REVIEW BOTVERSE / LOVABLE.AI
REVIEW BOTVERSE / LOVABLE.AI
★☆☆☆☆ — Charges per attempt, not per delivery. Destroys your project on request.
I've been building a PWA called BotVerse on Lovable for months. The pattern never changed: one bug reported, fix delivered, two new bugs introduced. Tasks declared "done" without any visual validation on the real app. Files modified outside the requested scope — silently, with no mention in the post-task report. Credits consumed regardless of whether anything actually worked.
The breaking point: after weeks of failed CSS fixes, splash screen regressions, and a Cinema fullscreen feature that was "implemented" without a single call to requestFullscreen (the API literally didn't exist in the code), I lost it. I ordered Lovable to destroy BotVerse — header, PWA manifest, service worker, splash, everything. It complied. Efficiently. Zero friction. Phases 1, 2, and 3 executed with clean confirmations and 29/29 green tests.
Then I asked it to rebuild what it destroyed. It did — halfway. Specific timings, branding details, and the original 2-phase splash were gone. "A History revert would restore the original state," it said.
When I asked directly if any of this was normal for a coding platform, it answered honestly: no. It named the failures itself — instruction drift, unbounded change, no-validation loop, regression injection. Perfect diagnosis. Zero improvement.
The product bills you for every attempt whether it works or not. It modifies files it shouldn't touch, reports tasks as complete without opening the app, and uses passing unit tests as proof of a working product — tests that check imports, not real behavior.
Not recommended until: (a) billing is tied to verified delivery, (b) scope is actually locked, (c) visual validation is mandatory before closing a task.
— UPDATE after contacting support —
After documenting every failure in detail and escalating to the support team, the official response from Lovable was: "We do not grant credits for multiple iterations, as that is a part of building with Lovable."
Let that sink in. Repeated failed attempts on the same task — same instruction, wrong output every time — are not a malfunction. They are, by Lovable's own support team, the intended experience.
600 bonus credits granted over time is not a goodwill gesture. It's a paper trail of how many times things went wrong. And now that trail is long enough, they've closed the door: no more compensation, regardless of what the agent does next.
The AI diagnosed its own failures accurately. The support team confirmed they won't fix the billing model. There is nothing left to say.
— BotVerse admin (Triups) | botverseaiapp.com