How GainFrame was built: a new dad, Claude Code and a one-person iOS app run like a company
Michael Rode has a full-time job and a baby at home. His second iOS app turns gym progress photos into body-fat estimates, and Claude Code writes most of the code. Nine months in it makes about $2,400 a month, RevenueCat-verified.
- MRR from 465 active subscriptions on Nov 9, 2026, RevenueCat-verified on TrustMRR
- $2,374
- all-time revenue on Nov 9, 2026, RevenueCat-verified on TrustMRR
- $20,161
- from first commit to the v1.0 TestFlight build, founder-reported
- 12 days
- spent on paid ads before he switched them all off, founder-reported
- $5,674
GainFrame is an iOS app that turns gym progress photos into body-fat estimates, muscle-group scores, side-by-side comparisons and an AI coach that answers from your own check-ins.[1] It is built and run by one person, Michael Rode, who describes himself as a full-time backend engineer of 15 years, a lifter for 20 and a new dad.[2] TrustMRR lists a team of one and says the app is bootstrapped.[1]
He writes about almost every step on GainFrame's own blog, with real numbers. This piece puts those posts together with the app's RevenueCat-verified revenue page on TrustMRR. Unless a line says it comes from TrustMRR, the numbers below are founder-reported.
Before GainFrame: an app nobody found
GainFrame is his second iOS app. The first, Screenshot Swipe, got 432 downloads and $57 in lifetime proceeds, and no longer shows up on the App Store.[3] He lists the reasons himself: no marketing, no user validation and not enough focus on making it simple.[3]
So he set three rules for app number two: anyone should see the value within five minutes, get real user feedback before launch, and start marketing from day one rather than after launch.[3] The idea itself came from his own camera roll: hundreds of progress photos with no weight, workout or notes attached, and no easy way to tell whether anything had changed.[3]
The build: a sprint, then testers from Reddit
The idea sat in his head for about a year. Then the first commit landed on February 11, 2026, and 737 commits later v1.0.0 went to TestFlight on February 23.[5] He found testers without promoting the app: he posted his own progress-photo screenshots in fitness subreddits, mostly the Hevy community, and people asked which app made them.[5] In March he put the funnel at about 150 mailing-list signups, about 100 TestFlight downloads, about 30 people who gave feedback and 5 to 10 power users who shaped the app.[3]
GainFrame went live on the App Store on March 8.[4] Day one brought about $250 in subscription purchases, every one from someone he already knew from TestFlight.[5]
How Claude Code fits in
His July playbook frames the business as departments, where every role a normal app company would hire for is either a paid service or a Claude Code skill he wrote.[2] On the engineering side he puts it plainly:
“Claude Code writes most of the code.”
What keeps that code consistent is a CLAUDE.md in the repo with the architecture rules: enum-driven navigation, the design-system components and what never to touch.[2] The part he says changed his work most is a skill called verify-on-sim. It builds the app, installs it in the simulator and drives it through RocketSim's CLI, tapping by element ID and screenshotting each step, so a feature is checked before he looks at it.[2]
The same idea runs through the rest of the business. A /release command reads the git log, drafts release notes, bumps the version, archives with fastlane, uploads to TestFlight and syncs App Store metadata in 9 localizations. /issues next picks the top GitHub issue and starts on it.[2] Keyword research is a five-stage chain of Claude Code skills, from a product-context file to article generation, with no paid SEO tools.[2] By August his weekly SEO audit was a written procedure that Claude Code runs end to end in about 20 minutes, and every mistake it caught became a new rule in that file.[9]
| AI coding | Claude Code, with a CLAUDE.md and custom skills (verify-on-sim, /release, /issues, SEO chain, weekly audit)[2] |
|---|---|
| App | SwiftUI + SwiftData, iOS 17+, project generated by XcodeGen, CocoaPods + SPM, fastlane[2] |
| On-device AI | Google MLKit (pose, segmentation, face detection) and Apple Vision[2] |
| Backend | Supabase: 30 Deno edge functions, 24 SQL migrations[2] |
| Product AI | Gemini 3.5 Flash for everything in the product, including a coach that can call 17 tools on the user's data[2] |
| Payments | RevenueCat (yearly plan with a 7-day trial, monthly with none)[2] |
| Analytics | PostHog, GA4 and Search Console, all wired into Claude over MCP[2] |
| Website | Next.js static export on Cloudflare Pages, $0 hosting[2] |
One practical detail from the playbook for anyone copying the stack: MLKit ships no arm64 simulator slices, so he gates every MLKit call behind a simulator check and strips its linker flags for simulator builds.[2]
Growth: ads off, owned channels on
Early growth looked steady: $100 MRR the week of March 29, $250 the week of April 19, $500 the week of May 10.[5] Part of it was paid. He spent $5,674 across Apple Search Ads, TikTok and Reddit, and worked out that he was paying about $114 to acquire a customer worth about $18, so he stopped all of it on May 10.[2]
Then came five flat weeks, $605 to $685 MRR between May 17 and June 14.[5]
“I 100% thought I was done.”
