Automation: overview

• Suggested — OpenFeather proposes something it could do.
• Prepared — the action is fully drafted, such as a draft message, and is visible to you for review.
• Confirmed — the action was actually executed and the product reported the result.

• Preparing drafts such as messages or documents for your review.
• Planning and scheduling proposals.
• Recurring research and summarisation runs over authorised sources.
• Notifications when a scheduled run completes.

Every automated action consumes a credential, touches someone else's information, or is difficult to reverse. OpenbirdAI therefore requires an explicit authorisation step and reports the outcome, rather than acting optimistically.

1. 1Pick a low-risk, read-only task you already do manually.
2. 2Define its schedule and the source it should read.
3. 3Review the first few results manually.
4. 4Only then let it run on its own.

An automated run is only trustworthy if you verify it occasionally. Unsupervised automation drifts: a source changes shape, a scope narrows, a prompt becomes stale.

• Check the first few runs manually before trusting the schedule.
• Spot-check sources every few weeks.
• Turn an automation off when its output stops being useful, rather than ignoring it.

Automation pays off when a task is repetitive, low-risk, and already understood. It fails when the task is still changing or when a mistake would be expensive to undo.

• Good first candidate: a weekly read-only summary you already produce manually.
• Poor candidate: anything whose output goes straight to a customer unattended.
• Rethink it: a task you cannot yet describe precisely enough to prompt for.