Start from a plain-English description
Open the AI sidebar on the workflow canvas and describe what you would like to automate. One or two sentences is enough. You can name an app, a trigger condition, and what the workflow should do next.
Keep the goal concrete. Instead of "make my work easier," try "when I press a Stream Deck button, post a summary of today's Linear tasks to Slack." The AI uses your connected apps and recent workflows to ground its suggestion.
- Mention the trigger you want: a Stream Deck button press, a schedule time, a webhook, or an app event.
- Mention the apps involved: Slack, Spotify, GitHub, Notion, Discord, or any of the 2,000+ integrations.
- Mention the end result: send a message, update a status, run a local command, or change a hardware state.
Answer clarifying questions
If your description leaves out details that matter, the AI will ask one or two short questions before it builds anything. Answer directly, then move on.
These questions help the AI pick the right provider, the correct action, and the appropriate variable wiring so the generated workflow does not need heavy edits later.
Review the generated workflow before you run it
The AI places nodes on the canvas, connects them, and chooses icons and labels that match the task. The workflow is a draft until you save it.
Check that the provider, action, and target settings match what you intended. If something is wrong, adjust the description and ask the AI to regenerate, or edit the node on the canvas directly.
- Verify the trigger node uses the right Stream Deck button or schedule interval.
- Check that action nodes use your connected accounts, not a placeholder.
- Open the variable picker and confirm values flow from earlier steps to later steps.
Run it and improve from there
Run the workflow once from the app to confirm it produces the expected result. When it works end to end, assign it to a Stream Deck button so you can trigger it from hardware.
A good AI workflow is not finished on the first try. Use the first version as a scaffold, then add error handling, timeouts, and fallback branches as the automation becomes part of your routine.