An automation is not finished when it works on the happy path. It is finished when the underlying service is unavailable, a device misses an event, or somebody uses the physical switch instead of the dashboard and the result is still unsurprising.
That is the standard I use with Home Assistant and Lutron switches. The reliable part of the system should remain reliable without requiring every other component to be healthy.
Separate control from coordination
The switch should handle the immediate action. Home Assistant is better suited to coordination: schedules, occupancy, scenes, notifications, and integrations. That division keeps a routine light action from depending on a complicated automation chain.
Assume events can be missed
Automations often begin with an event: a button press, a state change, or a time trigger. Events are easy to reason about in a demo and easy to lose in a real system. A restart, a network interruption, or a device that reports late can leave the system in an unexpected state.
For important behavior, I prefer conditions that describe the desired state rather than assuming every preceding event arrived. If the goal is “the outside lights should be on after dark when somebody is home,” the automation should be able to evaluate that state again.
Make recovery visible
Notifications are useful when they describe a decision or a failure, not when they report every routine event. I want to know when an automation could not complete, when a device has been unavailable for too long, and when a fallback was used.
Keep the fallback boring
A good fallback is usually less clever: the physical control still works, a schedule can be retried, and a failed notification does not prevent the light or scene from changing. The more important the action, the less it should depend on an elaborate chain of integrations.
The useful test
For each automation, I ask:
- What happens if Home Assistant restarts?
- What happens if the network is unavailable?
- What happens if the device reports the wrong state?
- Can the action still be performed manually?
- Will a failure be obvious without becoming noisy?
These questions produce quieter automations and a system that is easier to trust.
Current as of September 2026. This describes design principles, not a drop-in automation package.
Leave a Reply