AI Digest
← Volver

Digest curado

viernes, 11 de septiembre de 2026·weekly-deep·deep·10,768 tokens

🔥 TOP — lo que SÍ o SÍ tenés que ver

  • [Claude Managed Agents: permission policies con auto] — Nuevo modo donde el server evalúa cada tool call (agente o MCP) y decide correr, denegar o pausar para aprobación; los eventos agent.tool_use y agent.mcp_tool_use ahora reportan el resultado en un campo evaluation junto a evaluated_permission. Es la pieza que faltaba para meter agentes en producción sin bloquear el flow con human-in-the-loop manual. link
  • [Subagents vs Agent Skills: Executing Reusable Knowledge for Long-Horizon Agentic Tasks] — Paper que compara cargar skill packages en el contexto vs invocarlos como subagentes; muestra que a medida que crece el horizonte de tarea lo primero se vuelve frágil porque el razonamiento se degrada al acumular contexto. Directo a tu stack de agentes y a cómo diseñás arquitecturas multi-agente. link
  • [The Menu Is an Execution Prior: State-Path Tool Menus for Online Agents] — Propone tratar el "tool menu" (el subset de tools que ve el agente) como un prior de ejecución sobre rutas de estado, en vez de rankear por relevancia al request. Con miles de interfaces MCP/tools esto es la diferencia entre llamar la acción final sin sus prerequisitos y armar bien la secuencia. link
  • [Shopify vuelve a native: el criterio pasó de "no duplicar" a "el agente hace la traducción"] — Migran de React Native de vuelta a Swift + Kotlin porque los coding agents ya absorbieron suficiente implementación, traducción, testing y review como para que mantener dos codebases sea rentable otra vez. Cambio de razonamiento de fondo sobre cuándo tiene sentido la abstracción cross-platform. link
  • [Do Agents Know When They Succeed? Calibrating Agent Confidence from Internal Representations] — Dos métodos (Latent Trajectory Dynamics y Action Representation Probe) para predecir éxito de una tarea multi-turn desde las representaciones internas del modelo, no desde el output. Clave si vas a correr agentes en contextos donde un fallo silencioso te cuesta caro. link

📦 Claude / Anthropic ecosystem

  • [TeamAI CLI de Tencent] — Gestiona skills, rules, MCP y knowledge compartidos del equipo a través de Claude Code, Codex, Cursor, OpenCode y otros. Útil si estás pensando en estandarizar convenciones de agentes entre personas y no solo en tu máquina. link
  • [CloddsBot] — Terminal de trading autónomo construido sobre Claude, operando en markets de predicción, crypto y futuros con protocolo de pagos máquina-a-máquina. Más interesante como caso de estudio de agente de larga duración con manejo de riesgo que por el producto en sí. link
  • [How to Deal With Errors and Failures in LLM-Powered Applications] — Tratamiento sistemático de los modos de falla de un app con LLM: timeouts, respuestas malformadas, retries y degradación. Buen material para la parte no glamorosa del SaaS. link
  • [EP224: MCP vs RAG vs AI Agents] — Desglose conceptual de dónde termina uno y empieza el otro. Sirve para tener el vocabulario ordenado cuando discutís arquitectura. link

🛠️ Dev tools & coding

  • [Datasette 1.0a39 y 0.65.4 security releases] — Patches de seguridad para ambas ramas; aplicá si corrés Datasette público, especialmente si mezclás tablas públicas y privadas. link
  • [llmfit] — Inspecciona CPU, RAM, GPU(s) y VRAM de tu máquina y te recomienda qué LLMs open source corren con qué cuantización; ahora también mide tok/s reales y los contribuye de vuelta. Directo para decidir inference chica y local. link
  • [Any Nix package, live in your browser] — Una VM x86_64 Linux vía qemu-wasm corriendo en el browser, booteable con cualquier paquete de Nix de los últimos 13 años y direccionable por URL. Impresionante como demo de WASM como plataforma de ejecución. link

