would have. Here's the pre-flight for any agent with persistent memory, before it touches production.
Where Governance Can Actually Reach
Agent memory isn't one thing — it's four distinct surfaces, each carrying different privileges, each failing differently, each needing its control in a different place. Here's the map: every door, and where the lock goes.
The Prompt Is No Longer Where the Behavior Lives
On July 28, the protocol connecting AI agents to their tools stopped tracking who was talking — on purpose. It wasn't an isolated decision. Every layer of the stack is making the same move, and your audit surface hasn't followed.
Your Agents Remember Things Nobody Reviewed
Your system prompt went through security review. Your prompt library is versioned. The memory store now driving your agents' behavior went through neither — and your name is on the output either way.
Your Corpus Is Lying to Your Model
When an AI system gives a confidently wrong answer, everyone audits the model. Almost no one audits what you fed it. The failure is usually in the corpus — and corpus quality is measurable.