The systems this course studies change faster than the literature that describes them. A scheduling policy, a cache layout, or a disaggregation strategy usually appears in a serving engine's release notes, a vendor's engineering post, or an independent teardown well before anyone writes it up for a conference. This page lists the sources worth tracking in that gap, grouped by subject.
Nothing here is required reading. Use it when you are looking for a final project topic, checking whether an idea has already shipped in production, or trying to find a current number for something a paper measured two years ago.
Must watch is grouped by subject, not ranked. Those sources publish something relevant often enough to justify following the group you care about, and the most prolific and most technical of each group is listed first. Others collects systems blogs that seldom write about AI directly and reward a weekly scan for the storage, scheduling, and performance engineering instead. Every link was checked on 26 August 2026.
Grouped by subject. Within each group the sources that publish most often, and most technically, come first.
These are not AI blogs. They are here because agent infrastructure keeps rediscovering classic distributed-systems problems, and these groups solved the earlier versions at scale.
Scan these weekly. Most posts will have nothing to do with AI, and the ones that matter are about storage, scheduling, and performance engineering done carefully enough to be worth imitating.
These synthesize across companies instead of presenting one vendor's view, which makes them the fastest way to find out whether a claim in a vendor post holds up.
If you want the shortest list that still covers the course, monitor these twenty.
Summarizing every new post from twenty sources produces more text than anyone reads. If you build a monitor, filter on the terms below and summarize only what matches. Building one is a reasonable warm-up for a final project, since the hard part is not the fetching but deciding what deserves your attention.
The course runs one, at news.cs2680.com. It reads every source on this page once an hour and files each post under the topic it scores highest against, using the terms below. Its index is also served as JSON at /posts.json, so you can build on it rather than re-fetching forty-five blogs yourself.
Agents. agent, agentic, coding agent, harness, orchestration, multi-agent, subagent, tool use, MCP, A2A, skills, memory, context engineering, computer use, sandbox, isolation, agent evaluation, long-horizon, reliability, fault tolerance, prompt injection, agent security.
AI systems. inference, serving, KV cache, prefix caching, prefill, decode, disaggregated serving, speculative decoding, scheduling, batching, MoE, attention, linear attention, kernels, quantization, GPU, TPU, NPU, RDMA, networking, storage, checkpointing, model loading, autoscaling, distributed inference, RL infrastructure, energy, sustainability, cost, latency, throughput.
General systems. storage systems, caching, distributed systems, databases, networking, scheduling, serverless, virtualization, containers, hardware/software co-design, observability, failure recovery.