Tag: chatgpt

  • Prototyping > Vibe Coding

    Prototyping > Vibe Coding

    I’m not a fan of the term vibe coding. It sounds sloppy, unserious, and a little too close to “just prompt until something happens.” But the underlying idea, compressing the distance between an insight and something real you can interact with, is one of the most important shifts happening in product right now.

    For years, the path from idea to reality was slow and structured.

    Insight → wireframes → mockups → revisions → stakeholder input → engineering handoff → first working version → realization that we missed a lot.

    That cycle could take weeks. Sometimes months. And it often meant we were making major decisions based on static artifacts instead of something people could actually use.

    Now we can collapse most of that into hours.

    Tools like Lovable, Replit, and v0 let you spin up working prototypes in minutes. You can interact with them, tweak them, rethink flows, and explore directions before anyone commits to building production code. While these tools can generate full applications, I think their real sweet spot today is helping product teams and stakeholders spec projects faster and more clearly than ever before.

    Instead of describing the product, you can just show it.

    And more importantly, you can use it.

    From Spec to Something Real

    On our current platform project, we built a comprehensive prototype of the future product in about 20 hours and roughly $200. A few years ago that would have taken months and $30–40k in external design and engineering support just to reach the same level of clarity.

    But the part that really changed things wasn’t just the speed.

    We synced that prototype to GitHub. Then we had agents port the prototype pages into our actual codebase. At that point, the conversation shifted from:

    “Here’s a Figma file. Go build this.”

    to:

    “Here’s the UI. Your job now is to make this page work.”

    That’s a completely different starting point.

    The prototype becomes the spec.

    It already contains layout, interactions, component structure, and intent. Agents building against it already understand how things are supposed to behave. This is especially effective if your production stack shares the same front-end patterns. In our case, we chose building the with React, Tailwind, and shadcn because that is usually the default ecosystem most of these prototyping tools generate against. That alignment makes the handoff from prototype to production much smoother.

    The Tradeoff: Sameness

    Of course, there are tradeoffs.

    You can already see a certain sameness creeping into modern apps. Many AI-built interfaces use the same component libraries with slight visual variation. Dashboards start to look familiar. Patterns repeat.

    That’s the cost of speed.

    But it’s also the reason agents can move so fast. Standardized stacks mean less friction, less guesswork, and more momentum.

    And honestly, in early stages, speed matters more than visual originality. You’re trying to find something that works. Something useful, clear, and valuable. You can differentiate later.

    Why These Tools Matter Right Now

    Right now, tools like Lovable and Replit are incredible for getting an idea off the ground and iterating until something feels right. Not perfect. Not finished. Just right enough to build.

    One interesting thing I’ve noticed in my own workflow is how prompting style affects outcomes.

    Early on, I was extremely prescriptive. I’d use ChatGPT or Claude to generate long, detailed markdown instructions describing exactly what to build, then feed that into the prototyping tool. That works. Sometimes it’s necessary, especially when you have a clear vision.

    But I’ve also found value in being intentionally vague.

    Prompts like “make the dashboard useful” or “this page feels empty” can lead to unexpected ideas. Layouts, features, or small touches we hadn’t thought of. Occasionally, those end up being the best parts.

    There’s a balance there. Direction matters. But leaving room for interpretation can surface new thinking.

    The Real Shift

    To me, that’s what modern prototyping really is.

    It’s not about replacing design.
    It’s not about skipping engineering.
    And it’s definitely not about “vibes.”

    It’s about moving the moment of truth earlier.

    Instead of debating ideas, you interact with them.
    Instead of writing long specs, you explore working versions.
    Instead of waiting weeks to learn what you missed, you find out the same day.

    And once you have something real, agents can take it the rest of the way.