LangChain Blog: Two Types of Agent Authorization — Assistants vs Claws

LangChain Blog·medium signal

LangChain distinguishes two authorization models in LangSmith Fleet: Assistants (agents using the end user's own credentials) and Claws (agents using a fixed set of credentials). This is a critical architectural distinction for anyone building multi-tenant agent systems — the authorization model determines blast radius, audit trails, and compliance posture. The post provides patterns for each approach.