Tag
A paper arguing that instead of expanding models to handle hard tasks, developers should decompose them by extracting deterministic sub-tasks to a host, leaving only a small generative core to the model — demonstrating 'parity by construction' where model size no longer affects performance on extracted sub-tasks across nine models.
A practitioner discussion on whether AI agent workflows need an independent verifier (smaller SLM, deterministic rules, or hybrid) to catch rule violations and hallucinations, versus relying on self-auditing prompts/skills, weighing accuracy gains against cost and latency overhead.
一篇关于软件设计哲学的工程博客，论述优雅的系统设计往往来自于主动放弃某些需求（如一致性、并发写、低延迟等），并探讨了在业务与技术之间权衡取舍为何困难。
一篇关于 OpenAI 面试流程与经验的复盘：涵盖系统设计（Webhook 投递平台、CI/CD 系统）、编码题与行为面试，指出系统设计环节深入且难度极高，远超普通大厂标准。
A comprehensive, bookmark-worthy roadmap covering software architecture from fundamentals and design principles to distributed systems, DDD, evolution, and communication practices for developers aiming to become architects.
This post reflects on how agent development has mainly been an iterative process of patching weaknesses—adding retrieval, state management, retries, permissions, and evaluation—while noting the ongoing tension between offloading capabilities to supporting systems versus absorbing them back into the models themselves.
An AI engineering commentary arguing that an AI app is far more than an LLM API call, breaking the stack down into compute/storage, serving, models, context, orchestration, agents, and experience layers — advising builders to start with hosted models and only add complexity as needed.
Outcome School 公开了一篇详细教程，讲解如何设计实时语音 AI 智能体，涵盖级联管道（STT→LLM→TTS）、端到端 Speech-to-Speech 模型与混合方案、延迟预算、打断处理（barge-in）、工具调用、记忆与上下文、电话系统集成、规模化部署、边缘情况、可观测性、安全与成本估算。
The article distinguishes between an AI agent's authority to act and having sufficient evidence to justify actions, questioning where the check for adequate evidence should reside in agent systems.
Experienced developers discuss how AI coding agents are transforming their roles into system designers, with concerns about cognitive overhead and the impact on junior developers.
The article reviews the architecture of a chatbot system that converts conversational input into structured data for a constraint solver, emphasizing reliability by separating LLM use for language understanding from deterministic code for decision-making.
The post discusses issues with binding approvals to exact actions in AI agent systems, emphasizing the need for precise mechanisms to handle changes and prevent approval fatigue.
The author explains why a flat agent structure failed in building Hospilot, a multi-agent system for hospital operations, and how adopting a three-level hierarchy (agent → sub-agent → task) improved modularity, planning efficiency, and scalability.
An interactive web-based atlas providing worked examples and architecture diagrams for common system design problems like Twitter feed and chat systems.
A comprehensive guide to core system design building blocks, including client-server architecture, scaling, and databases, to help with system design problems and interview preparation.
An analysis of agent task queues reveals that duplicate claims by different agents can both be verified but only one is counted, leading to invisible waste and issues with self-reported timestamps and lack of exclusion events.
The article explores the concept of 'word coding,' where documents and concepts are structured for AI systems to navigate, offering potential applications in research, writing, and business.
The article lists 10 GitHub repositories essential for system design, including resources for learning scalable architecture and preparing for technical interviews.
The open-source project system-design-101, created by the ByteByteGo team, simplifies system design interview knowledge points through architecture diagrams, has received 84.1k stars, and is suitable for interview preparation and team training.
Google has released a free 1-hour course on Graph Engineering, detailing the transition from single agents to a full 24/7 system with practical guides and self-improving graphs.