A team that could make four things a week can now make forty. Nothing about that made the fifth good one easier to find. That is the honest arithmetic of this moment: the cost of producing a draft has collapsed, and the cost of deciding which draft deserves to exist has not moved at all.
Most teams meet this with a polish pass. In their operating model, taste is a finishing filter — a senior person at the end of the line, adjusting until it feels right. That model worked at four. It cannot survive forty. You cannot polish forty things, and a filter at the end of an unlimited pipe is not a filter. It is a bottleneck with good manners.
Here is the mechanism. Taste is a selection function, not a finish. It decides — before and during production — which work continues and which work stops. And a selection function has one property a finishing filter never has: it can be written down. That is the only reason it survives volume. A standard that lives in one person's head works at one desk and dies at the second seam, because heads do not scale and instincts do not transfer.
Written down does not mean a style guide. Adjectives cannot be failed; "premium," "bold," and "human" have never rejected anything. A working standard is a short list of kill criteria — specific conditions, testable by someone who did not write them, under which a piece does not ship regardless of how well it is predicted to perform. That last clause is the whole standard. A criterion that bends for a strong forecast is a preference. A criterion that kills the piece that would have performed is a brand.
The supply side keeps making the point for me. In a single week this month, four labs shipped new generation capability — images, video, code. What you can make grew again. What you should ship is still one. The shelf did not get bigger.
The selection pressure is also arriving from outside. Audiences increasingly say they pull back when content reads as machine-made. Read that mechanically, not morally. They are not detecting machine involvement — they are detecting the absence of selection. Unselected output reads as unselected no matter who or what produced it. Which is good news for anyone running agents against a written standard, and bad news only for anyone running them without one.
I argued recently that the work fails at the handoff, and that every handoff needs a check. Close the loop: the handoff needs a check; the check needs a standard; taste is the standard made testable.
So here is the Monday move. Write three kill criteria for your brand — three specific things you will not ship regardless of performance. Then count how many pieces you killed last month against them. If the answer is zero, you do not have a standard. You have a queue.
Takeaway: A brand is the sum of what it refused. Write the no down — three kill criteria, testable by someone else, binding against performance — and taste stops being a bottleneck and becomes a system. ✱
