A signal isn’t every interesting thing an AI could say. It’s a specific thing you decided was worth noticing, with a name, a purpose, and a reason to interrupt you. Anything looser just becomes another feed of noise.
Think about the dashboard light in a car. Thousands of things happen inside it, but you don’t need a report on every spark and temperature change. You need a clear light when the oil pressure matters or the tire pressure drops. A signal is that light saying pay attention here.
How it shows up
In a WorkDesk-style system, you define what you care about and the system watches the right source material for it. A transcript might contain a client risk. A log records what happened in a system. A weekly review might reveal a stuck project. A support inbox might show the same complaint three times. When the pattern appears, the signal surfaces where a person can act on it. Signals belong close to workflow, because a signal is only useful if it changes what happens next: a budget concern in three meetings might change the next sales conversation, or a repeated question might become a new SOP.
Why you care
The discipline is choosing the lights before you start driving. Too many and you stop trusting the dashboard; too few and you miss the thing that needed action. Signals can be powered by automation, but you still decide whether something’s actually important and whether the definition needs to get sharper. A good signal turns buried context into timely attention.