🏗️ Software engineering

  • [How we rebuilt Cloudflare Workers' module registry for Node.js compatibility] — Node.js compat ahora por defecto, apps hasta 64 MiB, registry de módulos por URL con import.meta, compilación lazy, caches de código compartidas y mejores errores. Lectura de internals sobre cómo se resuelve un problema real de bundling y compatibilidad a escala. link
  • [1.1.1.1 ahora soporta DNSSEC post-cuántico, los 2.420 bytes incluidos] — Validación de firmas DNSSEC con ML-DSA-44 de NIST, y cómo manejan firmas de 2.420 bytes y riesgos de downgrade a escala. Buen ejemplo de cómo aterrizar criptografía post-cuántica en infraestructura real. link
  • [Automatic Key Exchange: handshakes de origen más rápidos y post-cuánticos para 45 mil millones de conexiones diarias] — Probean qué algoritmos de key agreement soporta cada origen TLS 1.3 y lideran con el más seguro, prefiriendo post-cuántico donde se pueda. Patrón de descubrimiento de capacidades sin romper compatibilidad. link
  • [Built for Reliability: How American Express Processes Payments at Scale] — Cómo fluye una transacción por una arquitectura cell-based y cómo siguen procesando pagos cuando parte de la infraestructura se cae. Ejemplo concreto de diseño de aislamiento de fallas. link
  • [How Smart Model Routing Can Cut LLM Costs by 10X] — Cuándo el routing entre modelos efectivamente ahorra y de qué depende (tipos de request, diferencia de precios, calidad del clasificador). Relevante para el lado de inference del SaaS. link
  • [A Guide to Application Networking Basics] — Repaso de las capas de networking de aplicación de punta a punta. Buen material de repaso conceptual. link
  • [system-design-notes] — Notas del System Design Interview de Alex Xu (Vol 1 y 2), capítulo por capítulo: rate limiter, consistent hashing, key-value store, unique ID generator, URL shortener. Referencia rápida para repasar patrones. link

📚 Vale la pena leer

  • [Fragments: September 8 — La nueva frontera es medible vs no medible] — Christian Catalini argumenta que los primeros productos de AI exitosos fueron chat, imágenes y código no porque fueran los problemas más difíciles, sino porque sus outputs son fáciles de inspeccionar. El límite de automatización dejó de ser rutinario vs no-rutinario y pasó a ser medible vs no-medible. link
  • [OpenDiscoveryTrace: Process Traces for Evaluating AI Scientist Workflows] — Dataset público de 558 trayectorias completas de agentes científicos, con traces estructurados de 9 campos por paso (thoughts, tool calls, observaciones, errores, triggers de revisión, confianza autoreportada). Buen insumo metodológico para evaluar agentes por proceso y no solo por output. link
  • [Quoting Calif Research sobre WeWorm] — El primer gusano zero-click que se propaga por llamadas de WeChat en iOS y Android: el bug y el RCE se encontraron y escribieron en unos dos días con asistencia de AI, y el gusano tomó una semana más. Lo que antes requería un equipo y meses. Vale leerlo por la implicancia en superficie de ataque, no por el amarillismo. link
  • [Social Media Engagement: summer 2026] — Encuesta de Martin Fowler sobre qué servicio le trae engagement y cuál se desplomó desde principios de 2025. Datos concretos sobre el estado de la distribución de contenido técnico. link
  • [Do you even need a presentation?] — Sumeet Gayathri Moghe cuestiona si hace falta la presentación antes de discutir cómo hacerla. Aplica directo a las demos y pitches del SaaS. link

