What People Use It For

A tool that can do "anything" is hard to start with. The quickest way to see what OpenClaw is good at is to ask what starts each job. An agent that only answers when spoken to is only half of what OpenClaw offers.

The rule

Every job starts with a trigger: a message, a clock, or an event.

• A message. You write in a channel and the agent answers or acts: looking something up in your notes, drafting a reply, updating a file or a piece of documentation in its workspace.
• A clock. Cron jobs and heartbeats wake the agent on a schedule with nobody typing: a summary every morning, a clean-up of old downloads every Sunday, a check-in every half hour.
• An event. Another system sends an HTTP request, called a webhook, and the agent acts on it: GitHub reporting a new issue, or a form on your website receiving a submission. The agent pulls out the useful parts and does something with them.

Once triggered, a job can be more than one step. Inside a single agent loop the agent can read something, decide, run a tool, look at the result and carry on until the whole job is done.

And when one agent is asked to cover too many areas, you can run several specialised agents side by side, say one for home and one for work, with routing rules deciding which agent handles which messages so each stays in its lane.

The shortcut

Turn any idea into a job by filling three blanks: "When [trigger], do [steps], then tell me in [channel]." If a blank stays empty, the idea is not ready yet.

Drills

Answer fast — trust the rule.

Remember

• Every job starts with a trigger: a message, a clock or an event.
• Shortcut: "When [trigger], do [steps], then tell me in [channel]."
• Trap: an idea with no trigger or no channel is a wish, not a job.