玄子 @gavincoding
Prompt Engineer • AI Explorer Deep dive into AI, software engineering & engineering management. Teaching, learning, and scaling ideas that matter. hashnode.com/@harness Joined December 2016-
Tweets159
-
Followers10
-
Following223
-
Likes21
麻省理工这本爆火的深度学习教材，居然有完整免费中文版。 《深入理解深度学习》从基础一路讲到 神经网络、CNN、Transformer、扩散模型、强化学习等核心内容，而且每章都配了 PPT + Python 练习代码，非常适合系统自学。 重点是：原版免费公开，中文版也有人完整翻译整理好了。 原书地址：udlbook.github.io/udlbook/ 中文地址：github.com/exacity/deeple…
昨天看到一个变态的 Code Review skills 应该是 Cursor 官方一个非常严格的代码质量审查技能。Thermo-Nuclear Code Quality Review 热核级别代码审查.... skills.sh/cursor/plugins… 里面的 rule 翻译一下是这样的: 结构简化优先： 要主动找能让整段分支、helper、mode、条件判断直接消失的重构方案 1k 行红线：不允许 PR 把一个原本不到 1000 行的文件推过 1000 行，优先拆分，有点变态。 拒绝意面式增长：高度警惕往无关流程里塞临时 if、特殊分支、一次性布尔标志 这是设计问题，不是风格问题。 警惕串行编排与非原子更新：这个有用！review 的好多 bug 都是这个问题 还有几条不写了，大家可以自己去看 我给项目用了一下，然后就哭了，找到太多干脆不修了，新开一个项目我可能会用一下，假如大家希望用极高标准去找结构问题 强迫做大幅简化 可以用它试试看。
终于看到有人写 Agent 降本增效的文章了。腾讯技术工程团队这篇文章我觉得讲得非常的详细。我发现现在一个阶段就是大家在哐哐的造 Agent， 但是呢，怎么去优化成本，还没有完全顾虑到。 Agent 降本最有效的不是换便宜模型，是把 token 消耗当工程账来管。 文章里团队实测：一个中型需求从 70 万 token 压到 31 万，17 轮降到 9 跑，功能一行没砍。做法值得抄： 一、先拆账再动手。按调用链追踪，把消耗拆成六类：系统 prompt、工具返回、读的文件、系统提示词、历史消息、用户输入。大头是历史消息，append-only 滚雪球，来自工作流层面占位计费。 二、该看的按需加载。SKILL.md 只留骨架，40 行 spec 模板、38 行 DB 规则这种只在特定场景用得上的内容，六项全移进 references，用到才读。搜代码先查 AST 代码图谱锁定文件范围再 read，少几十轮盲搜，实测 -226.7%。 三、不该看的别进来。TAPD 需求原文、Figma 节点树这种上万行的 payload，主 Agent 不许自己碰，配短命子 Agent 只回传摘要，跑完销毁，后端 Agent 压根看不到 Figma 的 Schema。 四、重复的别反复计费。派发提示词动态内容后置，前缀稳定才吃满 KV Cache，缓存读取只要一折；进度看板外化到文件不堆会话；CLI 输出接 rtk 压缩来优化。 五、测试、变量强、修复轮次最多的 Agent 直接换便宜模型，再省 66%。 另外，最省钱的调用是不调用。确定性操作写死成脚本，LLM 只留给真正需要理解语义的环节。 你们的 Agent 账单拆开看过吗？😂 原文：mp.weixin.qq.com/s/TIdXNlrcAOUZ…
GitHub 这一篇 harness 实战值得学习
You're not behind. There's no secret everyone else has. There's just the harness, and it's mostly all you need. @burkeholland gives you a simple, repeatable workflow for GitHub Copilot. x.com/i/article/2039…
Want to master LLM Cache Management? Start with these resources. KV Cache: huggingface.co/docs/transform… hamzaelshafie.bearblog.dev/paged-attentio… Prefix Caching docs.vllm.ai/en/latest/feat… Continuous Batching anyscale.com/blog/continuou… Speculative Decoding research.google/blog/looking-b… KV Cache Quantization huggingface.co/blog/kv-cache-… Research Papers: 1. FlashInfer (Attention Engine) arxiv.org/abs/2501.01005 2. Zipage (Compressed PagedAttention) arxiv.org/abs/2603.08743 3. IceCache (Memory-Efficient KV Cache) arxiv.org/abs/2604.10539
AI 编程时代，最难放下的东西，可能是「我必须亲手看完每一行代码」。 Antirez 这篇文章挺有冲击力：未来程序员真正要控制的，可能并不是代码本身，而是代码背后的想法。 因为 AI 已经能生成大量局部质量不错的代码，如果每天都把时间花在逐行 review 几千行代码上，人的价值反而会被锁在低层细节里。 更重要的工作是想清楚软件到底要做什么，设计是否合理，关键抽象是否正确，测试是否足够，质量边界在哪里。 这不是鼓励闭眼 vibe coding。 相反它要求人对系统有更强的 mental model。你不一定要控制每个函数怎么写，但你要控制整个软件的方向、结构和判断标准。 我觉得这个观点有价值的地方在于： 以后优秀程序员的标志，不再是「我写了多少代码」，而在于「我能不能定义清楚什么代码值得被生成」。
Control the ideas, not the code: blog post here antirez.com/news/169
