The coordination gap
Why do agent swarms
lose coherence?
Agent swarms are powerful when each agent has a clear scope and clean handoff boundaries. They collapse when agents overlap, when routing logic is implicit, or when nobody has defined what "done" looks like for each specialist.
The hardest problem in swarm design isn't the agent framework — it's the operational design. Which agents should exist? What scope does each one own? When does one agent hand off to another? What context travels with the handoff? What happens when an agent gets stuck?
These are operational design questions, not infrastructure questions. Henry answers them through the same methodology it applies to human operations: map the process, specify each step, define execution modes, and assess fitness for delegation.
The result is a swarm design where every agent has a structured specification — scope, inputs, outputs, handoff conditions, and quality criteria — before a single agent is deployed.
Agent scope from step specification
Each Henry step specification defines a bounded unit of work. Steps grouped by domain become an agent's scope. The specification boundary is the agent boundary.
Handoff logic from process mapping
Eli maps the relationships between steps — dependencies, sequences, and parallel paths. These relationships become the routing logic that determines when Agent A hands off to Agent B.
Escalation paths from Fit Assessment
Ed's Fit Assessment identifies which steps an agent can handle autonomously and which need human oversight. These assessments become the escalation paths in the swarm routing — deliberate, not reactive.