How Agentic Workflows Know When to Run
A chat assistant has exactly one trigger: you. It works while you are typing at it and stops existing the moment you close the tab.
That is the real line between a chatbot and a working system. A system that only acts when asked is a tool; a system that knows when to act is staff.
Three Ways Work Starts
Beyond a direct request, work starts three ways: on a schedule, on an event, or when a watcher notices something changed. Almost everything an office repeats falls into one of the three.
Pick the wrong trigger and automation feels either late or frantic. Pick the right one and the work is simply done when someone goes looking for it.
The Clock Runs the Routine
A schedule fits any work that recurs whether or not anything changed. The morning briefing is the clearest case: it is assembled before your day starts precisely because you did not have to ask for it.
In FactoryOS, ingestion, maintenance, and report runs each fire on their own cron. The quiet hours do the recurring work, which is why agents should work overnight in the first place.
Events Beat Polling
An event trigger fires the moment something happens instead of the next time someone checks. An email arrives, a form is submitted, a workflow upstream finishes, and the next step starts itself.
Checking on a timer always loses to this. A renewal notice that sat unread since Tuesday is an event that should have fired on Tuesday.
Watchers Notice What Changed
A watcher compares the world now to the world last time and acts only on the difference. FactoryOS runs its ingestion this way: the pipeline is hash-gated, so a thousand unchanged files are skipped and the one edited contract is reprocessed.
The same pattern watches dates approach and thresholds get crossed. Nothing happens when nothing changed, which is what keeps an always-on system from becoming noise.
Triggers Do Not Grant Authority
A trigger decides when work starts, never how far it goes. A fired workflow can gather, draft, and stage everything, and still stop at the line where consequences begin.
That line is a human approval, presented with the agent's justification attached. The clock wakes the agent; a person still signs.
Why Timing Is the Product
Most automation demos show what AI can do when asked, and asking was never the bottleneck. The bottleneck is that nobody asked on time.
A workflow is agentic when it closes that gap on its own, starting the work while the humans are still busy. The output is the same document either way; the trigger is what made it early instead of late.
Look at your own week for the task that always runs late. Is anything actually hard about it, or is it just still waiting for a human to notice it is time?