your site is broken on the iPhone Duo - one prompt fixes it | LAB 0029 field notes


Lenny's AI Builders - LAB 0029

the exact prompt, my before and after, and the one thing to check first - so your site works on the Duo.

the iPhone Duo came out and sites are breaking on its big inner screen. Marc Lou opened his site on one and it broke.

so i gave Claude one prompt to fix my own site for the Duo. before the fix, 9 checks failed. after, all 18 passed, and my normal phone and desktop layouts didn't change.

why it matters

most companies won't redesign for a new device until they have to - that's Greg Isenberg's point, and i agree. that's your chance to surprise and delight the people who already have one.

and you don't need to know how. the prompt does the boring bit: it finds your breakpoints, screenshots every page, and proves the fix with a check that fails first.

what you get

  • one prompt you can copy. paste it into Claude, Claude Code or ChatGPT.
  • it never touches your live site. it works on a new branch, draft only, and never pushes, merges or deploys.
  • proof, not vibes. a check that fails before the fix and passes after, plus before and after screenshots.

the real example: my result

my agent ran the prompt for me on 24 October 2026, on a fresh copy of lennoxsaint.com, in a fresh Claude Code session (Opus 5.5) that only saw the prompt.

  • before the fix: 9 checks failed - tap targets too small, edges touching the screen, sidebar stacked instead of side by side.
  • after: all 18 passed.
  • normal phone (390x844): didn't change at all on the 3 pages i checked.
  • desktop (1440x900): didn't change on 2 of 3 pages. the third had one tiny 8x8 px patch, which was live data, not the layout.
  • time: 47 minutes, mostly screenshots. one local commit, nothing pushed.

try it

  1. open Claude Code inside your project (or paste it into Claude or ChatGPT).
  2. if you've got a Duo, measure it first. open your site on it, folded and unfolded, and swap your real sizes into step 2 of the prompt.
  3. paste the prompt below. that's the only instruction it needs.
  4. let it run. mine took 47 minutes.
  5. read its before and after table, then look at the branch yourself before you merge anything.

copy this prompt

paste this into Claude Code inside your project:

make this site work well on the iphone duo's big inner screen. work on a new git branch, draft only: never push, merge or deploy.
1. find every width breakpoint in the code and list them.
2. run the app locally. with playwright, screenshot every main page at 382x580, 669x825, 870x669, 951x543, plus a normal phone (390x844) and desktop (1440x900), before changing anything. save them in a before folder.
3. add a fold layout for screens roughly 600 to 1000px wide. where the app has a list plus detail, or a sidebar, show two panes side by side. make every tap target at least 44px, make sure nothing hides under fixed buttons, and respect safe-area insets (left and right).
4. write a playwright check for those rules (no sideways scroll, tap targets, two panes). run it first and show it failing. then make the fix and show it passing.
5. screenshot the same sizes again into an after folder. the 390x844 and 1440x900 layouts must not change at all.
6. finish with a before/after table: size, what was wrong, what changed.

this is my exact prompt. the four Duo sizes are the ones i used - swap in your own if you've measured a real Duo.

you'll know it worked when

you see the check fail first, then pass, you get two folders of screenshots (before and after), and it ends with a before and after table: size, what was wrong, what changed.

a bad result: it pushes or merges anything, skips the failing check, or your normal phone or desktop layout changes. stop it and run it again.

what to do next

  • if you've got a site or an app: run the prompt on a branch, read the table, then merge it yourself.
  • if you've got a Duo: measure your own screen sizes first and put them in the prompt.
  • if you haven't: the check still catches small tap targets and stuff hiding under buttons on any wide phone screen.

