Generation is becoming abundant. Confidence is becoming the bottleneck.

For most of the history of software, output was the constraint. You wanted more code, more analysis, more drafts, more coverage, and the limit was always how much a capable person could make in a day.

That limit is gone, or near enough. What did not change is the other side of the equation: how much a person can actually read, check, and be willing to put their name on. One of those curves went vertical. The other one did not move at all.

Nonsense is easy. It announces itself, somebody laughs, it gets thrown away. The expensive failures are the ones that look like competent work, survive a skim, get approved, and surface weeks later attached to a decision somebody has already made.

Every organization now producing machine output at scale is quietly accumulating these, and almost none of them can tell you how many.

Every serious field solved this at some point, slowly and unglamorously, and then forgot it had. The machinery that made work trustworthy became invisible precisely because it worked. Machine output arrived with an enormous amount of speed and none of that machinery at all.

We think the missing layer gets rebuilt, that it will look nothing like a chat window, and that building it well will matter more than owning the largest model. That is the bet. The rest we are keeping to ourselves for now.