GymBuddy: Guided Strength is now on Google Play. It is free, it works on Android phones, and it speaks English and Russian.
I designed it and built it on my own, with AI coding agents. I am a product manager, and I did not write the code. This post is about what the app does, how I worked with the agents, and what it took to get from a pilot APK to a public release.
Who it is for
Many people who go to the gym regularly do not have a program. They have a habit. They come in, look around, and do whatever they feel like today. Often it means the same exercises every week, a few muscle groups nobody trains, and no way to tell whether anything is moving.
GymBuddy is for them: people who know the main exercises but want a plan they can follow for the next few weeks. Beginners can use it too, but they are not the main group I designed for.
What it does
- Onboarding. You tell the app about your experience, your schedule and your equipment.
- A plan on real dates. The app builds a weekly program and puts each workout on a day in your calendar.
- Logging. During a workout you record every set: weight and reps.
- Adaptation. After each week, the next weeks adjust to what you actually did. Finished workouts stay as they were.
- Exercise cards. Every exercise has a short card in plain words: Setup, Movement, What to watch.
- Your data. You can see your history, export your workouts, and delete your account from the app or from a web page.
The part that picks the exercises and the weights runs on rules I wrote. The same inputs give the same program. I chose that on purpose: when the app suggests a weight, I want to be able to explain why.
My job was to decide
Before serious code, I wrote about fifteen requirement documents: the product brief, the decision log, the screen specs, the data model, and the rules for building and adjusting a program. I described that work in the GymBuddy case study. The first screen concepts came from Magic Patterns, an AI design tool. You can click through the final GymBuddy prototype I made there. I wrote about the tool separately: The design tool I want to thank: Magic Patterns.
AI coding agents wrote the code, mostly Codex and Claude. My own technical skills end with a few git commands that ChatGPT taught me: add, commit and push. ChatGPT and Claude also helped me check rules and edge cases.
The agents never got a free hand. Every change was a “slice”: one written task with the goal, the files it may touch, what it must not change, and how to prove it works. The agent did the work and wrote a report with the evidence it actually got. I read the report, and only then did the change reach the main branch. If a task needed a product decision, the agent had to stop and ask. I decided, and the agent did the work.
The first commit is from 23 March 2026. Today the main branch has more than 1,600 commits and more than 400 slice documents. About 1,900 tests run on every change, and a heavier set runs every week.
Shipping took more than code
Google Play has its own requirements. Most of them turned out to be product work.
- A health app. GymBuddy has an optional cycle section in onboarding for women, so Google treats it as a health app. I kept the section, because removing it would change the programs the app builds. Keeping it came with three conditions: no fertility or pregnancy words in the listing, honest Data safety answers, and the same disclaimer in the store, in the app and in the privacy policy.
- Deleting an account. Play requires that people can ask for deletion from a web page, not only inside the app. So deletion works in both places, and you can export your workouts before you go.
- Emails nobody mentions. Firebase does not send “your account is deleted” or “your first week is completed” . I designed and built those emails myself, in English and Russian.
- The closed test. New developer accounts must run a closed test before they can publish: at least 12 testers, for at least 14 days. I asked people I know in Belarus, Georgia and Canada, and they stayed for the whole test. The test started on 25 September. Google granted production access on 9 October, and the app went public on 14 October.
What changed because of the testers
- An optional section in onboarding crashed the app when it was left blank. Crashlytics showed me where, and it is fixed.
- Some fields were still in English in the Russian version. Fixed.
- A tester who has trained for a year told me the text cards do not help if you do not know the names of the machines. I did not want to turn GymBuddy into a video library. Now the exercise name opens its instructions, and I am working on one simple drawing per exercise.
What I am not claiming
In July I gave the first version to friends. I wrote about it in Testing your app on friends. The short version: people who install your app as a favour can show you defects, but they cannot tell you whether anyone wants it.
The closed test has the same limit, so I am not publishing numbers from it. The testers agreed to help. They were not looking for a training app.
The first numbers I will trust come from people who find GymBuddy on their own: on Google Play, through the GymBuddy site, or through its Instagram, YouTube and TikTok. When I have enough of them to say something honest, I will write about it, with counts, not percentages.
What comes next
For the next few weeks I am working on one question: do people who look for an app like this find it, install it, and come back? I will share what I learn, including what does not work.
In parallel, I have designed a separate app for personal trainers. It will use the same training engine. And I am also working on an AI component for GymBuddy: weekly adaptation, where a language model proposes the changes for next week. It will work inside fixed limits. Every proposal is checked against the exercise list and the progression rules. If a proposal breaks them, the current rules decide instead. I will write about it separately and in more detail soon. Until then, the specification is in my Upwork portfolio.
If your gym program is “whatever I feel like today”, you can try GymBuddy on Google Play. If you build products, the case study shows how the app was designed.