1) Load Balancing: x.com/system_monarch… 2) CDN: x.com/system_monarch… 3) Caching: x.com/system_monarch… 4) Cache Invalidation: x.com/system_monarch… 5) Rate Limiting: x.com/system_monarch… 6) API Gateway: x.com/system_monarch… 7) CAP Theorem: x.com/system_monarch… 8) Sharding: x.com/system_monarch… 9) Replication: x.com/system_monarch… 10) Partitioning: x.com/system_monarch… 11) Queues: x.com/system_monarch… 12) Microservices: x.com/system_monarch… 13) Microservices Vs Monoliths: x.com/system_monarch… 14) Fault Tolerance: x.com/system_monarch… 15) Database Scaling: x.com/system_monarch… 16) Service Discovery: x.com/system_monarch… 17) Consistency models: x.com/system_monarch… 18) Eventual Consistency: x.com/system_monarch… 19) Distributed Transactions: x.com/system_monarch… 20) Leader Election: x.com/system_monarch… 21) Horizontal vs Vertical Scaling: x.com/system_monarch… 22) Back of the Envelope Estimation: x.com/system_monarch… 23) Idempotency, Data Latency & Finale: x.com/system_monarch…
I've started a 25-day series on Scaling and Architecture. One topic per day. Day 1: Load Balancing Day 2: CDN Day 3: Caching (the 5 layers) Day 4: Cache Invalidation Day 5: Rate Limiting Today is Day 6: API Gateway. If load balancing is how you distribute traffic and rate
عن تجربة، هذه أفضل مهارات تصميم الواجهات مع الـ AI: • github.com/emilkowalski/s… • impeccable.style • tasteskill.dev بتفرق كثير بجودة الـ UI وتخلي النتائج أنظف وأذوق.
Andrej Karpathy spent 70 minutes explaining how top AI users actually work with LLMs. And most people are overcomplicating it. Bookmark it for later.
Anthropic 刚刚发布了一门关于 Loop Engineering 的免费课程，推荐学习。 x.com/precisox/statu…
Anthropic 官方团队把 Claude Code 从底层原理到真实项目实战，完整讲了一遍！ 如果你只是会让 Claude Code 写代码，却还没搞懂它如何读取文件、调用工具、管理权限、调度 subagents，这个视频值得收藏！ 我整理了重点时间线： 00:00｜Claude Code 底层如何工作 04:51｜Permissions：哪些操作允许、询问或拒绝 10:46｜如何把 Code Review、部署流程做成 Skills 15:08｜Intent-driven development 到底怎么做 22:40｜用语音 Prompt 从零构建 3D 游戏 32:55｜用 Branch、Draft PR 和 GitHub CLI 控制交付 38:15｜Auto Mode 如何减少重复授权 40:48｜用 Dynamic Workflows 并行调度多个 Subagents 46:53｜从 Demo 走向 Production 的工程原则 55:33｜Claude Code 如何用于研究、数据和媒体处理 我认为最值得看的是 Lydia Hallie 拆解了 Claude Code 的真实运行方式：模型本身没有长期状态。 文件内容、终端输出、Git 分支、工具调用结果和会话历史，都是由外层的 Agent Harness 维护，再在每一轮重新组装给模型。 这也解释了为什么真正影响 Claude Code 效果的，不只是 Prompt，还有：CLAUDE.md、Skills、Permissions、Verification、Git Workflow、Subagent 编排！ 视频最后还有一个很重要的提醒：AI 能一次生成大量代码，不代表这些代码都应该被合并。 架构、验证、安全、Review 和最终质量，依然由人负责。 这可能是目前理解 Claude Code 最完整的一小时。 建议先收藏，再跟着时间线看！
Building a Coding Agent? You need a fast, rock-solid base. 🤔 To avoid bloated frameworks, I built a high-performance CLI from scratch. Here’s my dev process & lessons learned 👇 medium.com/p/building-a-c… #AIAgent #CLI
Original video: youtu.be/GrNbuWWJYiI?si…
guge @gugeaigc
183 Followers 7K Following
Mei Nakajima @anwanoori
64 Followers 479 Following HSBC audits, invests in coffee shops and clothing stores
Wind @productdevbook
2K Followers 572 Following I am looking for a full-time job. Full-stack developer.
