Member-only story
Musings on Agentic AI Development
or “Vibecoding a Full-Stack Trading Platform: Lessons Learned on Prompt Engineering as a Core Developer Skill”
Note: This blog post was originally published on my personal blog at bgateb.com. It’s being posted here for posterity.
I’ve been using ChatGPT since GPT-3.0 — and while I know there are people out there doing truly wild things with AI and coding agents, building a full-stack web application from scratch wasn’t really something I’d seriously considered for myself.
I’m what I’d call a “legacy” developer — my career has been working with ERP systems — the old, entrenched, behemoths of systems that demand stability and security — not exactly the “new school” React-and-Node crowd that the modern web is built on. But over the course of about five days in my spare time, I built a fully-featured virtual pin trading platform for a Disney Pin Forum community, complete with custom OAuth2 authentication tied to the forum, APIs, a well-normalized mySQL database schema, implemented a complicated trade-state machine, real-time offer negotiation with counter-offers, shipment tracking integration, multi-theme UI, and a cron-driven backend — using Claude as my implementation partner.
Over 60,000 lines of production code. And it works really well.
So — what changed?
The discourse around AI-assisted development has split into two camps:
- Those convinced that LLMs will render software engineers obsolete,
