Agent babysitting
The unplanned work of watching, steering, and rescuing a coding agent mid-task, instead of handing off a task and reviewing a finished result.
Agent babysitting is delegation that failed halfway. The agent needs a human there the whole time. Approving each step. Re-explaining the task. Untangling a wrong turn. The task involves an agent, but the engineer never gets their attention back.
The term is meant to sting. Babysitting time is the honest bottom line in any claim that agents save time. An agent that writes code in five minutes but needs forty minutes of care did not save half an hour.
The fixes are structural. Clearer task specs. Places where the agent can fail safely with no human catching it. Machine checks that reject bad work on their own, so the human only sees output worth reviewing.