Build a routine

Routines are how you turn one-off operator help into reliable infrastructure. "Every Monday at 9am, give me a sales pipeline summary" is a routine. "Every weekday at 6pm, draft a release-notes update from today's GitHub PRs" is a routine.

This tutorial builds a daily marketing-metrics digest end-to-end.

Prerequisites

• A guild with a marketing-flavored operator (Solo or Marketing studio)
• An integration that gives access to a metrics source (Google Analytics or Mixpanel both work)
• A place to deliver the digest — a Slack channel or an email inbox (requires the corresponding integration)
• ~10 minutes

Step 1 — Open Work → New routine

Inside your guild → sidebar → Work → switch to the Routines tab → New routine.

You'll fill out:

• Title — "Daily marketing digest"
• Owner — pick the operator that should run it (your marketing strategist)
• Schedule — 0 9 * * * (every day at 9am UTC) or use the time picker
• Description — the prompt the operator runs each time

Step 2 — Write the description

The description is the prompt the operator receives every run. Be specific about output — operators are better at delivering exactly what you ask for than guessing.

Pull yesterday's marketing metrics from Google Analytics and Mixpanel:
- Sessions, new users, conversion rate (GA)
- Active users, key event triggers (Mixpanel)

Compare against the same day last week. Highlight changes >15%.

Produce:
- A 3-bullet summary (top win, top concern, recommended action)
- A markdown table with the raw numbers

Post the summary to #marketing on Slack.
Save the table as an artifact named "marketing-digest-YYYY-MM-DD.md".

Step 3 — Save and dry-run

Click Save. The routine appears in the Routines tab.

Click Run now to fire a one-off execution. This creates a fresh work item and the operator starts working. Watch the linked conversation for progress; tool calls, draft outputs, and the final delivery all stream inline.

If the dry-run posts to Slack and saves an artifact correctly, you're done — the routine fires automatically on its schedule from now on.

Step 4 — Monitor

Each scheduled run creates a new work item under the routine. The Routines tab shows:

• Last 10 runs with status (succeeded / failed / partial)
• Next scheduled run
• Average duration
• Credit consumption per run

Click any past run to see what happened. Failed runs include the error so you can fix the prompt or upstream issue.

Step 5 — Iterate

Bad output → edit the description, dry-run again, save. The next scheduled run uses the updated prompt.

Common iterations:

• Add more sources — append to the description, the operator picks up the new actions automatically
• Change delivery — swap Slack for email, or add a second target
• Tighten thresholds — change ">15%" to ">25%" if the digest is too noisy

Pause / resume

Routines tab → click the row → toggle Active. Pausing keeps the configuration but skips scheduled runs. Useful while you're iterating, or when you want to mute over a holiday.

What's next

• Work items — routines produce work items; learn the broader system
• Integrations and Actions — add more data sources for richer routines