Notes from the
trust layer

Engineering deep-dives, architecture decisions, and perspectives on what it takes to let software act on its own, written by the people building it.

ProductEngineeringArchitecturePerspective

LatestProductJune 9, 20267 min read

The approval workflow that finally got risk teams to say yes

Risk teams don't block agents because they hate automation. They block them because nobody could answer 'who approved this?' Here's the workflow that changed the answer.

Dana Whitfield

Head of Product

Read the post

EngineeringMay 12, 2026

Inside the Policy Enforcement Point: sub-10ms decisions at the gate

How we built a policy engine that sits in the hot path of every agent action and keeps the median decision at 8.4 milliseconds: compilation, caching, and the discipline of a deny-safe fallback.

Jonas Reinholt9 min read

ArchitectureApril 16, 2026

Hash-chained receipts: audit logs your auditor can verify

Traditional audit logs ask your auditor to trust your database. Signed, hash-chained receipts let anyone verify (independently, offline) that the record is complete and untampered.

Priya Natarajan8 min read

PerspectiveMarch 4, 2026

Why AI agents need passports, not API keys

The API key was designed for software that does what it's told. Agents decide what to do, and lending them your credentials erases the one thing accountability requires: knowing who acted.

Maya Chen8 min read

Prefer receipts to blog posts?

Everything we write about is running in production. Mint a passport and read your own signed audit trail instead.

Mint your first passport Talk to us