There is a shallow version of the question "what remains human?" It fixates on which tasks people will keep doing by hand. Chapter 10 of my book argues that this is not the most important distinction.
Humans may type less. They may implement less boilerplate directly. They may supervise parallel flows of machine work, edit generated artifacts, compose systems from agentic parts, and step in mostly at moments of ambiguity or consequence. All of that is plausible. None of it names the deeper role.
A list of responsibilities
The deeper human role is responsibility. The chapter lists what that covers. Responsibility for standards. Responsibility for scoping. Responsibility for the boundaries within which machines act. Responsibility for deciding when evidence is sufficient. Responsibility for noticing when the system is producing motion without progress.
The last item matters most to me. A system that generates a lot of work looks busy. Someone has to judge whether the work moved anything. A machine can report its own activity. It cannot take the blame for it.
This fits a theme that runs through the whole book. Cheap generation shifts value upward, toward framing, discrimination, prioritization and the willingness to throw away seductive garbage. As output gets cheaper, standards do not become quaint. They become load-bearing. People who hold those standards are doing the human work, even if they never touch a keyboard.
A test you can run today
The chapter gives a fast test for whether responsibility has actually been placed, not merely assumed. Name the person accountable when a delegated system's output turns out to be wrong six months from now, after whoever wrote the prompt has moved to a different team.
If no name comes to mind, the responsibility was never assigned. In the book's words, it evaporated along with the context window that produced the output.
I like this test because it is hard to argue with. It does not ask whether your team cares about quality. It asks for one name. Teams often discover that the person who launched an agent assumed a reviewer owned the result, and the reviewer assumed the person who launched it did. Both are acting reasonably. Nobody owns the outcome.
The same logic explains why the book stays skeptical of both utopian and defensive stories. AI neither leaves engineering unchanged nor abolishes the need for engineers. It relocates value toward those who can design, govern and continuously improve systems of delegated work.
Here is the takeaway. Pick one workflow where an agent produces output that others rely on. Write down a name, not a team, for who answers for that output in six months. If you cannot do it, treat that as the next piece of engineering work, ahead of any new feature. Then write the name where the next reader of the system will see it.
This post draws on Chapter 10 of From Copilot to Colleague.