Instead of a new idea he fixed what existed. His analytics had logged 10 trial conversions where RevenueCat counted about 48.[5] Once he could measure, the problem was trial-to-paid at 20.7%. He moved two Pro features he had accidentally made free back behind the paywall, tested a paywall that shows the yearly plan as a monthly price, and put previews of the best Pro features into onboarding. Five weeks later the last four weekly cohorts blended 44%, with no ad spend.[6]
The AI bill got the same treatment. In the week of June 14 the Gemini bill was $237.14 against $488.36 of gross revenue, 48.6%. After tagging every model call by feature, trimming the coach's prompt, moving photo-history imports to Gemini's Batch API and caching the coach's starter questions, the week of July 5 came to $160.04 against $955.63, 16.7%.[7] He is clear that about half of that improvement is revenue growing, not cost cuts.[7]
The channels he kept are ones he owns: a blog that had 141 posts and 11 free tools by July, and comic carousels on TikTok and Instagram.[2] In the carousels the AI draws only the art and a Python script sets all the text, because letting the image model render text made every slide look different.[2] In August he reported organic clicks going from 2,153 to 7,205 between two 28-day windows, and free-tool clicks from 302 to 2,799 between two matched 30-day windows.[9][10]
Results so far
His own snapshots: $1,000.04 MRR on July 12, $1,536.09 with 296 active subscriptions on August 2, $1,920.18 with 330 on August 12, and $2,208.95 with 391 on September 10.[5][8][10][11] In that September post he wrote that growth had slowed and he was trying Instagram creator partnerships while he worked out where to spend his time.[11]
On November 9, 2026, TrustMRR showed $2,374 MRR from 465 active subscriptions, $2,512 in the last 30 days and $20,161 all-time, verified with a RevenueCat API key.[1] The app is also listed for sale there. His message on the listing says he is selling because he is a new dad with a full-time job and another project, and the app deserves more hours than he has.[1]
Timeline
Before 2026
First app, Screenshot Swipe: 432 downloads, $57 lifetime proceeds.[3]Feb 11, 2026
First commit.[5]Feb 23, 2026
v1.0.0 to TestFlight after 737 commits; testers found through Reddit.[5]Mar 8, 2026
May 2, 2026
AI Coach ships in v2.0.[5]May 10, 2026
All paid ads stopped after $5,674.[2]May 17 to Jun 14
Flat at $605 to $685 MRR; analytics and trial-conversion fixes.[5]Jul 10, 2026
Publishes the playbook: $944 MRR, 203 active subs, Claude Code writing most of the code.[2]Jul 12, 2026
$1,000.04 MRR.[5]Aug 12, 2026
$1,920.18 MRR; free tools credited for much of the jump.[10]Sep 10, 2026
$2,208.95 MRR; growth slowing, creator partnerships next.[11]Nov 9, 2026
$2,374 MRR from 465 subscriptions, RevenueCat-verified on TrustMRR; listed for sale.[1]
Lessons for builders
- 01
Write your rules down for the agent
A CLAUDE.md with architecture rules and what never to touch is how he keeps every Claude Code session writing code that matches the app.[2] - 02
Make the agent check its own work
verify-on-sim drives the real app in the simulator and screenshots each step, so a green build isn't mistaken for a working feature.[2] - 03
