Turn Agent Discoveries Into Repeatable Workflows
Agents are useful for finding a path through unfamiliar work, but repeatable execution belongs in a bounded workflow.
Developed from a conversation between Pete Winn, Paul Itoi and Andy David
An agent can explore an unfamiliar task, make decisions and discover a route to an answer. That flexibility is valuable while the route is unknown. Once the same task must be performed again, however, repeating the open-ended search wastes effort and allows settled decisions to be reconsidered.
A workflow captures what the agent discovered as a sequence of bounded actions. It can direct the agent to complete one step, check in, then move to the next step and check in again. The agent still handles the parts that require judgement, but the surrounding process determines where those parts belong and preserves the work that has already been done.
Paul compares this to a robotaxi finding a house without a map. Letting the car explore until it reaches the destination is impressive the first time. Sending it out to rediscover the same house every day is unnecessary once its turns can be recorded. The agent should find the route, while the workflow makes that route dependable enough to travel again.