When AI agents do the work, who has the ball?

In 2025 and 2026, "AI agents" moved from demos into everyday work tools. Project platforms let you assign tasks to an AI teammate. Chat assistants can research, draft documents, update tickets and send messages. Vendors talk about digital workers sitting alongside human ones. See AI agents in project tools.

That raises a question every small team will face soon, if it has not already: when an agent is working on something, who has the ball?

Agents do tasks. People own outcomes.

An agent can draft the report, summarise the research or prepare the campaign. It cannot be accountable for whether the report is right, the research is sound or the campaign is appropriate. Accountability stays with a person.

The risk is that assigning a task to an agent feels like handing it off. It is not. It is closer to delegating to a very fast, very confident assistant whose work you must check.

Three states, not two

With agents in the loop, an action can be in one of three states:

1. A person is doing it.
2. An agent is doing it, and a named person will review the result.
3. It is waiting on someone else, human or agent, and you need to chase.

The middle state is new, and it is where accountability blurs. The fix is simple: every agent task has a named human owner who reviews it before it moves on.

Practical rules for small teams

• Name a human owner for every task an agent touches
• Review before it leaves the team: anything sent to clients or published gets a human read
• Track agent work like delegated work: what was asked, when, and when the result is due
• Do not let agents act on irreversible things (payments, deletions, external emails) without approval

The NIST AI Risk Management Framework talks about human oversight in formal terms. On a small team, it means someone always knows they are responsible.

Where a who-has-the-ball board helps

On a 2D board, an action stays in Mine while an agent drafts it, because you still own it. If the agent's output needs someone else's review, it moves to Other with that person's name. The board keeps showing a human owner at every step, which is exactly what agents make easy to forget.

The bigger point

Agents will change how much a small team can get done. They will not change the need for clear ownership. If anything, the faster work moves, the more it matters that someone is watching where it goes. See will AI replace project managers and AI task management in 2026.