Building

How to build the first version

Everything in a first version is a trade against time. These are the trades I’d make again.

1. Cut until it hurts, then cut once more.

   Jobs came back to Apple in 1997, found roughly 350 products, drew a two-by-two grid and shipped four computers. Profitable within a year. Your version of that grid is smaller and just as necessary.

2. One thing on the screen at a time.

   Every additional choice on a screen is a chance to leave. The phone is a one-thing-at-a-time device and fighting that costs you How to keep people coming back you’ll never see in a chart.

3. Ship the embarrassing version.

   Reid Hoffman’s line about being embarrassed by your first release holds, and it still feels bad every single time. The polished version shipped in nine months teaches you less than the rough one shipped in three weeks.

4. Constrain yourself on purpose.

   M50 Speed Run is 708 KB, the entire game, one file. The constraint wasn’t a limitation, it was the reason the thing got finished and loads instantly on a bad connection.

5. Scope to the life you have.

   Seven apps got built around a full-time job. That only works if each one is scoped to evenings and weekends from the start, not scoped to a fantasy about quitting.

6. For content apps, the content is the product.

   Over The Bar carries 1,741 real GAA clubs and 6,400 playable levels. The engineering took weeks; the content took far longer and is the part nobody can copy in a weekend.

7. Make the first sixty seconds the whole design job.

   Most people who leave never see minute two. Whatever the app is for, that has to be visible, understood and enjoyable before the first minute is out.

8. Don’t build accounts until you have to.

   A sign-up wall in front of an unproven product is a filter that removes the users you were trying to learn from. Let people play first and ask for identity when there’s something to save.

9. Instrument it before launch, not after.

   Analytics cannot backfill. A day that goes unmeasured stays unmeasured forever, so the instrumentation has to land before the users do, not in the tidy-up sprint afterwards.

10. Ship weekly, or watch it fade out.

    A slow cadence hides a dead product from you for months. Weekly is fast enough that you cannot lie to yourself about progress.