Zines / Death by Choices

Let’s take a test.

I will ask you to draw a face. You draw an oval. Two dots. A curve.

That’s a face. Done. Go touch grass.

I then go and ask people one question.

If you could add one thing to your face, what would it be? Be creative.

A cyclist wants eyes on the side because cars come from the side. A founder wants a second mouth. Multitasking, you see. Lunch and a call. One stone, two birds.

Ignore the silliness of this example, but if you just look at the reason, it might make sense. If it doesn’t, I’m hopeful it will by the end.


Your turn

You are the observer who can see how everyone’s request shapes the face. Every click is the next person in line, making their case.

Drag it around too. Some of them wanted things on the back.

drag to turn it around ↻
features: 0states to support: 1last review: —

Did you feel it?

Now read the list and find the person who was wrong. Play along, we are almost there.

Let’s say they wouldn’t do it to themselves. But in a sim game, they wouldn’t mind it for the advantage they listed.

Every ask made sense to the person asking.

Yet nobody asked for this face.


Why a face

You’ve been reading faces since the day you were born. You don’t need a design review to know that one is wrong. You just know.

That’s why I picked it.

Software doesn’t have a face.

Nobody sees the whole thing at once, so the monster sits in plain sight:

  • The Homer. Homer Simpson got to design a car once. Every feature he ever wanted. It cost $82,000 and sank the company.
  • The TV remote. You use four buttons. Yet it has fifty. Except the Apple TV remote.
  • The hotel shower. Zero ways to get warm water yet so many knobs.
  • The settings page. Nobody remembers the defaults, including the people who wrote them. I am guilty of this too.
  • The approval process. Each one a scar from a single incident in 2019.
  • The agent setup. Twelve MCP servers, four memory layers, three ways to run the same tool.

Every one of them looked like progress on the day it shipped 💪

Features add. States multiply.

Ten toggles is not ten things but 1,024 combinations.

2×2×⋯×2⏟10 toggles=210=1,024\underbrace{2 \times 2 \times \cdots \times 2}_{10\ \text{toggles}} = 2^{10} = 1{,}024

You’ll test nine. Your users will find the rest 💥

How many of you have experienced this in prod?


So why do we keep doing it

Each person sees their own request. The cyclist never meets the plumber.

Review looks at the diff. And the diff sure does look fine.

How often do you file a ticket to remove? Adding gets you a demo or the feeling of shipping something. Removing? It probably gets you a meeting about who might still be using it.


The computer says yes

Adding used to cost something. You had to write it, defend it, and stay up until it worked. The friction was good, it killed a lot of bad ideas early.

That friction is gone.

The computer says yes to the cyclist, the plumber and the product manager. Simultaneously. Worktrees per sub-agent, anyone?

Ask for a second mouth and you get one.

With a compliment: you're thinking exactly right. And:

  • a PLAN.md
  • a README on why it was always part of the vision

You’d never do this to a face.

So why do it to your product?

The best things I shipped this year were the features I deleted.


#DeathByChoices #FeatureCreep #TheHomer #Vibecoding #NobodyFilesATicketToRemove