💤 Skippeable pero conviene saber

  • [Claude Platform release notes — resto de los ítems] — El feed trae más entradas que la de Managed Agents; vale escanearlo entero por si hay cambios de SDK o Console que te afecten. link
  • [god's-eye-view] — Simulador de satélite espía en el browser con datos reales de fuentes públicas: vuelo, barcos, satélites, terremotos, cámaras y control por voz con un agente realtime. Buen showcase de visualización + datos abiertos. link
  • [.blend URL Viewer] — Herramienta de Simon Willison para ver archivos .blend desde una URL, en la línea de usar coding agents para generar tooling chico y útil. link
  • [XAI-Arena: Can LLMs Assess the Quality of XAI Explanations?] — Framework LLM-as-a-judge para evaluar explicaciones de XAI en múltiples dimensiones. Interesante si en algún momento tenés que justificar decisiones de modelo ante clientes. link
  • [An Autonomous GeoAI Agent for Arctic Eco-Navigation] — Sistema multi-agente human-in-the-loop para ruteo marítimo en el Ártico que integra criterios operativos, físicos, ecológicos y comunitarios. Caso de aplicación multi-agente fuera del scope típico. link
  • [Valerant: An Automatic Navigable Game Map Generator] — World Action Models aplicados a juegos 3D con geometría persistente, un terreno bastante menos explorado que el 2D. link
  • [Decision-Focused Active Learning for Scale-Aware Critical-Materials Recovery] — Active learning vs space filling para elegir experimentos en recuperación de tierras raras; encuentra el mejor resultado registrado con menos experimentos. link
  • [Adaptive Entangled Game Modules in Artificial General Intelligence] — Marco de probabilidad-onda para comportamiento colectivo de agentes adaptativos, testeado contra datos intradiarios del mercado chino. Especulativo, pero con validación empírica. link
  • [Gradland: On Phenomenal Experience, Differentiated Across Many Dimensions] — Exploración teórica de si la estructura de gradientes y Jacobianos caracteriza la experiencia fenoménica, en un mundo idealizado de redes neuronales. Filosofía de la AI, cero aplicabilidad práctica inmediata. link
  • [awesome-gpt-image-2] — 530+ casos y 20+ templates de prompts para GPT Image 2 / 2.5, con comparativas del mismo prompt entre versiones. Útil si generás assets para el SaaS. link
  • [armorpaint] — Software de pintado de texturas PBR 3D, repo orientado a devs. Solo si te metés con gráficos. link
  • [AINews: Collusion.wiki y un segundo incidente no divulgado de agent swarm de OpenAI] — Reporte de incidentes de enjambres de agentes. Vale tenerlo en el radar por el ángulo de seguridad, pero es agregación de noticias sin fuente primaria. link

Artículos fetched (36)

  • OpenDiscoveryTrace: Process Traces for Evaluating AI Scientist Workflows
    arxiv-ai· 11-sept

    arXiv:2609.09203v1 Announce Type: new Abstract: Existing benchmarks for autonomous AI scientists evaluate only final outputs---generated code, hypotheses, or papers---yet discard the reasoning process by which those outputs were obtained. This makes it impossible to audit scientific methodology, diagnose failure modes, or distinguish systematic reasoning from fortunate guessing. We present \textbf{OpenDiscoveryTrace}, a public dataset of 558 complete AI scientific agent trajectories that captures how models reason, not just what they produce. Each trajectory records a structured 9-field-per-step trace---including thoughts, tool calls, observations, errors, revision triggers, and self-reported confidence---as models execute 124 scientific tasks spanning drug discovery, materials science, g…

  • Adaptive Entangled Game Modules in Artificial General Intelligence
    arxiv-ai· 11-sept

    arXiv:2609.09226v1 Announce Type: new Abstract: We introduce a probability-wave framework for modeling the collective behavior of interacting adaptive agents, deriving testable eigenmodes through a generalized behavioral intelligence (GBI) nonlocal probability-wave equation. This framework captures a broad range of human intelligence behaviors with analytical mechanisms and offers an indirect method to examine the Liu-Chen-Ao (LCA) hypothesis of nonlocal entangled nerve fibers in the brain through collective trader behaviors. Our empirical analysis of Chinese intraday stock market data demonstrates that adaptive entangled game modes explain 82-94% (89% overall) of observed decision patterns, a sharp contrast to the predictions of neoclassical finance based on independent rational agents. …

  • Subagents vs Agent Skills: Executing Reusable Knowledge for Long-Horizon Agentic Tasks
    arxiv-ai· 11-sept

    arXiv:2609.09233v1 Announce Type: new Abstract: How can language model agents effectively leverage libraries of reusable knowledge to solve long-horizon tasks? Recent work has increasingly focused on agent skills: reusable capabilities represented as skill packages, i.e., multi-file bundles containing instructions, scripts, and other resources that help agents perform specific tasks. Agent skills are typically executed by loading their skill instructions into an agent's context and relying on the agent to follow them. As task horizons grow, however, this approach becomes increasingly brittle, because reasoning quality degrades as more information accumulates in the context window. We investigate an alternative approach in which skill packages are instead invoked as subagents. Rather than …

