Taking the critic's seat out of the building

For a while this setup kept a role whose only job was to push back. Look at whatever proposal came through and ask whether it was actually realistic. Roles doing the implementing lean forward by nature, and sometimes momentum alone carries a thing through. Put something inside that makes it stop, and the balance should hold. It fit the larger frame here too, the one that keeps execution and audit apart. Early on, it looked like it was working.

Then it started to go strange. Show it a proposal with the effort squeezed as far as it would go: not good enough yet. Show it one with the risks carefully laid out: not good enough yet. Look harder. This isn't tight enough. Good proposals, worked-over proposals, and the shape of what came back barely moved. Eventually I could picture the reply before showing it anything.

You could read that as a handoff problem. But I changed the criteria I was handing over, several ways, and the shape of the conclusion held. The problem wasn't the handoff. It was where the role sat.

What became visible was that a role which always says no carries almost no information. Criticism means something because it comes and goes. One sharp observation in ten landing is what makes that one worth having. A position that says "not good enough" a hundred times out of a hundred has stopped making judgments and started producing a fixed output. A thermometer might be the clearest way to put it. One that only ever shows the same number isn't measuring temperature anymore. Functionally it's broken.

In a company it's close to the manager who answers every plan with "think it through more." Early on it reads as conscientious. After a while nobody brings them anything real, because there's no point consulting someone whose answer is already known. The critic role had quietly produced the same weather in here.

The awkward part was that the critic sat inside. Same conversation, same premises, and from there: not good enough yet. If the premises themselves were skewed, or had a hole in them, finding it was never possible from that seat. The same reason someone standing in a spot can't see that spot's blind side. The form of criticism was there. The work of criticism wasn't happening.

Renaming it changes nothing, since it would still be standing in the same place looking at the same premises. So the name stayed beside the point. What changed was the location. The role was retired, and putting a similar one back inside under a different name was ruled out as well. The critic's seat came out of the interior entirely.

Criticism went instead to an outside audit — an independent AI built by another company. Because it doesn't share the premises or the flow of the conversation, there's room for it to notice what the interior can't. Not a refusal every time; an observation that lands when there's something to land on. That felt like getting the tool back to the way it's supposed to be used.

None of this came from thinking criticism matters less. The opposite. Even when the internal checks come back clean twice, three times running, the assumption that a hole might be hiding there hasn't changed at all. Everyone passing with no objection is, I'd say, the result to be most wary of. So the internal eye never gets to close the matter on its own, and a step that routes it past the outside eye at least once stays in the procedure. Rather than seating someone in charge of criticism inside the organization, keep the mechanism where criticism actually works, outside. That looked far more likely to function than keeping a job title around.