Nodsend field notes

Engineering accountable autonomy.

Practical patterns for putting human authority into production agent systems.

Architecture01

Why approval is a control-plane concern

A model asking itself for permission is not the same as an application-enforced decision boundary. Here is how to separate intent, authority, and execution.

9 Aug 20268 min read

Read article

Integrations02

Durable human input across agent frameworks

LangGraph interrupts, CrewAI feedback providers, and guarded AutoGen tools expose different extension points. The approval lifecycle underneath them should remain the same.

9 Aug 20269 min read

Read article

Operations03

Exactly once starts with an atomic decision

Concurrent clicks, network retries, and repeated events should never execute a consequential action twice. Design for duplicates at every boundary.

9 Aug 20268 min read

Read article

Engineering accountable autonomy.

Latest field notes

Why approval is a control-plane concern

Durable human input across agent frameworks

Exactly once starts with an atomic decision

Follow the product as the decision layer evolves.