Parallel tasks

Ask the agent to work through independent items in parallel when each item can be handled with the same instructions. Good examples include summarizing a set of PDFs, comparing vendors against a rubric, or researching a list of companies. If one task depends on another task’s answer, ask for those steps in sequence.

​

Give each task a clear contract

Include the sources, criteria, output format, and what to do when evidence is missing. For example: “Review each selected proposal independently. Create a note for each with scope, price, delivery date, and risks. Cite the proposal, and mark missing information as unknown. Then compare the results in one table.” A relevant Skill can supply the shared rubric or format. A Canvas task batch can create up to 50 items at a time. Larger requests can be split into batches.

​

Watch progress

For Canvas task batches, the process tray shows per-item status. Open a task to inspect its conversation and result. Outputs can arrive progressively, or the agent can combine them after the individual reviews. Say whether you need one artifact per item, a combined result, or both.

​

Recover a failed item

Ask the agent to retry failed or incomplete items and retain completed results. Check the failure first: a missing source, unavailable connection, or provider limit may need attention before retrying.

​

Compare after the independent review

Independent tasks are useful for collecting comparable evidence. Ask the main conversation to combine their findings and explain differences after the individual tasks finish. Parallel execution can reduce waiting, but completion time depends on task size, available concurrency, provider limits, and retries. It isn’t guaranteed to take the time of one item. See Research a list in parallel for a complete example.