Indie iOS developer. I write about App Store search, ASO mechanics, and what actually moves downloads. Evidence over growth hacks.
İstanbul, Türkiye
Joined December 2016
- Tweets239
- Following130
- Followers38
- Likes229
Pinned Tweet
I feel like nobody's shown what Gemini 4 Argon can do with something you actually use every day ❤️🔥
I gave it one image of an Istanbul çaydanlık and asked for an interactive 3D exhibit.
162,324 triangles, all generated in code. No Blender, no 3D models, no textures.
The steam, the pour, the cutaway with tea leaves and boiling water, the copper → silver → brass swap. One prompt, no fixes.
format inspired by @thebuggeddev 🫡
5 two-minute checks before you submit to App Review, all from this week's threads:
1. run your iPhone-only app on an iPad simulator (2.4.1)
2. hangs only in review? test on an IPv6-only network (2.5.5)
3. no rating prompt during onboarding (Apple's HIG: wait for real engagement)
4. the amount you bill is the biggest price on your paywall
5. demo login + every paywall variant written in Notes for Review (2.1, 2.3.1(a))
your iPhone-only app still gets reviewed on an iPad.
guideline 2.4.1: iPhone apps should run on iPad whenever possible. on iPad it opens in a scaled iPhone window.
if login, loading or layout breaks there, that's a rejection. someone in my feed waited 9 days to learn that this week.
2-minute check before you submit: run it on an iPad simulator
Apple updated the App Review Guidelines yesterday. 2 new rules matter if you ship apps:
2.5.19: no hidden or dormant features. code that's off during review and switched on later can get the app removed and the account terminated
2.3.14: the new header and search images can't show prices, discounts or claims you can't verify
Apple doesn't mention paywalls. if you test them remotely, list every variant in Notes for Review
full breakdown with specs, a pre-submission checklist, and how I made the Homiva header in image 4: bytswise.com/blog/apple-app-…
ran Gemini 4 and Claude Opus 5.5 through the same 4 App Store jobs:
ASO metadata: Gemini used 58 of 60 chars, Claude 51
paywall redesign: Gemini added a ★4.9 rating that doesn't exist. Claude didn't
SwiftUI refactor: $0.48 vs $1.13, both built first try
one-shot game: Claude finished 17s faster
Gemini was cheaper on 3 of 4. Claude was faster on 3 of 4. only one made up reviews
one run each, same prompt, no follow-ups. Gemini costs are API-price estimates. you can play both games here: bytswise.com/blog/gemini-4-v…
asked Gemini 4 and Claude to write my app's App Store metadata. both followed every rule.
Gemini used 58 of 60 name+subtitle characters. Claude left 9 on the table.
in ASO, empty characters are rankings you never get
the chart says Argon is pricier per task.
tried a real refactor from my app: Argon ≈$0.48, Opus $1.13. both built first try.
cheap tokens ≠ cheap work… except when they are
named after a gas that doesn't react with anything. the timeline did not get the memo 😂
tl;dr for devs: $2/$10 per 1M tokens until Nov 30, then same price as Opus
asked Gemini 4 and Claude Opus 5.5 to redesign my app's paywall. same photos, same prompt, one shot.
left is what's live today. which one would you ship?
(Gemini also invented a 4.9 rating for me 😂)
everyone's sharing the cornhole robot. this is the real one:
Argon split 6M lines of legacy code out of Chromium in weeks, not months.
that's the job most of us actually have
Our engineers used Gemini 4 Argon to eliminate an entire category of browser crashes in @GoogleChrome.
Argon helped separate core webpage handling from a 6-million line legacy folder in Chromium's 40-million line codebase, completing months of work in just weeks. Here's how ↓
Gemini 4 vs Claude Opus 5.5.
One prompt, one shot: build me a mobile game.
Which one would you play?
If you're adding AI to your app, here's what I'm stealing: let the model pick from your components instead of drawing UI, check its output before users see it, and keep interactions local.
Wrote it all up with sources and 5 prompts to try: bytswise.com/blog/chatgpt-in…
A couple of details I couldn't find anywhere else, like the ~70 components count. And one's off: some components do keep their state on reload, OpenAI's own docs say so.
Read @rabi_guha's teardown of ChatGPT's new Intelligent UI and couldn't help myself. Went digging through OpenAI's public runner code to see what actually holds up.
