Skip to main content
An automation is a rule that launches an agent when something happens. It connects a trigger (what happened) to instructions (what to do) and permissions (what actions are allowed). Automations do not do the work themselves — they spin up agents that operate inside your environment and then report back with results. Key characteristics:
  • Event‑driven: Automations start from real signals like errors, tickets, or messages.
  • Goal‑oriented: Instructions focus on outcomes, not step‑by‑step scripts.
  • Controlled: Permissions define what the agent can and cannot do.
Automations vs agents: Automations decide when to act. Agents do the actual work.

​
Common use cases

Sentry error triage

Investigate production errors and propose fixes automatically.

Intercom support triage

Route high‑impact tickets to agents for faster resolution.

GitHub issue response

Draft responses, reproduce bugs, and suggest fixes.

Linear ticket automation

Triage and take a first pass at implementation.

PostHog session replay

Watch replays, identify root causes, and suggest code changes.

​
What happens when it runs

1

Trigger fires

A system event happens — a ticket is created, a new error appears, or a message arrives.
2

Agent starts

A coding agent spins up in a real environment with access to the right context.
3

Work happens

The agent investigates, implements, and prepares outputs.
4

Results are delivered

Outputs show up in the client where the request started.