LangChain Launches LangSmith Fleet for Enterprise Agent Management
Summary
- • LangSmith Fleet rebrands Agent Builder into a full enterprise agent management platform
- • Non-engineers can now create agents from a plain-language prompt description
- • Tiered permissions let teams control who can run, edit, or clone each agent
- • Agent identity model defines how agents authenticate with company tools on behalf of users
- • Two authorization modes: "Assistants" act on-behalf-of users; "Claws" have fixed dedicated credentials
Details
LangSmith Fleet: rebrand and expansion of Agent Builder
Agent Builder launched in October to enable knowledge workers to create agents via natural language. Fleet extends it into a full enterprise workspace covering creation, sharing, permissions, credentials, an activity inbox, and observability — reflecting how teams' needs evolved from building one or two agents to managing many.
Tiered permission system for granular agent access control
Three permission levels — Can Clone, Can Run, Can Edit — can be layered and applied to individual users or the full workspace. Permissions can be changed or revoked at any time, giving centralized IT or AI teams a governance mechanism before publishing agents organization-wide.
Two authorization types: Assistants (on-behalf-of) vs Claws (fixed credentials)
Assistants act on-behalf-of their end user, using each individual's own credentials when accessing tools like Notion or Rippling — so Alice can't access Bob's data. Claws have a fixed, dedicated set of credentials (often a purpose-built service account) that all users share, enabling consistent access control regardless of who triggers the agent. This distinction addresses credential bleed in multi-user agent deployments.
Channel support added: Slack, Gmail, Outlook, Teams
Fleet adds channel integrations so agents can be accessed via communication platforms. Assistants require a mapping of the end user's channel identity (e.g. Slack user ID) to their LangSmith ID to support per-user credentials, limiting which channels currently support on-behalf-of authorization.
Inbox enables human-in-the-loop approval of agent actions
The Inbox tracks all agent activity and surfaces actions that require human sign-off before proceeding. This targets high-stakes or irreversible tasks where full autonomy is not yet appropriate, providing an auditable trail of what each agent did and the reasoning behind it.
Governance and identity are now the hard problems in enterprise AI agents
Fleet's positioning reflects a pattern mirroring cloud infrastructure and SaaS: democratized creation generates sprawl, which generates demand for auditability, access control, and identity management. LangChain is betting the next competitive layer in agent platforms is enterprise governance, not model capability.
Key features, technical details, and strategic context from LangChain's LangSmith Fleet enterprise agent management platform launch, including the two agent authorization models.
What This Means
LangSmith Fleet marks a maturation point in enterprise AI tooling: the conversation has moved from 'how do we build agents' to 'how do we govern the agents we already have.' The explicit distinction between Assistants (on-behalf-of user credentials) and Claws (fixed service-account credentials) gives enterprise teams a concrete framework for designing multi-user agent deployments without credential bleed. For AI and IT leaders, Fleet offers a concrete answer to the audit, credential, and access-control questions that have blocked broader agent rollouts. Teams evaluating agent infrastructure should now weigh governance, identity model, and observability capabilities — not just developer ergonomics — as first-class criteria.