Agreement is the least useful thing a review produces
A plan that everyone approves can still be several plans. The approval hides which.
In 1974 Jerry Harvey published The Abilene Paradox, built on a family story. On a hot afternoon in Coleman, Texas, his father in law suggests driving to Abilene for dinner. Everyone agrees. The trip is miserable, and on the way home each person admits they never wanted to go. They agreed because they assumed the others did. Harvey's point was about organisations: they get into trouble less often through open conflict than through agreement nobody actually held.
Talking it through does not fix it
The obvious remedy is discussion. The evidence says discussion is worse at this than it feels. Garold Stasser and William Titus had groups choose between candidates (for student body president, in the original study) where the facts favouring the best one were split across members, so no single person could see the whole case. Groups spent their discussion on information everyone already shared, and mostly missed the better choice. The private knowledge that would have changed the answer rarely got said out loud.
That is the mechanism behind a quiet review. Each person brings an assumption that is obvious to them and so never mentioned: that rollback takes minutes, that the release goes to ten customers, that "ready" means the migration is done. The meeting circulates what the plan says, which everyone read. The assumptions stay private. The vote is unanimous.
Agreeing is cheap. Explaining is not.
The second problem sits inside each reviewer. Leonid Rozenblit and Frank Keil asked people to rate how well they understood everyday devices such as a zipper or a flush toilet, then asked them to write out how each one works, step by step. Ratings fell once people tried. They called it the illusion of explanatory depth. A thumbs up never triggers it. Asking someone to explain their reasoning does.
Gary Klein's premortem pulls the same lever. Before a project starts, the team imagines it has already failed, and each person writes down why, alone, before anyone speaks. Klein cites research by Deborah Mitchell, Jay Russo and Nancy Pennington finding that imagining an event has already occurred increases the ability to correctly identify reasons for future outcomes by 30%. Writing alone is the point. Each private assumption lands on paper before the group starts to converge.
Agents made approval cheaper still
None of this is new. What changed is volume. A coding agent will draft a launch plan, a migration design and a pricing model in an afternoon, and each lands in a channel where the cheapest response is an emoji. We have written about the reading side of this before: output grew, attention did not. When attention runs short, approval becomes the default result of a review, which is the condition Harvey was describing.
What we changed
Inputs is our attempt to make the cheap response the useful one. A page asks each reader for a choice and the assumption behind it, saved privately with what they were looking at, so the publisher sees the spread before the meeting instead of after the launch. Until today it was limited to a few teams. It is now on for every artifact on thing.
Two decisions follow from the research. Answers stay between the reader and the publisher, because the point is to collect them before the group converges. And when an agent answers on someone's behalf, which it now can, the answer is marked as an agent's. An agent agreeing for you is the Abilene trip with one more passenger.
It has real limits. Readers must sign in, which costs answers from people outside the team. Nothing reminds anyone to answer, so silence still looks like agreement. And collecting assumptions does not resolve them. A team can read every disagreement in the report and still drive to Abilene.