Most teams don’t lose work inside a step. They lose it between steps.
Think about a relay race. It’s not just about how fast each runner is. It falls apart at the baton pass, and a fast first runner who drops the baton still costs the team. A real handoff includes all three parts: the baton (the output from the last step), the runner (whoever owns the next step), and the lane (the rules for what “ready” means).
How it shows up
In a workflow, a handoff is often where the work needs a different kind of judgment: a salesperson collects the client need, a designer turns it into a concept, operations turns it into a schedule. We talk about three types. A role handoff happens inside the same phase, usually person to person. A functional handoff moves across functions, like sales to design. An external handoff leaves your team and waits on a client, vendor, or outside reviewer, and it’s usually the hardest, because someone has to own the waiting and follow-up. This matters even more with agents. An AI finishing a draft and needing your approval is a handoff. A subagent reporting back is a handoff. When the AI hits a decision that needs judgment, you want a human in the loop handoff instead of a guess.
Why you care
Handoffs are where messy work hides. Vague input means the next person rebuilds what happened. An unclear owner means everyone assumes someone else has it. A non-standard output means the next step starts with cleanup. When we map work, handoffs become milestones that show where agents can run in parallel and where a human needs to review. That’s why delegation routing is more than who gets the task. It’s who gets it, with what context, at what moment, and with what definition of done.