亚当 @yx465676231
1 Followers 6 Following
chany khan @Ishkhan_001
26 Followers 450 Following
@levelsio @levelsio
955K Followers 3K Following 💸https://t.co/sQ0aiU7v02 📸https://t.co/lAyoqmSBRX $90K/m 🎮https://t.co/tF49atJ396 $44K/m https://t.co/8rYr9A594z $25K/m 🏡https://t.co/1oqUgfD6CZ $21K/m 👙@X $20K/m 🌍https://t.co/UXK5AFqCaQ $12K/m 💾@pieter 🏩https://t.co/4p4dzTDVN6
Jim Fan @DrJimFan
586K Followers 3K Following NVIDIA Director of Robotics & Distinguished Scientist. Co-Lead of GEAR lab. Solving Physical AGI, one motor at a time. Stanford Ph.D. OpenAI's 1st intern.
swyx @swyx
191K Followers 4K Following achieve ambition with intentionality, intensity, integrity & insanity. affiliations: - @smol_ai - @dxtipshq - @cognition - @aidotengineer - @latentspacepod
Hamel Husain @HamelHusain
55K Followers 3K Following Evals Evals Evals - https://t.co/Zrmp6LRd9c About Me: https://t.co/P6WyeKkyTa
Riley Goodside @goodside
227K Followers 4K Following Screenshots of chatbots since 2022. Formerly: Google DeepMind, Scale.
Jerry Liu @jerryjliu0
84K Followers 2K Following Parsing the world's hardest PDFs @llama_index. cofounder/CEO Careers: https://t.co/EUnMNmbCtx Enterprise: https://t.co/Ht5jwxSrQB
Harrison Chase @hwchase17
130K Followers 587 Following @LangChain Always hiring: https://t.co/D5Ut3loFO7
Demis Hassabis @demishassabis
1.8M Followers 179 Following Nobel Laureate. Co-Founder & Chair @GoogleDeepMind; Chief Scientist of Alphabet. Founder & CEO @IsomorphicLabs. Building the future...
SpaceXAI @SpaceXAI
2.1M Followers 6 Following
Ian Goodfellow @goodfellow_ian
380K Followers 1K Following Co-founder of stealth startup. Inventor of GANs. Lead author of https://t.co/M6vl8pEQ4I Founding chairman of @pubhealthaction
François Chollet @fchollet
728K Followers 829 Following Co-founder @ndea. Co-founder @arcprize. Creator of Keras and ARC-AGI. Author of 'Deep Learning with Python'.
Google Cloud Tech @GoogleCloudTech
1.3M Followers 2K Following Follow along for how-tos, demos, product news, and more. For company updates, check out @GoogleCloud. Watch #GoogleCloudNext on demand ⬇️
OpenClaw🦞 @openclaw
545K Followers 26 Following Personal AI that actually does things. Your agent, your machine, your rules. Now a 501(c)(3) non-profit foundation — open and independent, forever. 🦞
Ian Nuttall @iannuttall
81K Followers 102 Following Building stuff on the internet since 2005. Working on https://t.co/qi6VVnpjtE and other things.
