Tất cả bài viết
-
Designing an MCP server for internal systems: don't just port your REST API over Auto-generating an MCP server from an OpenAPI spec is the shortest path to an agent that wastes tokens uselessly. REST serves developers, MCP serves LLMs — these two targets require different designs.
-
Spec-Driven Development with GitHub Spec Kit: specify first, generate code later Features requiring multiple man-days that are handed over to a coding agent via ad-hoc prompts will sooner or later turn into vibe coding. Spec Kit provides a 4-phase process to ensure that the spec — not the prompt — is the single source of truth.
-
Spec-Driven Development với GitHub Spec Kit: đặc tả trước, sinh code sau Feature nhiều ngày công mà giao cho coding agent bằng prompt ad-hoc thì sớm muộn cũng thành vibe coding. Spec Kit đưa ra quy trình 4 pha để spec — chứ không phải prompt — là nguồn sự thật.
-
Voicebot dưới 800ms: vì sao độ trễ, không phải giọng nói, quyết định voice AI thành hay bại Giọng TTS đã đủ giống người từ lâu. Thứ khiến khách cúp máy là 1,5 giây im lặng sau mỗi câu hỏi. Mổ xẻ latency budget của một cuộc gọi voicebot và vì sao pipeline STT→LLM→TTS cũ không thể thắng.
-
Thiết kế MCP server cho hệ thống nội bộ: đừng bê nguyên REST API sang Auto-generate MCP server từ OpenAPI spec là con đường ngắn nhất đến một agent đốt token vô ích. REST phục vụ developer, MCP phục vụ LLM — hai đối tượng này cần thiết kế khác nhau.
-
Tetragon replaces Falco + auditd: consolidating runtime security into a single eBPF agent Running both Falco and auditd on the same node is redundant and costly. Evaluating the direct move to Tetragon 1.4: inline enforcement in the kernel, practical TracingPolicy, and the event data overhead that few people talk about.
-
Production-ready Webhooks 2026: Standard Webhooks, JWKS, and the replay problem The difference between a webhook that "works" and a webhook that "runs reliably" is a checklist: Standard Webhooks, rotating signing keys via JWKS, retries with DLQ, and idempotency keys on the receiver side.
-
Self-hosting LLMs with vLLM: cost analysis and a checklist before leaving APIs At stable volumes, self-hosting is 8–18x cheaper per token than APIs — but only if you calculate the break-even formula correctly and can afford the operational costs. A financial + technical problem for VN tech leads in 2026.
-
Wasm beyond the browser: when WebAssembly replaces containers for lightweight microservices WASI and the Component Model are mature, with cold starts of 1-5ms and memory usage at 1/10 of Node.js. However, server-side Wasm is only suitable for a specific layer of workloads — this article helps you decide where to pilot.
-
When the customer is an AI agent: preparing websites and catalogs for agentic commerce before 2027 AI agents are shopping on behalf of humans — and they select brands based on data quality, not banners. What technical steps must Vietnamese e-commerce businesses take to avoid becoming invisible to this wave?
-
42% returning to monolith: re-evaluating the microservices problem through a 2026 cost-centric lens 42% of organizations are merging microservices, and service mesh usage has dropped from 18% to 8%. It's not that microservices are wrong — it's that the real bill has arrived. The framework decides for a team of 10-50 engineers.
-
Tetragon thay Falco + auditd: gom runtime security về một agent eBPF Chạy Falco lẫn auditd trên cùng một node là chồng chéo tốn kém. Đánh giá thẳng việc hợp nhất về Tetragon 1.4: enforce inline trong kernel, TracingPolicy thực tế, và chi phí event data ít ai nói.
-
Webhook đạt chuẩn production 2026: Standard Webhooks, JWKS và bài toán replay Webhook 'chạy được' và webhook 'chạy đáng tin' cách nhau một checklist: chuẩn Standard Webhooks, signing key xoay qua JWKS, retry có DLQ và idempotency key phía nhận.