Human agency is usually treated as an attribute of the person. Someone possesses agency when they can form intentions, make choices, and act on them. But as decision-making becomes distributed between humans and increasingly autonomous machines, meaningful agency also depends on the architecture surrounding the person.
A human may nominally remain in control while being presented with decisions they cannot meaningfully evaluate, actions they cannot reverse, or processes they cannot inspect. Conversely, substantial machine autonomy may coexist with strong human agency when objectives and authority remain clear, consequential actions are visible, intervention remains possible, and responsibility is connected to genuine decision-making power.
In this sense, human agency becomes partly a system property. It depends on what the architecture allows the person to know, authorize, challenge, interrupt, change, and reclaim. Designing for human agency therefore requires more than inserting approval buttons or requiring periodic review. It requires asking whether the relationship between human and machine preserves a meaningful human role in determining what happens.
This creates a useful principle for agentic AI: the quality of a system should be assessed not only by what the machine is capable of doing, but by what forms of human agency remain possible around it. Machine autonomy and human agency need not be opposites. The architectural challenge is to increase the former without quietly diminishing the latter.