A workflow you dispatched is halfway through when it reaches a decision you would normally have been asked about — which of two shapes an output should take. Your session is responsive throughout. What happens?

The principle — dispatching a run is committing to its decisions in advance. Everything it needs from you has to be in the script before it starts.

Why the key is correct — the limit is stated as flatly as it can be: no mid-run user input. Read against how a run works, that settles the rest. Claude writes the script for the task you describe, and a runtime executes it in the background while your session stays responsive — so the conversation you are still using is not the run's control surface. The workflow runtime executes the script in an isolated environment, separate from your conversation, and intermediate results stay in script variables instead of landing in Claude's context. A decision between two shapes of output is therefore a thing the script has to already know, whether that means encoding the choice, making it a parameter, or splitting the run in two so the second half is dispatched after you have seen the first.

Why the others are wrong — the first has the run waiting on an answer it cannot receive. The second expects a question that will never be asked. And the third mistakes a responsive session for a steering wheel: it stays usable because the run is elsewhere, which is precisely why you cannot reach into it.

Remember this — decide it before you dispatch, or split the run so you decide between the halves.

Sources — Anthropic's Claude Code documentation on orchestrating subagents at scale with dynamic workflows.