I like tinkering with new product ideas. I struggle sometimes with the quality of my prompts – taking my idea and prompting tools like Bolt or Lovable to create a prototype. It’s that classic cold start problem where you don’t know what you don’t know. Just for fun, I wanted to see if I could build a vibe coding assistant that helps people generate more detailed prompts, giving their AI tools the right input.

In the absence of a simple tool to get over my first cold start problem, I used Claude to generate a quick PRD outlining the key requirements and milestones for my vibe coding assistant idea (see here if you want to learn more about using AI to generate PRDs).

My first prompt is simple enough: I want to implement the PRD and focus on front-end capability only.

I’m not unhappy with the first output. The chat interface feels intuitive and asks the right questions for users to get started with their idea.

I tweak things further by adding extra questions that I want the vibe coding assistant to ask users (like adding real-life examples or designs to enrich the prompt generated by the assistant).

Because of my instructions to only focus on the front-end, none of the sessions in the vibe coding assistant are persisted. I want users to be able to go back to previous sessions and retrieve prompts they’ve generated.

This is the step where I learn the most – about things like database structure, statement management updates and auto-save of key events.

Or when things go wrong and the data doesn’t persist between user sessions on the vibe coding assistant.

After a bit of back and forth, the AI comes up with the required fix and the session data persists correctly.

Main learning point: I deployed my vibe coding assistant here. There’s a nice irony in building a “vibe coding” assistant”; even if my goal is to help people build on the fly and remove onerous planning upfront, I still needed some time to figure out how my assistant would complement (and not duplicate) the prompting capabilities already built into tools like Bolt and Lovable. The foundation of critical product think still matters, even just to provide the AI with the right level of context.