Are you the operating system,
or have you built one?
Agents now report the status, write the draft, and run the check. So which meetings still need a person in the room? Only the ones where somebody decides something.
(The test for any recurring meeting: name the last decision it changed. If it only produced a document, an agent can do it, or nobody needs it. And when a number becomes the target, it stops telling you anything.)
Three meetings agents replaced. Did anyone pick up the judgment?
The PRD review meeting
Gone, because Claude writes a passable spec in 90 seconds. The judgment it produced (other senior people catching scope creep, reframing the problem, naming the risk) never moved anywhere else. Most teams simply stopped reviewing. The drift shows up in production three cycles later.
The competitive teardown
Gone, because every agent now does competitor research on request. The judgment it produced (someone forming a real opinion about why we are different) does not happen on its own. The agent gives you an even-handed summary. Having a view is a human job.
The status update Friday memo
Gone, because the tools now report progress as it happens. The judgment it produced (one person separating signal from noise and naming what was actually at risk) is worth ten dashboards. Most teams replaced it with raw feeds. Now everyone reads everything and nobody reads the situation.
Cycle A shipped 14 features. Cycle B shipped 8.
Same team size. Same tools. Somebody has to say why, and the answer changes what you do next.
If the cause is how the work was described, it shows up in the quality of the specs the agents were given. If the cause is one person carrying the cycle, it shows up in how the team runs. If nobody looks, next quarter chases a guess instead of the cause.
A year ago you could put the swing down to who was out that month. Now that agents do much of the building, headcount no longer explains it.
How long does a customer signal take to reach a roadmap change?
If a real risk shows up in week 1, when does the plan change? Two weeks means the team learns as it goes. Six weeks means the team reports. Eleven weeks means the signal got lost in a quarterly review deck.
The number is knowable. Take a handful of roadmap changes, find the signal that caused each one, and date both ends. Most teams have never measured it, which is why it slips: nobody tracks the interval, so nobody notices it doubling.
Right now, can you name the biggest gap in the system and who owns closing it?
When the team can do more than the product shows, that is one situation with one fix. When the product asks for more than the team can support, that is a different situation with a different fix. When how the team runs, how it builds, and what it ships all sit at the same depth, the work compounds without you in the room.
If nobody can name the current gap, that is the answer. One gap, one owner, every cycle.
Take two weeks off and don’t answer Slack.
When you come back, can you tell what happened from the system, or only from what people remember?
A real operating system tells you what shipped, what slipped, which calls got made and by whom, how the team did last cycle, which gap opened up, what the agents already handled, and what your lead chose to defer. Two weeks away is the cleanest version of the question: does any of this survive without me?
The work is to move judgment out of your head and into the system, one ritual at a time, until the answer to every question above lives somewhere other than you. Agents do the read, the draft, and the check. A person still owns each call. That is the job now, and it is what makes the two weeks possible.
If you are sizing up your own job: Hire for the PM job that exists now. If you brief the board: What you say in 60 seconds.