Always-on AI agents like OpenAI's Dots: who owns the work they pick up?
At its DevDay at the end of September 2026, OpenAI introduced Dots: persistent agents that keep working after you close the tab, each with its own cloud computer and browser, reaching users through ChatGPT, Slack and Microsoft Teams, as reported by SiliconANGLE. The pitch is that you hand an agent a goal and it carries the job forward.
We wrote earlier about AI agents and who has the ball. Always-on agents push that question further, because the agent is not just answering a prompt; it is holding work over days.
The ball cannot sit with software
When an agent picks up a task, it is tempting to say the ball is in the agent's court. That is a trap. An agent cannot be accountable to a client, cannot decide when a trade-off is acceptable, and cannot be chased in the same way a person can. If the agent stalls, misunderstands or acts wrongly, a human still answers for it.
A workable rule: every agent-assisted action still has one human owner. The agent is a tool that person uses, not a new court.
How to show agent work on a board
On a 2D board, an action handed to an agent stays in Mine, owned by the person who delegated it, with a note that an agent is working on it. When the agent produces a draft or result, the owner reviews it and moves the action forward.
If the agent is waiting on someone else, say a client's reply it requested, the action belongs in Other with that person's name, not the agent's. The waiting is still on a human.
Watch for invisible work
Always-on agents can create work nobody sees: messages sent, files edited, tasks opened. Ask:
- What can the agent do without asking?
- Where is a record of what it did?
- Who reviews its output before it reaches a client or colleague?
Teams that keep agents in draft mode, with humans approving actions that leave the team, avoid most surprises.
Ownership is the scarce skill
As agents get better at doing, the human work shifts to deciding and owning: what should be done, by when, and whether the result is good enough. Those are exactly the questions a board built around ownership makes visible. For the wider trend, see AI agents in project tools and will AI replace project managers?