Articles
Worth reading
A short take on pieces worth your time — architecture, AI, system design, and engineering careers — each one linked back to its original source.
42 articles
In-House LLM Serving at Netflix
Most companies just call a hosted LLM API and move on; Netflix's AI Platform team explains why and how they run the full inference stack themselves instead, inside their existing production environment rather than spinning up a separate ML silo. What's genuinely useful here is the decision-making, not just the architecture — the tradeoffs they weighed on cost, latency, and operational ownership before choosing to build rather than buy. Good grounding if you're evaluating whether your org actually needs to run its own inference infrastructure.
How Netflix Simplified Batch Compute with Kueue
A concrete look at how Netflix simplified its batch compute stack by adopting Kueue, the Kubernetes-native job queueing system, instead of maintaining a bespoke scheduler layer on top of it. It's a good case study in a pattern I see a lot in mature platform teams: replacing homegrown infrastructure with an increasingly capable open-source primitive once the ecosystem catches up to what you built out of necessity years earlier. Useful if you're running batch or ML workloads on Kubernetes and wondering whether you still need your own queueing layer.
Securing the future of AI agents
Rohin Shah and Four Flynn lay out Google DeepMind's roadmap for securing their own internal systems against AI agents that are capable enough to be genuinely useful but not reliably aligned: access control, anomaly detection on agent trajectories, and human escalation for suspicious behavior, treated as security engineering problems rather than alignment philosophy. What struck me is that this isn't speculative; they describe an internal prototype already monitoring coding-agent trajectories in production. A useful, concrete counterpoint to AI safety discussions that stay abstract, this is what "securing an agent deployment" looks like as actual infrastructure.
Scaling beyond one: How Airbnb evolved its data architecture for a multi-product world
Airbnb's data engineers walking through what happens to a decade-old data model when the product stops being "one thing" — Homes — and becomes three (Homes, Experiences, Services) at once: the consistent, flexible modeling framework they built so the same metrics and pipelines don't fork three ways. It's a useful case study in designing for extensibility in a data platform, not just a services layer, which is the part of "architecture" that gets talked about far less than it should. Relevant well beyond travel and marketplace companies — anyone with a growing product surface will recognize the problem.
The Death of Microservices Hype: When Modular Monoliths Win
A sober look at the 2026 correction away from "microservices by default": teams that adopted them before they had the scale to justify the operational cost are quietly moving back toward modular monoliths — a single deployable unit with real internal module boundaries, rather than either a tangled legacy monolith or a fleet of distributed services. The honest takeaway isn't "microservices were wrong," it's that the decision needs to follow actual scale and team-topology pressure, not architecture-conference hype.