Passing every RLS test with extra grants still live is the classic AI-built app gap: the agent tests what it built, not what it left behind. A privilege audit belongs on every launch checklist.
Most software doesn't fail in the code. It fails in the first conversation, when nobody asks what the product is actually for.
Hi Contra 👋 I'm Arshad, a software architect in Karachi. With my team at 360Xpert Solutions, I design and build software products, from the first sketch to a live app.
The video is our process in 20 seconds: design it, build it, add the AI, ship it.
The 8 blurred tiles at the end are real products we've shipped. I'll unlock them here one by one: what each one does, the hard parts, and what we'd change.
#1 is next: a legal AI built for Pakistani law, where every answer cites the statute it came from.