An Agent Automation is a standing instruction to an agent, plus the triggers that start it. “Every Monday at 8am, summarise the week ahead” is one: the brief is the instruction, Monday 8am is the trigger, and from the moment it exists your agent runs it without being asked. Each Agent Automation belongs to exactly one agent. It’s the unit you create, pause, and delete.

​
The parts

  • The brief. What to do, written in plain English. It combines with the agent’s own instructions on every run: the agent’s identity sets the how, the brief sets the what, and the brief wins if they conflict
  • Triggers. The conditions that start it. There are three kinds: a schedule (a recurring local time or a fixed interval), an event (something happening in a connected app or inside Cando), and manual (you press Run now). One Agent Automation can have several triggers, each toggled on and off independently
  • A sandbox. Every Agent Automation gets a private computer: a filesystem and shell that persist between runs. Files it made last week are still there this week, which is what lets it build on its own work

​
Three ways to create one

  1. Describe it in chat. Tell your agent the job; it proposes the Agent Automation and the triggers it heard
  2. Promote a chat. When a conversation turns out to be recurring work, turn it into an Agent Automation in one step. It keeps the chat’s context and sandbox
  3. Start from a Template. Prewritten Agent Automations on the Home screen; picking one creates your own copy to edit

​
What happens when it runs

Each trigger firing becomes a run, every run is recorded in Agent history, and each run produces a chat you can read and reply to. Nothing your agent does is off the record.