I write here as BlueCat. Tsinghua CS PhD. Former Huawei TopMind. Currently independent.
Systems is the keystone layer — it carries applications above and exposes hardware below, and the good ideas inside it cross-pollinate constantly across security, compilers, OS, networking, virtualization, and runtime. Working anywhere in the field for long enough, you start noticing the same primitives recur in different costumes.
My background cuts a grid across that landscape: data-centers through single servers down to edge devices on one axis; GPUs, high-performance networking, OS, virtualization, dynamic binary translation, compilers, and NPU toolchains on the other. Each cell has shipped work behind it — high-performance network optimization, cross-ISA virtualization and JIT, GPU, RDMA container networking, microkernel network paths, on-device OS work, NPU compiler toolchains. Breadth like that used to be a curiosity. In 2026 it's load-bearing — it's the breadth that lets a single human ask cross-field questions a current model can actually push forward.
What I'm increasingly curious about is how short the innovation cycle becomes when those two ingredients are stacked: a person with real grid-spanning experience, paired with a top-tier model. What compresses from the PhD-year scale to the day scale, what doesn't, where the bottleneck moves next. That's the thread running through everything posted here.
Notes for myself, kept here in case any of it is useful to anyone else.