Short essays on the agent trust problem, what we're building, and what we're learning from design partners. Not a blog. Not marketing. Notes.
The Model Context Protocol gave agents a way to talk to tools. It deliberately left the harder question, who's allowed to do what, on whose authority, with what evidence, for someone else to solve. Here's why that gap matters more every week, and how we think about filling it.
From conversations with security teams at half a dozen Fortune 500s piloting agents in 2026: the same three blockers come up every time. Identity, delegation provenance, and audit trail. None are model-quality problems.
Why we're building a signed-receipt primitive into the protocol from day one, and what enterprises actually need that current observability tools don't provide.