  • Gradland: On Phenomenal Experience, Differentiated Across Many Dimensions
    arxiv-ai· 11-sept

    arXiv:2609.09306v1 Announce Type: new Abstract: This paper investigates the hypothesis that the first-order structure of physical interactions, i.e. gradients or Jacobians, characterizes the structure of phenomenal experience. It does so in an idealized world inhabited by neural networks, Gradland, where the physics are known and the functions are (mostly) differentiable. The paper introduces two measures of Jacobian structure: effective rank and cohesion, based on Kirchhoff complexity. Applying the measures to a series of worked examples shows the hypothesis accounts for: (1) the duration of experience, that it can prolong over hundreds of milliseconds; (2) the difference between what is experienced vividly and obscurely; (3) the experience of texture; (4) the blooming buzzing confusion …

  • An Autonomous GeoAI Agent for Arctic Eco-Navigation
    arxiv-ai· 11-sept

    arXiv:2609.09374v1 Announce Type: new Abstract: Arctic maritime navigation is becoming increasingly important as changing sea-ice conditions expand seasonal accessibility while simultaneously introducing substantial operational, environmental, and community risks. Arctic route planning is inherently a multi-criteria problem: routes that improve vessel safety or efficiency may increase exposure to sea ice, sensitive ecosystems, or nearby communities. Existing routing methods prioritize travel time, fuel use, and navigational risk, often overlooking ecological and community impacts. We introduce a human-in-the-loop, multi-agent GeoAI system for Arctic eco-navigation that integrates operational, physical, ecological, and community-related criteria within a unified routing framework. Multiple…

  • The Menu Is an Execution Prior: State-Path Tool Menus for Online Agents
    arxiv-ai· 11-sept

    arXiv:2609.09395v1 Announce Type: new Abstract: Language models act through tools, yet practical agents face libraries containing thousands of interfaces. We introduce the tool menu as the short, ordered subset of available tools shown to an agent before execution. The agent can call only tools in this menu. Multi-step tasks require the final action and the prerequisite tools that create its inputs in a usable order. Current constructors rank tools by request relevance, which can surface the final action while omitting or delaying less obvious producers. We introduce the state path, a pre-execution route from the observable request state to the desired outcome, and propose State-Path Tool Menu to learn it. Our framework treats the menu as an execution prior over these routes. Its encoder …

  • Decision-Focused Active Learning for Scale-Aware Critical-Materials Recovery
    arxiv-ai· 11-sept

    arXiv:2609.09413v1 Announce Type: new Abstract: Choosing a recovery process for scale-up requires connecting laboratory results with product requirements, process costs, and scale effects. We analyze records from Pacific Northwest National Laboratory's Computer Intelligence for Critical Element Recovery and Optimization (CICERO) workflow for autonomous selective precipitation. Active learning uses prior results to choose experiments. In a conditional retrospective benchmark with fitted models and recycled neodymium-iron-boron (NdFeB) magnet records, active learning finds the best recorded result with fewer experiments than nonadaptive space filling. Enrichment is the selected rare-earth-to-iron ratio relative to that in the feed. Adaptive policies reach the recorded enrichment maximum by …

  • Valerant: An Automatic Navigable Game Map Generator via Action-Conditioned World Model Exploration
    arxiv-ai· 11-sept

    arXiv:2609.09418v1 Announce Type: new Abstract: World Action Models (WAMs) couple predictive world modeling with action generation, allowing anticipated future states to guide agent behavior. Although WAMs are rapidly advancing embodied AI, general-purpose counterparts remain largely unexplored in games. Existing game-oriented approaches often combine action-conditioned world models with external policies and reward functions to realize WAM-like decision-making, yet they operate mainly in 2D visual observation space and do not instantiate persistent 3D geometry. Extending this paradigm to 3D games introduces a distinct challenge. In autonomous driving and robotics, the physical environment exists independently of the model, providing a persistent 3D world in which selected actions can be …

