Occasional posts on things I've built, broken, or learned.
Alibaba's Qwen3 brings thinking-mode reasoning, a 235B mixture-of-experts architecture, and state-of-the-art benchmarks — at a size businesses can actually self-host.
Qwen gives businesses a credible path to running powerful AI on their own infrastructure — no data leaving your servers, no usage bills, no vendor dependency.
Building a demo agent is easy. Keeping one running reliably in a real business is not. Here's what actually breaks and how to design around it.