What I Got When I Didn't Win
NMA didn't win the Qwen Cloud hackathon. Then I read the winners' code — and the prize-money regret evaporated.
Opus at Home, or Benchmaxxing?
Qwen's 27B dense model posts coding scores that touch frontier models, and Hacker News spent 624 comments arguing whether a 100x smaller model can really be that good - then someone made it think for seventeen minutes about an owl.
The Weights Are the Release
Qwen3.8-Max, the most capable model in the Qwen family, dropped today - and for the first time the flagship itself is going open weights. The Hacker News thread split right down the middle.
Qwen Drops Two More: The 'Always A-Tier' Strategy
Qwen released two new open-source models today, continuing a cadence that has the community calling them 'consistently in the A-tier' — a reminder that in AI, velocity is a strategy of its own.
Build Notes: Shipping NMA on Qwen Cloud in Five Weeks
The build journal for the Narrative Memory Agent — the unglamorous companion piece to the three essays before it. What actually happened between git init and the submission deadline.