Problem it solves
Multi-agent runs show a single spinner or "processing" state. Users don't know if agents are running, blocked, or failing independently.
When to use
During any multi-agent run involving two or more concurrently operating agents.
When not to use
For sequential single-agent runs. The parallel state surface is specific to network execution.
Governing principle
Each agent's state is independent. A blocked agent must be surfaced even if other agents are still running. Partial failures must not be hidden behind an aggregate status.
Required Components
Interaction Flow
Network execution begins
Multiple agents start running. Each is tracked independently.
Network state surface active
The execution state surface shows all active agents simultaneously, each with their own state indicator.
Individual state updates
Each agent's indicator updates independently: RUNNING, WAITING (on dependency or approval), BLOCKED (on error), COMPLETED.
Attention trigger on anomaly
If any agent encounters an anomaly — behavioral drift, scope escalation, decision fork — an Attention Trigger surfaces it immediately, even while other agents continue.
Human intervention available
The user can interact with any individual agent's state: pause it, redirect it, or abort its portion of the run without stopping the whole network.
Governance requirements
Each agent's state transitions must be logged independently. Network-level state must not mask individual agent failures. Intervention events must be logged per agent.
Accessibility notes
Agent state changes must be announced via ARIA live regions. When multiple agents update simultaneously, updates must be queued to avoid overwhelming assistive technologies. State must not be communicated through color alone.