  • XAI-Arena: Can LLMs Assess the Quality of XAI Explanations?
    arxiv-ai· 11-sept

    arXiv:2609.09428v1 Announce Type: new Abstract: Evaluating the quality of explanations produced by explainable AI (XAI) methods remains challenging because existing approaches often rely on subjective human judgment, limiting reproducibility, scalability, and comparability between studies. We examine whether LLMs can serve as a reproducible and scalable mechanism to make comparative assessments of the quality of XAI explanations. We introduce XAI-Arena, an LLM-as-a-judge framework for scalable, reproducible, multidimensional, and stakeholder-sensitive evaluation of XAI explanation quality. XAI-Arena then allows us to compare XAI explanations along various dimensions, namely, perceived simplicity, clarity, task adequacy, trust calibration, actionability, transparency, faithfulness, and ove…

  • Do Agents Know When They Succeed? Calibrating Agent Confidence from Internal Representations
    arxiv-ai· 11-sept

    arXiv:2609.09448v1 Announce Type: new Abstract: As agentic systems getting adopted rapidly in safety critical applications, it is vital to measure the confidence associated with the agentic actions. In comparison to the traditional machine learning systems, agentic workflows have complex failure modes with planning, tool invocation and dynamic environment interactions. In this paper, we investigate whether model's internal representations provide stronger signals of eventual task success in multi-turn agentic setups. We introduce two complementary methods: Latent Trajectory Dynamics (LTD), which summarizes changes in residual-stream representations across an an interaction trajectory, and the Action Representation Probe (ARP), which predicts success from representations formed at action d…

  • A Guide to Application Networking Basics
    bytebytego· 10-sept

    In this article, we will look at the various aspects of networking in detail.16 hrs ago • ByteByteGo774

  • Built for Reliability: How American Express Processes Payments at Scale
    bytebytego· 08-sept

    In this article, we will try to understand how the transaction runs through such a cell-based architecture and how the payments are processed even when…Sep 8 • ByteByteGo23435

  • EP224: MCP vs RAG vs AI Agents
    bytebytego· 05-sept

    An AI agent is kind of an AI system where the agent performs the task autonomously and takes the decisions.Sep 5 • ByteByteGo304711

  • How Smart Model Routing Can Cut LLM Costs by 10X
    bytebytego· 09-sept

    Cost reduction isn’t a given. It also depends on the types of requests the application receives, the price difference between models, and how well the…Sep 9 • ByteByteGo246410

  • How to Deal With Errors and Failures in LLM-Powered Applications
    bytebytego· 07-sept

    Apart from normal processing, the application also sends data to a large language model (LLM). It then uses the model’s response to carry out a task.Sep 7 • ByteByteGo27859

  • Claude Platform release notes
    claude-changelog

    Release notesSubscribeCopy pageUpdates to the Claude Platform, including the Claude API, client SDKs, and the Claude Console.SubscribeCopy pageThe Claude Platform release notes list changes to the Claude API, the client SDKs, and the Claude Console, newest first. For release notes on Claude Apps, see the Release notes for Claude Apps in the Claude Help Center.For updates to Claude Code, see the complete CHANGELOG.md in the claude-code repository. September 10, 2026 Claude Managed Agents permission policies now include auto: the server evaluates each agent or MCP tool call and runs it, denies it, or pauses for your approval. agent.tool_use and agent.mcp_tool_use events report how each call was evaluated in an evaluation field alongside evaluated_permission. See Let the server evaluat…

  • 1.1.1.1 now supports post-quantum DNSSEC, all 2,420 bytes of it
    cloudflare· 10-sept

    1.1.1.1 now validates DNSSEC signatures using NIST’s post-quantum ML-DSA-44 algorithm. Here is how we manage 2,420-byte signatures and downgrade risks at scale.

  • How we rebuilt Cloudflare Workers’ module registry for Node.js compatibility
    cloudflare· 09-sept

    Workers now enables Node.js compatibility by default, supports applications up to 64 mebibytes, and adds a URL-based module registry with import.meta, lazy compilation, shared code caches, and clearer errors.

  • Automatic Key Exchange: faster, post-quantum secure origin handshakes for 45 billion daily connections (and counting)
    cloudflare· 08-sept

    Automatic Key Exchange probes TLS 1.3-capable customer origins to learn which key agreement algorithms they support. We then lead with the most secure algorithm when connecting to the origin, preferring post-quantum connections wherever the origin supports it.

