I've been experimenting with AI-assisted development lately — mostly "vibe coding," where you describe what you want and iterate through conversation rather than writing every line yourself. Wanted to share what that process actually looked like on a real project.
The Project
A free numerology calculator called Destiny Matrix Chart — it takes a birth date and generates a visual chart (SVG-based octagram) with calculated positions, all client-side, no backend.
What "Vibe Coding" Actually Looked Like
- Described the core calculation logic (digit-reduction algorithms) in plain language, iterated on edge cases together
- Built the SVG chart generation through back-and-forth — describing the visual layout, adjusting positions, colors, and spacing conversationally
- Debugged a genuinely annoying WordPress content-filter issue where
<script>tags kept getting mangled on paste (wpautop stripping things it shouldn't). Ended up base64-encoding the payload and decoding it at runtime to sidestep it entirely — not something I would've thought to try without talking through the problem out loud first.
What Surprised Me
The bottleneck wasn't writing code, it was describing intent clearly enough. Vague requests got vague results. Being specific about edge cases, data structures, and constraints upfront saved a lot of back-and-forth.
The Result
Everything runs client-side — vanilla JS, SVG generation, no dependencies. You can see it live here:https://destinymatrixchart.co/
Curious if others have run into similar friction points with AI-assisted builds, especially around CMS quirks like the one above. What's tripped you up?
Top comments (0)