The fastest way to understand CodeSky AI is to build something with it. So we did — a small but real SaaS dashboard for tracking gym member churn, end to end, in a single afternoon. Here is exactly what happened, including the bits we had to fix.
1. The prompt
We started with one sentence:
"A SaaS dashboard for boutique gyms to see member churn by week, with an Arabic UI option and Stripe-style billing pages."
CodeSky asked two clarifying questions — preferred LLM and database engine — and then began generating. Within four minutes we had a working React frontend, an Express API, a Postgres schema with seeded sample data, and a deployed preview URL.
2. Customizing without losing the AI
The generated app was 80% there. The next 20% was where the real value showed up:
- We tweaked the chart palette in the editor — saved, hot-reloaded, done.
- We asked the AI to "add a CSV export to the churn table", and the diff was applied with tests in under a minute.
- We swapped the LLM mid-project from a frontier model to a smaller fine-tuned one, just to see the trade-off in code style. Both produced clean, idiomatic React.
3. Going production
One click pushed the project to a custom domain with SSL, and a parallel push committed everything to a fresh GitHub repo we had connected. From the GitHub side, future commits sync back into the CodeSky workspace automatically — useful when an engineer wants to handle a tricky migration directly in code.
4. What surprised us
The Arabic + RTL experience was the part we expected to be roughest, and it was the easiest. Toggling the locale flipped the layout, the menus, the chart legends and the date formats with no manual adjustments.
5. What we still had to do by hand
Two things: the privacy policy copy (we wrote it ourselves) and the "real" Stripe credentials (CodeSky scaffolded the test integration; production keys are an explicit human step, on purpose).
Total time from first prompt to a real customer using the dashboard: 3 hours, 41 minutes. Worth a Friday afternoon.