How we tested v0 and bolt.new for MVP development and what we learned.

We took the success of low-code AI dev tools very seriously. At the end of 2024, I and TITL team decided to see if we could hire low-code developers, integrate AI tools, and launch a new service for prototypes and MVPs development. In other words, could we make it truly cheap?

‍

Since we’re a design-first team, we saw this as a natural extension of our high-fidelity Figma prototypes: those detailed, clickable mockups used to validate user flows, get feedback, drive sales and hand off to developers effectively.

‍

So we tried to build a mobile and a web app with v0 and bolt. Here’s how it went:

‍

Designs import

Our typical workflow starts with designs, so the mockups import and their further adjustment became the first major challenge. Imagine a mediocre FE dev coding your design from memory: misplaced buttons, inconsistent spacing, ignored styles. When you asking the AI to correct the cart button moved out, it can move the button to the right place, but something else will go. And so on around, one corrected, the second, the third ... at this point it's not even about specific styles (which it’s not eager to apply correctly as well), but in general there're too many nuances for AI.

‍

Because we prioritize a UX and cohesive brand identity, these issues were a headache.

‍

Debugging

The rest goes smoothly until bugs appear, then things get complicated again. You get a lot of code that is hard to navigate because it's the 1st time you see it. You won’t immediately know which part of the generated code is causing the error. There's no easy way to refactor the code, and we got stuck telling the AI "please fix this," without knowing if the fix is correct or if it breaks something else.

‍

In contrast, in a powerful IDE (like Cursor), you know your code and how it’s structured, so it’s straightforward to set breakpoints, view logs and so on. You can quickly find and fix a problem.

‍

Key takeaway

I believe the current strategy shouldn’t be “let AI write everything,” but rather using AI to generate pieces of code and then assembling them yourself. It’s clearly a great booster for the process...in hands of a strong developer 🙂.

‍

Some might ask, Why start with design first? Well, sure, you could throw together a simple, template-based app, but the clients could do that themselves, so there’s no business in that alone. And we wanted to see how this approach holds up against real-world tasks we face every day: ensuring frictionless UX, upholding cohesive branding, and delivering clear messaging (visually and conceptually).

‍

As for the new service, of course we won't rely on low-code developers to cut costs. Instead, we decided to proceed with: prototype/MVP development led by a senior developer, boosted by AI tools. And we’ve already started building a mobile app under this new model.