Multi-Fleet Coordination

June 10, 2026 · View on GitHub

When teams run separate fleets (platform, GTM, support, security).

When you have multiple fleets

Different data boundaries (PII, production, customer)
Different platforms (LangSmith Fleet vs self-hosted)
Different autonomy policies

This is normal at F2+. Fleet engineering applies per fleet and between fleets.

Between-fleet rules

Rule	Rationale
No shared service credentials across fleets	Attribution breaks
Handoffs require typed contracts	Same as multi-agent
Each fleet has its own registry	No shadow catalog
Cross-fleet actions go through inbox	Sovereign control
Economics attributed per fleet	Chargeback clarity

Bridge from loop multi-loop

loop-engineering multi-loop coordinates loops within one repo/team.

Multi-fleet coordinates populations across teams:

Team A Fleet          Team B Fleet
├── daily-triage      ├── pr-babysitter
├── ci-sweeper        └── gtm-research
└── registry A        └── registry B
         │                     │
         └──── handoff contract ────┘
              (typed, audited)

Priority when fleets collide

Sovereign control (kill / pause)
Identity / compliance boundary
Budget caps
Choreography contracts
Topology changes

Anti-pattern: Monolith Fleet

One enterprise fleet for every agent everywhere. Simple early; brittle at scale. Prefer federated fleets with explicit handoff contracts.