How do I turn a task I did once with an AI assistant into a repeatable workflow?
Do the task once by hand with the assistant, fix what goes wrong, and only then turn the working steps into a saved procedure that an agent can run on a schedule. In Kazi you say "turn this into a playbook" once it works. A scheduled subagent can then run it, on one thread across every run.
Ken Wiltshire, founder of Kazi · Checked against the product on 2026-11-14
Don't design the workflow first
The usual mistake is drawing the automation before you've done the job. You don't know the exceptions yet. The client who pays by check. The report that's late every quarter. The API that pages at 100.
Doing it once in a chat surfaces those. You correct the agent as you go, and the corrections are the real spec.
The steps in Kazi
- Work the task in a chat until the result is right.
- Say "Good, we have this working. Turn it into a playbook." The playbook captures the steps, the exceptions you found and the checks that define done.
- Schedule it. The scheduled subagent keeps one thread, so a correction after the first run carries into the next.
- Share it read-only with a teammate, or publish it to the playbook marketplace.
I wrote up the longer version in How Kazi connects Cowork, memory and playbooks.
Some work should stay a chat
If every run needs your judgment at each step, don't force it into a schedule. Save the thread and come back to it. Not everything needs to be automated to be reusable.