The demo that sells
I sat through a demo of an AI project-management tool recently. It watched the tickets, the commits and the call notes, and produced a clean status report in a few seconds. Green where things were green, flags where there were flags. It was genuinely well made. And it was solving a problem I do not have.
The report was the part I could already produce in an hour. The hour was never the constraint. What takes the real effort is knowing whether the report is true before I send it, and no tool in that demo touched that.
To be fair to the tool, it did the thing it promised, and did it well. My problem is not that it fell short. It is that the thing it does well was never the thing that was hard, and watching software do the easy part beautifully has a way of making you forget the hard part is still sitting there, undone, waiting for a person.
That was never the hard part
Look at what these tools mostly automate: status updates, meeting summaries, task tracking, a tidy dashboard. All of it is the visible output of project management, the packaging the job comes wrapped in. None of it is the job. The job is the judgment those outputs are supposed to reflect. Is "on track" actually true. What gets cut when the date will not move. Which of a dozen true facts is the one that will matter next week, and belongs at the top.
A tool that generates a faster report is polishing the surface of a decision it had no part in making. It makes the evidence of the work look better without touching the work.
You can hear it in the language around these tools. They promise to give you visibility, to keep everyone aligned, to surface risks. All of that is reporting. None of it is deciding. A risk does not become less dangerous because a dashboard displayed it more clearly, and two people are not aligned because a tool summarised the meeting they both misremember. The surfacing was never the missing piece.
A tool that writes the update faster is solving the part that was never slow.
Why it cannot reach the half that matters
The signal a real call needs is usually not in the data the tool can see. It is in a hesitation on a call, a developer who has gone quiet, a client whose last question was sharper than the ones before. A model can arrange everything that was written down, and arrange it beautifully. It cannot weigh what nobody wrote down, because that lived in the room and never made it into a ticket.
So the tools cluster around the half that demos well. A status report is legible, visual, easy to show off in a sales call. Judgment is invisible and hard to put on a slide. The market is automating the demonstrable half, which happens to be the half that was never the bottleneck.
There is nothing wrong with automating the reporting, to be clear. I am glad the update writes itself. The mistake is calling that project management, and quietly implying the judgment underneath it comes along for free. It does not. You have automated the paperwork of the job and left the actual job exactly where it was.
What I would actually want
Not something that hands me a finished artefact to rubber-stamp, which just adds a thing I now have to check. Something that helps me interrogate the situation, that argues back, that asks the question I forgot. The useful AI on a project is the one I drive to think harder, not the one that produces a report for me to nod at.
That tool is harder to build and harder to sell, because its value does not fit in a screenshot. It shows up as a slightly better decision three weeks later, which no demo can capture. So the market keeps shipping the version that looks impressive in the room and helps least once the room clears.
I do not think this stays true forever. Eventually someone builds the tool that helps with the hard half, that sits in the decision with you instead of typing up its conclusion afterwards. But it will be a harder thing to sell, because you cannot screenshot better judgment. Until then, the honest move is to treat these tools as very fast clerks rather than project managers, and to keep the deciding firmly on the human side of the desk.
So the next time a tool promises to run your project management, ask it the only question that separates the useful from the decorative: does it help you decide, or does it just help you report the decision faster? Which half is it really automating, and which half is still yours?