tcgworld.gg
TCGworld is an online collectibles platform that allowed users to digitally open Pokemon packs and physically redeem the cards they pulled. Basically GACHA lootboxes.
The platform was built in roughly six months by a team of two operating AI-first from the beginning. As coding agents rapidly improved, we evolved alongside them - adopting frontier models, refining workflows, and learning how to orchestrate high-velocity software development with a small number of operators.
What started as experimentation with Claude Code and Sonnet quickly evolved into a broader agentic development stack. Today, much of our development workflow is built around leveraging specialized models and tooling for different parts of the system.
Our stack:
- Next.js
- NestJS
- Claude Code + Opus (frontend workflows)
- Codex + 5.5 (backend workflows - much, much better than Opus in this domain imo)
- Custom internal operational tooling
- Vercel (frontend deployment)
- Fly.io (backend deployment)
- Cloudflare (firewalls, IP proxying, geo-blocking)
- AWS RDS + S3
- Redis
One unexpected outcome of building this way was how quickly we moved away from low-code tooling.
I originally built many of our internal operations tools in Retool because I had prior experience deploying it in enterprise environments. But as coding agents improved, the tradeoffs of low-code and no-code platforms became increasingly difficult to justify. The limitations, abstraction layers, and idiosyncratic behaviors of GUI-based systems often created more friction than speed.
We eventually rebuilt our operational tooling entirely in-house, including:
- order fulfillment management
- ticket resolution systems
- user dashboards
- gem drops management
- promo systems
- internal admin tooling
We also deployed AI agents beyond software development itself. Using market data from the PriceCharting API, we built internal workflows that analyzed sales velocity, pack popularity, pricing movements, and thematic trends to generate new product and SKU ideas. What would traditionally require hours of manual merchandising and market research became a largely automated creative and operational process.
The result was faster iteration, tighter integration with our core systems, and significantly more flexibility over product behavior and operations.
In just two months after launch, the platform reached roughly $3.5k MRR before we ultimately decided to shut it down.
The decision wasn't really driven by growth or technical challenges, but by a growing misalignment between the product and our values. As we spent more time with customer behavior data, we became increasingly uncomfortable with the compulsive spending patterns the platform could encourage.
Despite the traction, it no longer felt right to continue building a product optimized around that dynamic, so we chose to walk away.
You can still check out the platform at tcgworld.gg.