Project tools are hiring AI agents. So who has the ball now?

Every major project management vendor spent the past nine months putting autonomous agents inside their product. Almost none of them have answered the obvious follow-up question.

The shift is real and it happened fast. ClickUp shipped version 4.0 in December 2025, a full platform rebuild around a converged workspace tying tasks, docs, chat and AI together, with a "Super Agents" feature that runs workflows across a workspace while carrying context between them. monday.com spent early 2026 repositioning itself as an AI work platform and, per trade coverage from UC Today, built infrastructure that lets AI agents sign up, authenticate, and operate inside the platform as participants rather than as a chat sidebar.

Asana has AI Teammates. Adobe Workfront lets a project manager add an agent to a plan as an assignable resource, which is a sentence that would have read as satire in 2023. Across the first quarter of 2026, the category as a whole moved from tracking work to attempting to do some of it.

The good version of this

We should be fair, because there is a genuinely useful version. The most credible thing these agents do is reduce routine coordination: flagging that a schedule has slipped before a human notices, suggesting a workload rebalance, turning a meeting recording into a list of action items so that nobody has to be the person who volunteers to write the notes.

That last one alone is worth something. The action items from a meeting are exactly the commitments most likely to evaporate, which is why we wrote about follow-ups dying in notebooks in the first place. If a machine reliably extracts them, good.

The question nobody is answering

Here it is. When an agent is assigned a task, who has the ball?

This is not a philosophical puzzle, it's an operational one. Every project board in existence encodes an implicit promise: if a card is assigned to a name, that name is going to move it, and if it stops moving you can go and ask that name why. The whole social machinery of a project runs on that. An agent assigned to a card breaks the machinery in a specific way. The card looks owned. It has an avatar and everything. But there is nobody to ask.

Worse, agents fail quietly and confidently. A human who is blocked eventually says so, usually with visible irritation. An agent that hit an ambiguous instruction three days ago will either have done something plausible and wrong, or done nothing, and in both cases the card sits there looking exactly like a card that is fine.

We have written a lot about how projects die from small unnoticed silences rather than from dramatic failures. Assignable agents are a new and efficient source of small unnoticed silences.

Delegation only works if you can see it

Axtio's whole model is built on one distinction: work in your court, work in someone else's. The Other court exists because handing something off is not the same as finishing it, and pretending otherwise is how things vanish. We covered the psychology of that in The Psychology of the "Other" Column.

Delegating to an agent is delegation. It belongs in Other, alongside "waiting on the client" and "sent to legal", and it needs the same treatment: a visible marker that this is out of your hands, and a habit of scanning that column for things that have been sitting too long. The mistake is treating an agent handoff as a completion. It's a handoff, and handoffs come back.

Whatever tool you use, the practical rule is the same. If an agent is doing something for you, you should be able to look at your board and see (a) that it's out with the agent, and (b) how long it's been out. If your tool cannot show you the second one, the agent has not saved you work, it has moved the work into a place you can't inspect. That's the same failure mode as an invisible blocker, just with better branding.

A note on the numbers

You'll see a lot of adoption statistics attached to this trend, including a widely quoted monday.com survey finding that 82 percent of respondents use work or project management software to improve organisational efficiency. Read those with the usual caution: they are vendor surveys about the vendor's category, and they measure enthusiasm rather than outcomes. As of August 2026 there is not much independent evidence yet on whether agentic features actually shorten project cycles. Ask again in a year.

Where this leaves small teams

If you're two people and a contractor, most of this is not aimed at you. Agentic coordination features are built to solve the problems of organisations large enough that nobody knows what anyone else is doing, and they come attached to platforms priced accordingly. We have made the general case for that skepticism in what happens when project tools try to do everything.

Use the agents that genuinely save you a chore. Meeting notes, drafting, summarising a long thread, all fine. Be much more careful about the ones that take ownership of a piece of work, because ownership is the one thing your board is actually for. A tool that quietly blurs who has the ball has broken the only feature that mattered.

Sources

monday.com repositioning and agent infrastructure as reported by UC Today. ClickUp 4.0 release details via the ClickUp changelog. Category overview and the monday.com survey figure via Businessmap's 2026 trends report. Accurate as of August 2026.