Automation
QwenWorkAutomation with real limits
The interesting part is not that it can run on a schedule. It is that failures report which step broke, because an automation you cannot debug is an automation you switch off in week two.
What can start a run
- Schedule: hourly, daily or weekly, in the workspace time zone.
- File event: a new sheet or document appears in a watched folder.
- Form or message event: a submission arrives, a keyword is mentioned.
- Manual: a button a member presses, which is how most automations actually start their life.
The numbers that decide whether it fits
| Limit | Value | Consequence |
|---|---|---|
| Runs per hour | 200 per workspace | A tight loop will hit it; batch jobs should be spread out |
| Active tasks, free | 10 | Enough to test, not enough to run a department |
| Retries | Automatic for transient failures | A deterministic failure will retry and then stop |
| Concurrency | Task-dependent | Two runs writing the same file will collide |
Designing one that survives
- Start manual
Build it as a button. Watch it run ten times before you put it on a schedule.
- Make the output idempotent
If the task reruns, it should overwrite the same file or append a dated one — never duplicate rows into a summary.
- Write the failure path first
Where does the output go when the input is malformed? A task with no failure path fills a folder with broken documents.
- Add an owner
Every automation needs a person who reads its failures. Orphaned automations fail silently for months.
Do not automate a decision you have not made manually ten times. Automation freezes a process. If the process is still changing weekly, you are freezing the wrong version of it.
Automation output usually lands in the workspace. Permissions decide who sees it.