How to get started building your own app with AI
Posted by
Latest Posts

This week I ran a 2-hour training for founders in the Sabi community who are building their own apps with AI. Here are my notes.
When a company hires Inovo, we start with a 2-week discovery. We go through their code and put some best practices in place. This is the short version of that list, for a founder building with Lovable, Replit, Base44 or Devin.
1. Pick a builder and start small
Lovable, Replit, Base44 and Devin all do the same job. You describe what you want and an AI agent writes the code.
They all use the same AI models from Claude and ChatGPT. The difference is simply user experience. They all have free trials, so paste the same prompt into all 3 and see which works best for your workflow.
A good first project is your own website. Give the builder your live URL and ask it to recreate the site. You already have approved copy and graphics, so about 20 minutes later you have a site you can change by chatting with AI.
If this is your first time, start with Lovable. It's the easiest to learn. For a bigger app or a mobile app I recommend Devin.
2. Save your code in GitHub
GitHub is a private folder where your code lives. It's free. Most AI builders have a GitHub connection. Turn it on once and every change you make is saved there.
- You own the code. If you cancel the builder, the app is still yours. You can move it to another builder.
- You keep every version. If you break something after a week of building, you can go back.
It is so cheap to start building now that the hard part is choosing what to build. Before a new feature, we write a frame.
- Problem - What's not working today, and for who?
- Outcome - What will be different when this is done?
- Not doing - What are we saying no to? This is my favorite section.
- Appetite - How much time is this worth?
And a frame gives the project an end. Without one, it's easy to prompt from morning to night.
For more info on framing, here our guide on how to frame a project before you build it.
4. One feature per session
Treat the agent like a junior hire. Give it one feature per session, then close the session and start a new one. Don't build the whole app in one session.
3 things I ask for all day:
- "Let's plan this before we build it." I want to see the plan first.
- "Give it to me in plain English." I can read code and I still ask for this.
- "Make me 3 options." The agent builds 3 clickable prototypes and I pick a direction.
5. Ask your agent to write tests
Automated tests are checks that your agent writes. They run every time the code changes, and they catch the change that fixes one page and breaks another.
Sabi has 10,000 tests. They run in under a minute, and every one has to pass before a change goes live.
Paste this into your builder once:
Set up a test framework for this project and write unit tests for the core business logic: the code that handles money, user accounts, and anything that talks to other services. Add a single command that runs all the tests. From now on, run the tests every time I make a change.
There is more in Intro to unit tests and evals for your AI project.
6. Let the agent do the setup
As the app grows, you will hear about these tools:
- Vercel - where your app runs. It has a free plan.
- Supabase - where your data lives. It has a free plan too.
- Sentry - watches the live app and reports bugs back to your agent.
The same goes for our articles. You don't need to read them. Hand the link to your agent and say "set this up."
What to do this week
- Create a GitHub account and connect your builder.
- Paste in the test prompt.
- Frame your next feature before you build it.
As always, if you have questions, send me a reply!
AI-First App Founders
Don't miss a post
Practical notes on building AI-first apps from the Inovo team.