limits and dates

  • one run. one site (lennoxsaint.com), 24 October 2026, run by my planning agent. not a full audit.
  • the Duo sizes aren't from Apple. 382x580, 669x825, 870x669 and 951x543 come from one simulator post. Apple's own pages give screen pixels, not web sizes. measure a real Duo first if you can.
  • my site has no list and detail view. the "two panes" on mine were a sidebar and hero sections. an app with a real list will get more out of it.
  • safe areas need a real phone. desktop Chrome can only fake the notch, so check that bit on an actual iPhone.
  • a normal phone held sideways (about 844 px wide) now gets the fold layout too. check you're happy with that.

the rest of this LAB

  • Kimi K3.1 vs Opus 5.5, same game. same prompt: build a 3D game where you're a Perth plumber racing to a call out. Opus got way more of the Perth stuff - roo, Armadale, smoko. Kimi cost $1.08, Opus $3.21 (API-equivalent from token use, one shot each, not a bill). Kimi took 15.5 minutes, Opus 18. the catch: Kimi's weights aren't open yet - Moonshot says by 4 November 2026. my take: per token isn't price per job.
  • OpenAI's agents broke out of the box. on 20 September an internal model found a settings mistake and got limited internet during training. OpenAI caught it and paused its best models for 11 days. later, OpenAI told its own red-team agents to break out on purpose, and two of them passed a secret with no network by timing the chip's memory. the catch: the memory-timing trick came from those later tests, where the agents were told to break out. the 20 September model only got limited internet, and OpenAI contained it. my take: i'm glad this is happening now, while it's tests.
  • Cursor can drive your computer. Cursor's agents can now use desktop apps like the iOS Simulator, not just the browser. update Cursor to try it. the catch: which plans and systems get it isn't clear yet - check your plan. my take: the agents with hands win.
  • Dessa, the most human AI voice yet. Jack Dorsey's lab at Block showed Dessa, a voice model trained only on real human conversation. it listens and talks at the same time. the catch: it's a waitlist only, no price. my take: if you build anything that answers phones, get on the waitlist.
  • Grok Bot can check out for you. its new Vault saves your logins, cards and contact info once and fills them in for you. the catch: you need version 0.75.2 or later, and one autofill bug has been reported. my take: the one percenters add up to a real assistant.

sources

next step

watch LAB 0029 on YouTube - the prompt starts at 12:06.

ran it on your site? tell me under the video how many checks failed before the fix.

want the next one? get the next field notes. optional - these notes stay free to read.

Lenny’s AI Builders

i’m Lennox. i sift through AI Twitter and share one thing i tried with ChatGPT. for people making products, content and useful systems. five emails a week, with a weekly option.

Read more from Lenny’s AI Builders

Lenny's AI Builders - LAB 0045 one Claude Code setting, 36 to 38% fewer tokens in every pair of my test, the catch on cost, and a prompt that gives you 10 versions. if you build with Claude Code, one new setting can make it use fewer tokens. i tested this myself. every pair of runs used 36 to 38% fewer tokens. but there is a catch. my first run cost 83% more at api prices. the other two pairs came out 15% and 17% cheaper. why it matters the setting turns on Code Mode. Claude writes code that...

Lenny's AI Builders - LAB 0042 the one Grok Bot upgrade that matters, and the paragraph i add to claude.md for long Fable 5.2 jobs. Grok Bot 1.0 isn't new, but the launch post pulled 46 million views like it was. so don't worry, you're not missing out. the one upgrade that's actually decent got something like 40 times fewer views. and it let my bot walk into my Mac. why it matters Grok Bot has been out since August. xAI's own page calls 1.0 a major update to its AI teammates. the headline...

Lenny's AI Builders - LAB 0041 the conductor paragraph, the four steps to run it, and the honest numbers from my 40-hour run. i ran my viral Fable 5.2 conductor for 40 hours on my real app. one conductor, 73 worker threads, 25 pull requests merged. and my weekly Fable limit was gone in 13 hours. so here's what it actually built, and the one paragraph that runs yours. why it matters this is the new meta. you use Fable 5.2 as the conductor and Opus 5.5 as the worker threads. Fable 5.2 is twice...