What shipped fast
Lovable handled the first-pass screens, onboarding, and dashboard structure shockingly fast. The team had something demoable on day one and a believable client flow by the end of the week.
Operator teardown4 days to a pilot-ready MVPFounder with light frontend experience
Built a client portal MVP in Lovable, then moved the risky backend work into Cursor
A service business needed a client-facing portal with onboarding, document upload, project status, and a paid premium support tier they could demo to pilot customers fast.
What broke
The moment payments, file access, and Supabase policies mattered, the generated backend stopped being something I wanted to trust blindly. Stripe and access state were the obvious pain points.
What they would do differently
I would still start the first pass in Lovable, but I would move sensitive backend boundaries and payment state into a code-first setup earlier instead of pretending the generated backend was done.
Related failure modes
Read these before you repeat the same mistake
Why builders get stuck at auth and databases
The real reasons auth, RLS, schema design, and database assumptions stall AI-built products.
Read the failure mode ->
Why Stripe, subscriptions, and webhooks break so many AI-built apps
The core failure modes around checkout, webhook drift, stale access state, and subscription logic.
Read the failure mode ->
Why builders get stuck at deployment
Why apps that work locally fall apart at domains, env vars, hosting, and production setup.
Read the failure mode ->
Learn the workflow
If you are trying to do this well, read these next
A Landing Page
The hard part is not the layout. It is knowing what promise the page makes, who it is for, and why someone should trust it enough to click.
Read the workflow ->
A Newsletter
The hard part is not the signup form. It is deciding what the newsletter is actually about, what angle it owns, and what makes it worth opening next week too.
Read the workflow ->
A Blog
The hard part is not the page shell. It is creating content that is sharper than the average AI sludge and structuring it so search and humans can both trust it.
Read the workflow ->