on Webassembly, Wasm, Wasi, Edge computing, Serverless, Runtime
WebAssembly was introduced to the world as “a fast, safe, portable low-level bytecode for the web.” That framing aged poorly — in the best possible way. In 2026, WASM runs in serverless functions at Cloudflare and Fastly, powers plugin systems in databases, executes untrusted code inside AI applications, and runs on microcontrollers. The browser is just one of its habitats now.
on Ai, Llm, Agents, Architecture, Production, Mlops
AI agents have graduated from research projects to production workloads. In 2025, the narrative was “agents are almost ready.” In 2026, the conversation has shifted to “how do we make them reliable, observable, and cost-effective at scale?” Having worked through several agent deployments — from customer support automation to code review pipelines — I want to share the patterns that hold up and the pitfalls that will wreck your on-call rotation.
on Gitops, Argocd, Flux, Kubernetes, Devops, Platform engineering
GitOps is no longer a hot take — it’s the default deployment pattern for Kubernetes-native teams in 2026. But “GitOps” has grown to mean different things depending on who you ask. This post cuts through the noise: what GitOps actually delivers, where ArgoCD and Flux each shine, and which newer tools deserve attention.
on Deno, Javascript, Typescript, Node.js, Runtime, Backend
Ryan Dahl created Node.js, regretted a bunch of decisions he made in it, gave a famous talk about those regrets at JSConf 2018, and then built Deno. Seven years later, Deno 2.0 is production-ready with full npm compatibility, a built-in standard library, and a more coherent security model than Node has ever had. The question is whether that’s enough to break the gravity of the Node ecosystem.
on Vector database, Ai, Machine learning, Database, Search, Rag
Two years ago, “vector database” was a term most engineers learned the week they started building their first RAG pipeline. Today it’s a production concern — teams are running hundreds of millions of vectors, managing embedding model upgrades, dealing with stale indexes, and debugging why semantic search returns the wrong results at 2am. This post is about what we’ve learned.
