Some tasks must run in order. When agent B depends on what agent A produces, you run them sequentially: wait for A to finish, read its output, then launch B with that output included in the prompt.
This is still useful even though it doesn't save time. Sequential agents let you break a long task into focused stages, each with a narrow scope. I'll use a concrete example: a research agent might return lines of notes. You then feed those notes to a writing agent that turns them into structured documentation.
Each agent stays within its area of responsibility. The handoff between them is explicit, and you can inspect the output at each stage before the next one starts.