What's worth dispatching, what isn't
The right Dispatch task: long enough to make leaving worthwhile, well-scoped enough that it can finish without you. Bad Dispatch tasks: short and cheap (just do them at your desk), or open-ended and ambiguous (you'll need to redirect mid-flight, but you can't).
The pattern that works: well-defined inputs + well-defined outputs + pre-approved tools. "Take the 8 PDFs in ~/inbox/contracts/, extract dates and amounts, write ~/Desktop/contracts-summary.csv." Inputs known. Output known. Tools obvious from the verb list. Dispatch can run unsupervised.
The pattern that fails: "Look into our database performance and figure out what's slow." Open-ended. Will hit ambiguity. Will pause for clarification you can't provide. Save those for when you're at the desk.