Blog
Notes on deciding what to build, written while building Appray.
-
29 October 2026
I thought Appray had no competition. I had the wrong rivals in mind
Search the App Store for anything like Appray and you won't find much. I spent twelve days in June building a feature for a rival Appray doesn't actually have, before four months of support emails told me who it was really losing to. Read -
26 October 2026
People asked for a Face ID lock. That wasn't the actual problem.
Eleven emails over four months asked for the same lock screen. I'd nearly built it before rereading what people actually described, which turned out to be an unlocked phone changing hands for ten seconds, not a stolen one. Read -
22 October 2026
Appray has no analytics. Here's how I know what to fix anyway.
No account, no sign in, nothing leaves the device, including to me. That pledge means I genuinely don't know what's working, and twice this year the way I found out anyway took six weeks instead of an afternoon. Read -
19 October 2026
When is an MVP actually ready to launch?
I had a working version of Appray by late July. I submitted it in September. The six weeks in between didn't make the app better, they made me feel less sick about pressing submit, and I can prove it. Read -
15 October 2026
How many user interviews do you actually need before you build?
I did fourteen interviews before I wrote a line of Appray and felt thorough. I was wrong about the thing that mattered, and the number had nothing to do with it. Read -
12 October 2026
Appray used to run on GPT-4. I pulled it after three weeks.
For its first three weeks, Appray's verdict engine was a GPT-4 API call, not a rule. Then I ran the same eight answers through it forty times and got fourteen different verdicts back. Read -
5 October 2026
One-time payment or subscription: how I actually decided
Appray charges 4.99 dollars once for Pro, not monthly. I started out certain it would be a subscription because every app I admired ran one. Two spreadsheets and one imagined cancellation screen later, I'd changed my mind. Read -
1 October 2026
Should I translate Appray into other languages?
Twenty six support emails arrived in German, Croatian or Italian, always written in careful English. Germany is my second biggest market and nobody there has asked for a translation. What stopped me wasn't cost, it was Croatian grammar. Read -
28 September 2026
I built the feature people asked for most. Then came four months of silence.
Thirty four emails asked for a way to export a verdict, more than any request except Android. I shipped it in March. Because I build a private app with no analytics, I had no way to know if it worked until I emailed twelve people myself and asked. Read -
21 September 2026
Two five-star reviews arrived the same week, asking for opposite things
One wanted the interview shorter, the other wanted it longer, both about the same eight rounds of questions. I build a private app with no analytics to settle it, so I had to read both reviews properly instead. Read -
17 September 2026
What I actually do with every feature request that comes in
There is no roadmap for Appray, just a plain note that grows and shrinks. The board I tried and abandoned, the three questions I actually ask, and a pattern that hid across three channels for eight weeks before I caught it. Read -
14 September 2026
How many people need to ask for a feature before you build it?
Someone told me the rule is five. I tracked every request for four months, and the count nearly cost me the feature that turned out to matter most. Read -
10 September 2026
Should I build an Android version?
Nineteen emails asking for Android since launch. The honest revenue maths said no. What actually changed my mind was reading all nineteen properly and noticing most of them weren't asking for what I thought. Read -
7 September 2026
Should you build a feature just because your competitor shipped it?
Three competitors added an AI priority score in the same six weeks. I got four hours into copying it for my no-AI app before I checked what my own reviews actually said. Read -
3 September 2026
When to kill a feature you already shipped
Appray hands out Kill verdicts for a living. It took me five weeks to give one to my own feature, and the number that proved it was pointless wasn't what finally moved me. Read -
31 August 2026
How to prove a feature before you build it
I once put a fake button in a test build to measure demand and didn't like how it felt. What I use instead, and the two-week rule that stops "prove it" quietly becoming "never". Read -
30 August 2026
A customer asked for a feature. Should you build it?
Someone emailed twice in a week asking for accounts. Here is the test I now run before building anything a user asks for, including the one I said yes to too fast. Read -
16 August 2026
The features you build because they are fun to build
Nobody sets out to build the wrong thing. There is a specific reason the unnecessary features always feel more urgent, and it is not laziness. Read -
15 August 2026
Feature prioritisation frameworks, honestly compared
RICE, MoSCoW, Kano, value versus effort. What each is genuinely good at, where each falls apart, and the problem all of them share. Read -
14 August 2026
How to decide what to cut from your MVP
My first feature list for Appray had thirty one items on it. The version that shipped had nine. Here is how the other twenty two went away, and which cut I got wrong. Read -
13 August 2026
Jobs To Be Done interview questions that get real answers
Everyone said they would use my first product. Nobody did. The problem was not the idea, it was that I asked people to predict the future instead of describing the past. Read