Hey everyone,
To pick up where I left off in my previous post, everything was working smoothly in my Gym Scheduler app.
The login worked. The schedule worked. The deployment worked.
Life was good π
Until... I decided to add a new feature.
The Feature That Started It All πͺ
I wanted users to be able to create and manage multiple workout programs depending on their goals.
For example:
- Push Pull Legs (PPL)
- Upper / Lower
- Full Body
- Strength-Focused Program
| Before | After |
|---|---|
| One simple schedule | Multiple programs users could easily switch between |
Simple enough, right?
Yeah... not exactly π
Claude Had a Plan π
Of course, I turned to Claude.
I started by creating a detailed implementation plan and adding it to my planning.md file.
After reviewing everything, Claude confidently told me:
"Looks good, let's implement it."
Perfect. Everything seemed to be going great.
Claude started:
- Creating the necessary files
- Installing required packages
- Updating the database logic
- Writing the code
I was feeling pretty confident.
And then I made my biggest mistake.
I Let Claude Run Unsupervised π
The code looked fine. The plan looked fine. Everything seemed fine.
But when I ran the project locally...
everything fell apart.
Everything Broke π
Instead of getting a shiny new program-switching feature, I got:
- β My old schedule disappeared
- β The styling changed
- β The dropdown was missing
- β The feature didn't even exist I just sat there staring at the screen thinking:
"What on earth happened?"
Attempt #2 π
Naturally, I asked Claude to review the code.
A few moments later:
"Everything looks good π"
And then...
it broke the app again π
At this point, I wasn't sure whether to laugh or cry.
Time To Take Control π¨
I finally decided to stop trusting Claude blindly.
First, I reverted all the changes.
Then I started approaching the problem differently:
- Cleared the conversation context
- Explained the expected behavior more clearly
- Shared screenshots
- Updated the implementation plan
- Narrowed down what was actually broken
And suddenly, things started making sense.
This time Claude rewrote the affected files differently. The implementation looked cleaner. The logic looked better.
And for the first time...
it actually worked π
What I Learned π€
This experience taught me something important.
AI is incredibly powerful.
But when it gets stuck, throwing more prompts at it usually isn't the solution.
What helped was stepping back, understanding the problem myself, and giving better context.
The better I explained the issue, the better Claude's solutions became.
Turns out AI isn't a mind reader after all π
Feature Showcase π
After all that debugging, the app ended up much better than before.
Some of the new additions include:
- βοΈ BMI Calculator
- π₯ Calorie Calculator
- π Current-Day Highlighting
- π Exercise Reordering
- ποΈ Multiple Workout Programs
- β¨ Better UI & UX
Final Thoughts
At the end of the day, the feature did work β despite all the stress it caused π
And honestly, the app became much better because of it.
The biggest lesson I took away from this:
AI can write code faster than most developers ever could.
But developers still need to communicate clearly, think critically, and understand what they're building.
The code might be generated by AI.
The direction still comes from you.
Enjoyed this one? Drop a β€οΈ and follow along β more parts coming soon! π




Top comments (0)