Turn chores into one command
Releases, issue triage and SEO research became skills, which is how one person with about two focused hours a day runs it.[2] - 04
Fix measurement before optimizing
- 05
Tag every AI call by feature
One afternoon of tagging showed where the Gemini money went and made every later cost cut possible.[7] - 06
When AI output drifts, move it into code
His carousels only became consistent once the image model drew the art and Python set the text.[2] - 07
Test paid ads against payer value
About $114 to acquire a customer worth about $18 is why he stopped and put the effort into the blog, free tools and TikTok.[2]
More stories
All stories- CollectionAI-built apps making under $10K a month, with revenue you can check13 real products built with AI coding tools that earn somewhere between a few hundred and ten thousand dollars a month, every figure read from the payment provider.10 min read
- How they built itHow MimikFlow was built: three months in n8n, one month of code, and a product that finds its own customersMo Alani automated his own LinkedIn prospecting, sold the setup by hand, proved it in n8n, then rebuilt it as a SaaS with Claude Code. It now brings in about $11K a month, Stripe-verified.8 min read
- How they built itHow Theanna was built: a non-technical founder, her engineers and Claude CodeNomiki Petrolla doesn't write code, yet she builds every screen her users see. Here's her workflow, her $241-a-month AI stack, and how Theanna reached $232K ARR.8 min read
The newsletter
More stories like this, by email
Occasional email with new projects on the leaderboard and new build stories. Every number sourced, like the site.
Sources
- [1]GainFrame - TrustMRR (revenue verified with a RevenueCat API key)TrustMRR · checked Nov 9, 2026
- [2]Every Tool and System Behind My $944 MRR iOS AppGainFrame blog (Michael Rode) · Jul 10, 2026 · checked Nov 9, 2026
- [3]My First App Flopped. Here's How I Launched My Second One in 2 Months.GainFrame blog (Michael Rode) · Mar 28, 2026 · checked Nov 9, 2026
- [4]GainFrame Is Live on the App StoreGainFrame blog (Michael Rode) · Mar 8, 2026 · checked Nov 9, 2026
- [5]My App Hit $1,000 MRR Today, Five Months After the First CommitGainFrame blog (Michael Rode) · Jul 12, 2026 · checked Nov 9, 2026
- [6]My Trial Conversion Was 21%. Five Weeks Later It's 44%.GainFrame blog (Michael Rode) · Jul 7, 2026 · checked Nov 9, 2026
- [7]My App Was Spending Half Its Revenue on AI. Four Weeks Later It's 17%.GainFrame blog (Michael Rode) · Jul 16, 2026 · checked Nov 9, 2026
- [8]I Thought $1,500 MRR Meant I'd Figured It Out. It Just Gave Me Better Problems.GainFrame blog (Michael Rode) · Aug 2, 2026 · checked Nov 9, 2026
- [9]My SEO Is a Weekly Audit Loop Now. Organic Clicks Tripled in a Month.GainFrame blog (Michael Rode) · Aug 8, 2026 · checked Nov 9, 2026
- [10]Free Tools Helped Add $920 MRR in 31 DaysGainFrame blog (Michael Rode) · Aug 12, 2026 · checked Nov 9, 2026
- [11]GainFrame Is Around $2,000 MRR. I'm Still Figuring Out How to Grow It.GainFrame blog (Michael Rode) · Sep 10, 2026 · checked Nov 9, 2026
Something wrong or out of date? Tell us and we'll fix it.