shipped, 2025
WorkPortfolio.io
A full SaaS, shipped by directing AI coding agents.
This was my first real software product! It tanked. But you learn by doing, not by watching. I’m still proud of it :)
Showing is better than telling. It’s better to ship a product than to take a course. WorkPortfolio is a place to do exactly that.
Build a professional portfolio of your projects, and keep it current. Each update takes less time than a LinkedIn post.
At a glance
Shipping production software usually demands months or years of prerequisite learning before the first real system goes live. That’s a slow path when the goal is to deliver, not to collect credentials.
I built a real multi-tenant SaaS — auth, billing, storage, a custom analytics pipeline, SEO, and a hardened deploy — end to end by directing AI coding agents, instead of the conventional learn-then-build path.
The binding skill was verification: learning enough of each concept — webhooks, RLS, serverless lifecycles, billing state — to catch confident-but-wrong agent output before it shipped.
It’s a production system on Vercel, Supabase, and Clerk. It missed product-market fit — ~70 users, thin retention, and I’ll say that plainly — but it proved I can ramp on an unfamiliar stack fast and ship, and it converted directly into a first paid software-build engagement.
AI components
None in the product — there is no model provider in the dependency tree. This system was built with AI, not powered by it.
AI coding agents drove the build — ChatGPT for planning, then a deliberate move to Cursor’s agentic workflow for repo-aware development. The AI lives in the development process, not the runtime.
The lived discipline of verifying agent output — catching hallucinated APIs, missing routes, and unsafe defaults — was the project’s core skill, and where AI-assisted development quietly fails without it.
Technology
Next.js • Supabase • Clerk • Vercel • Cursor