I’m not a traditional developer.
I’m learning to build software by actually building with AI tools — and Lovable has become one of the tools I use most.
One thing became obvious surprisingly quickly:
Getting more AI credits is useful. Learning not to waste them is much more useful.
So if you’re just starting with Lovable, here are the lessons I wish I had known before using my first serious batch of credits.
And if you haven’t created an account yet, my current invite gives new users 10 extra Lovable credits:
👉 https://lovable.dev/invite/SDWBXFW
More on that — including what I get from the referral — at the bottom.
1. Don’t start by asking Lovable to “build my app”
The tempting first prompt is something like:
Build me a platform for X with authentication, profiles, payments, a dashboard and a beautiful responsive interface.
It feels efficient.
It usually isn’t.
The larger the instruction, the more assumptions the AI has to make. Every wrong assumption becomes something you’ll have to repair later.
Start by defining:
- what the product must actually do;
- what the main entities are;
- what the first user journey looks like;
- what absolutely does not need to exist yet.
Then build one meaningful slice at a time.
2. Separate thinking from building
One of my most useful habits is not asking the coding tool to solve every product question while it is also writing code.
First decide what you want.
Then ask Lovable to implement it.
For a larger feature, I often work in this order:
problem → specification → implementation → test → refinement
That extra thinking step can save several rounds of rebuilding.
3. Tell it what must remain untouched
This is a surprisingly powerful addition to prompts.
Instead of:
Change the profile card.
Try:
Change only the profile card. Do not alter navigation, authentication, database structure or the existing mobile layout.
AI development tools are very good at changing things.
Sometimes they’re a little too enthusiastic about it.
Defining boundaries reduces accidental regressions.
4. One clear objective beats five repair prompts
When something doesn’t work, it’s tempting to keep throwing little correction prompts at it.
That can create a loop:
fix → new problem → fix → regression → another fix
When that happens, stop.
Describe:
- the current behaviour;
- the desired behaviour;
- the relevant constraints;
- what has already been tried.
Then solve the underlying problem once.
5. Functional first, beautiful second
It is incredibly easy to spend credits tweaking:
- spacing;
- border radii;
- gradients;
- button positions;
- typography;
- tiny responsive details.
Those things matter.
But not before your important flows actually work.
My preferred order is:
architecture → functionality → reliability → usability → visual polish
Otherwise you may lovingly polish a screen that gets rebuilt two days later.
6. Make big changes deliberately
Before a significant change, know how you can get back to a working state.
Version control is not something only “real developers” need.
AI-assisted building actually makes it more important, because large amounts of code can change very quickly.
Before a risky change:
- make sure the current version works;
- commit or checkpoint it;
- define the scope of the next change;
- test immediately afterwards.
Being able to undo a bad idea is a feature.
7. Don’t confuse “the page looks right” with “the product works”
A generated interface can look remarkably finished while important things underneath are still incomplete.
Test the actual user journey.
For example:
create account → sign in → perform core action → save data → return later → find it again
Click the boring buttons.
Try bad input.
Refresh the page.
Use mobile.
Use a different account.
The last 20% of building often contains 80% of the things users will eventually complain about.
8. Treat your database as part of the product
For non-developers, the visual interface is much easier to understand than the data model underneath it.
That makes it tempting to postpone database thinking.
Don’t.
Before adding lots of features, understand at least:
- what data you’re storing;
- which data belongs to which user;
- which relationships exist;
- what users may read;
- what users may change;
- what should never be exposed.
A beautiful application sitting on a confused data model becomes painful surprisingly quickly.
9. Use another AI as a thinking partner
A coding agent doesn’t have to be your only AI.
For complicated changes, I often find it useful to work out the logic separately first and then give the builder a much cleaner instruction.
In other words:
use one AI to think with you and another to build with you.
The better your specification becomes, the less expensive experimentation tends to become.
10. Spend credits on learning, not just output
Some “failed” prompts aren’t actually wasted credits.
If a failed attempt teaches you:
- how your application is structured;
- why a certain approach breaks;
- how authentication works;
- how your database behaves;
- how to write a better specification;
then you’re becoming a better builder.
The real waste is repeatedly making the same mistake without changing your process.
That’s why I keep small notes about what worked and what didn’t.
Over time, those notes become your own development playbook.
Want 10 extra Lovable credits?
If you’re new to Lovable and were planning to experiment with it anyway, my current invite gives new users 10 extra credits:
👉 https://lovable.dev/invite/SDWBXFW
Use them to build something small enough that you can understand what is happening.
Then build the next piece.
Referral disclosure
This is my personal Lovable invite link.
According to the referral offer currently shown in my Lovable account:
- you receive 10 extra credits when signing up through the invite;
- I receive 100 Lovable credits if you subsequently subscribe to a Pro plan with 100 credits/month or more.
It doesn’t make your subscription more expensive.
I’m sharing the link because I use Lovable myself, and the referral credits help fund further building and experimentation.
If that arrangement works for you, feel free to use it:
https://lovable.dev/invite/SDWBXFW
Happy building.
Top comments (0)