quilspace @quil_space
34 Followers 27 Following Quilspace is the first AI author that creates ready-to-sell ebooks
Santiago @svpino
456K Followers 600 Following Computer scientist. I teach hard-core AI/ML Engineering at https://t.co/THCAAZcBMu. YouTube: https://t.co/pROi08OZYJ
孙宇晨（去过太... @sunyuchentron
632K Followers 284 Following 👨💻 企业家 | 🤵♂️ 外交官 | 👨🚀 宇航员 #712 | 🍌 艺术收藏家 | 创始人 @trondaoCN | 顾问 https://t.co/UaNI0eB7Vc https://t.co/KLklaEG0qB https://t.co/Hkyam1WH8d | 英文@justinsuntron
Aaron Levie @levie
3.4M Followers 946 Following ceo @box - your business lives in content. unleash it with AI
NVIDIA GeForce @NVIDIAGeForce
3.1M Followers 958 Following Introducing GeForce RTX 50 Series graphics cards & laptops, powered by NVIDIA Blackwell and AI.
Sam Altman @sama
6.2M Followers 1K Following The mission of OpenAI to ensure that AGI benefits all of humanity
Lex Fridman @lexfridman
5.5M Followers 693 Following Host of Lex Fridman Podcast. Interested in robots and humans.
OpenAI @OpenAI
5.3M Followers 4 Following OpenAI’s mission is to ensure that artificial general intelligence benefits all of humanity. We’re hiring: https://t.co/dJGr6LgzPA
Bun @bunjavascript
82K Followers 7 Following Bun is a fast, all-in-one toolkit for installing, bundling, running and testing JavaScript & TypeScript. To install: `npm i -g bun`
OpenCode @OpenCode_AI
1K Followers 0 Following This was the account for the original OpenCode project, which has joined @charmcli with its original author, @h_kujtim.
DAN KOE @thedankoe
1.0M Followers 960 Following the content brain for creators and agents: @edendotso
Claude Code Community @claude_code
48K Followers 73 Following Community account for sharing ClaudeCode related projects and releases. Views/shares independent from @AnthropicAI positions.
Mike Krieger @mikeyk
467K Followers 271 Following Building at Anthropic Labs @anthropicai. Before: CPO at Anthropic, co-founder & CTO of @instagram and @artifact_news
Simon Willison @simonw
221K Followers 6K Following Creator @datasetteproj, co-creator Django. PSF board. Hangs out with @natbat. He/Him. Mastodon: https://t.co/t0MrmnJW0K Bsky: https://t.co/OnWIyhX4CH
宝玉 @dotey
248K Followers 2K Following AI Engineer, dedicated to learning and disseminating knowledge about AI, software engineering, and engineering management.
Maryam Miradi, PhD @MaryamMiradi
9K Followers 125 Following VP & Chief AI Scientist | LinkedIn 100k+ | 20+ Years in AI | Built 400+ AI Agents | Founder & Instructor, AI Agents Course (3,200+ Students) | 46k+ Newsletter
Femke Plantinga @femke_plantinga
14K Followers 610 Following learn with me about AI. growth @SliteHQ
Victoria Slocum @victorialslocum
10K Followers 616 Following learning cool stuff, machine learning engineer at @weaviate_io 💙
Olivia Mitchell @OliviaMitc4151
522 Followers 137 Following AI Enthusiast |AI Art & SEO Explorer | AI工具种草 | 副业搞钱探索
yv @yvbbrjdr
10K Followers 455 Following exists as 451; opinions are my own; Creator of @LANDropApp, @AthenaAGI, LMRouter; MTS @MicrosoftAI pretraining; Ex-@NVIDIA RISC-V sec; EN @realJerryZhou
LangChain @LangChain
264K Followers 190 Following Own your intelligence. Makers of LangSmith, @LangChain_OSS, and @LangChain_JS.
LXDAO @LXDAO_Official
6K Followers 54 Following LXDAO is an R&D-driven DAO building an Infinite Cycle to help sustain open-source projects and public goods. Telegram: https://t.co/1SX0Pp9WOj
Nuxt Nation Conferenc... @NuxtNation
3K Followers 23 Following The #1 online event for all things @nuxt_js! Follow @frontendnation for updates on all future Nuxt events.
Aurimas Griciūnas @Aurimas_Gr
36K Followers 793 Following 🔨 Founder & CEO @ SwirlAI 📖 Writing about #LLM, #AI, #DataEngineering, #MachineLearning and #Data ✍️ Author of SwirlAI Newsletter.You might like
