Coordination Loop

Assembly operates as a continuous loop rather than a linear pipeline.

Coordination emerges from the repeated interaction between intent, planning, and actuation, driven by execution feedback.

The coordination cycle

At runtime, Assembly follows a simple cycle:

• Intent defines what should happen

• Planning proposes how it could happen

• Actuation executes actions in the real environment

• Feedback updates the system state

This cycle runs continuously, allowing the system to adapt without restarting execution.

Event-driven behavior

The coordination loop is event-driven. Changes in execution state, environment conditions, or resource availability trigger updates to plans and, when necessary, intent.

This approach avoids rigid scheduling and allows the system to react to partial success, delays, or failures in real time.

Local vs global failures

Assembly distinguishes between local failures and global failures.

Local failures affect individual tasks or executors and can often be handled through replanning. Global failures impact shared intent or system-wide constraints and may require intent updates.

This separation helps prevent cascading breakdowns in multi-agent systems.

Preventing cascading effects

By continuously isolating failures and propagating feedback through the loop, Assembly reduces the risk of system-wide collapse.

Coordination is maintained not by enforcing perfect execution, but by adapting faster than failure spreads.