Most people want to become: «AI engineers.» Very few are thinking about becoming: «AI infrastructure engineers.» That might be one of the biggest opportunities in tech right now. Because as models…

Most AI infrastructure today still assumes: «one request one response one execution path» FARA is interesting because it pushes toward something much more agent-native: «adaptive runtime…

Most AI agents fail for a boring reason: «bad infrastructure.» Not bad prompts. Not weak models. Infrastructure. stablyai/orca is interesting because it points at a shift a lot of builders eventually…

Ray 4 万+ stars，解决的是 AI 和数据任务的分布式计算问题。 当任务从单机脚本变成训练、推理、调度、并行执行，Ray 这种运行时就开始有价值。 适合：AI 计算、分布式任务、模型服务。 不适合：普通小脚本过早上分布式。 入口：

Somewhere during all this I realized I enjoy high ownership environments where I can touch everything from backend to product decisions. Still figuring things out, but excited for what’s next.