I get asked, occasionally, how I decide what to ship out of the lab. The honest answer is that I have four rules. Three of them are easy; the fourth is the one I keep failing at.
1. Build for someone you can name#
The fastest way to ship something nobody wants is to build for an imagined audience.
Every experiment in the lab starts with one specific person — me, a friend, an owner of a factory I've spent time in. If the spec sheet doesn't have a name at the top, I don't start.
It sounds limiting; in practice it's the opposite. Building for a named person produces decisions. "Would Pradeep use this on the floor?" is a sharper question than "would users find this valuable?"
2. The first version should be embarrassing#
A first version that doesn't slightly embarrass me is a version that took too long.
Embarrassment is a useful signal. It means the work is in front of real conditions before the polish has had time to hide what's wrong with the idea. The polish should come after — for the second version, when you know the idea is right.
3. Honesty about state is the brand#
The most premium move the lab makes is labelling a paused experiment as paused.
You can't fake working on something. The audience the lab is calibrated for — premium consumers, senior builders — has been to enough product launches to read between the lines of "we're focusing on customer success" and know what it means.
So when something is paused, the page says paused. When something is a sketch, the page says sketch. When something is shipping, the page says shipping. The motion design and the visual polish — all of which take real work — are there to honour the work. The labels are there to honour the audience.
4. The hard one: let things go#
Most of what I sketch dies. Most of what I prototype gets put down. Most of what I ship gets pushed into maintenance and then quietly retired.
This is the rule I fail at. I keep paused experiments running longer than they deserve, because a paused experiment with a beautiful page feels alive even when it isn't. The rule, when I follow it, is simple: a sunset announcement is a feature, not a failure.
I'm trying to write one a quarter.
A small lab is honest in proportion to how much it lets go.
That's the method. It's not magic. It's mostly just choosing not to fall for the failure modes that catch you when you don't.
The next essay will be about the first rule — why "build for someone you can name" is the rule the rest of the practice rests on, and how to find that person without inventing them.