  • AlexsJones/llmfit
    github-trending

    Hundreds of models & providers. One command to find what runs on your hardware. llmfit English · 中文 · 日本語 Find out which open-source Large Language Models (LLMs) your hardware can comfortably run. llmfit inspects your CPU, system RAM, GPU(s), VRAM, and accelerator configuration to recommend models across popular quantizations. 📊 New: benchmark & share — real numbers from your machine, better estimates for everyone. Download a model, serve it, and measure real tok/s on your hardware — then contribute the results back to the project as a PR, straight from the TUI. No gh CLI, no third-party account. Every run is saved locally first, your own measurements replace estimates in the fit table, and each merged submission ships in the next release: anyone on identical hardware gets measured ✓ num…

  • alsk1992/CloddsBot
    github-trending

    Open Source AI trading agent that operates autonomously across 1000+ markets - Polymarket, Kalshi, Binance, Hyperliquid, Solana DEXs, 5 EVM chains. Scans for edge, executes instantly, manages risk while you sleep. Agent commerce protocol for machine-to-machine payments. Self-hosted. Built on Claude. https://www.cloddsbot.com AI-powered trading terminal for prediction markets, crypto & futures Claude + Odds = Clodds Clodds CA: 2puc76ehVHyPXhZmDprtP2phDSFE4kzZKDT4JgAWpump Quick Start • WebChat • Features • Channels • Markets • Launch • Forum • Docs Clodds is a personal AI trading terminal for prediction markets, crypto spot, perpetual futures with leverage, token launches, and Bittensor subnet mining. Run it on your own machine, chat via any of 21 messaging platforms, trade across 10 predic…

  • armory3d/armorpaint
    github-trending

    Graphics Creation Tools https://armorpaint.org armorpaint ArmorPaint is a software for 3D PBR texture painting - check out the manual. Note 1: This repository is aimed at developers and may not be stable. Distributed binaries are paid to help with the project funding. All of the development is happening here in order to make it accessible to everyone. Thank you for support! Note 2: If you are compiling git version of ArmorPaint, then you need to have a compiler (Visual Studio with clang tools - Windows, clang + dependencies - Linux, Xcode - macOS / iOS, Android Studio - Android) and git installed. git clone https://github.com/armory3d/armorpaint cd armorpaint/paint Windows (x64) ..\base\make # Open generated Visual Studio project at `build\ArmorPaint.sln` # Build and run Linux (x64) ../ba…

  • bilawalsidhu/gods-eye-view
    github-trending

    A spy satellite simulator in your browser, except the data is real. Live open source spatial intelligence on a photorealistic 3D globe. https://maptheworld.ai/ 🌐 God's Eye View A spy-satellite simulator in your browser — then you realize the sources are public and the data is real. Photorealistic 3D globe. Live aircraft, ships, satellites, earthquakes, traffic, and public cameras. Hands-free voice control powered by a realtime AI agent. No place left behind. ▶️ From the project behind the viral God's Eye View series (formerly WorldView) — 5M+ on YouTube · 25M+ across socials 🏆 Reached #1 on GitHub Trending, daily and weekly · August 2026 #8 Product of the Day · Hunted by Chris Messina, creator of the hashtag “pretty cool” — Brendan Eich, creator of JavaScript and co-founder of Mozilla a…

  • freestylefly/awesome-gpt-image-2
    github-trending

    Prompt as Code | GPT Image 2 / 2.5 提示词与案例库，530+ 个案例、20+ 套工业级模板与可复用 Skills，新增 2.5 同提示词对比专区，附完整提示词与生成记录，持续更新。 https://gpt-image2.canghe.ai Prompt as Code | GPT-Image2 Industrial Prompt Engine & Template Library, 500+ Reverse-Engineered Cases, 20+ Industrial Templates English | 简体中文 | 日本語 GPT Image 2.5 Spotlight Explore the 2.5 spotlight: learn about Sunburst and Flare, then explore a shared-prompt comparison with a draggable divider, side-by-side view, image enlargement, and generation settings. Sunburst: image generation and precise editing. Flare: fast, high-quality everyday image generation. 4 real recreations: #532 lemon campaign, #527 Rio diorama, #523 Manhattan watercolor, and #510 Bichon Shop icon. Each preserves its original gallery image and uses the full gallery prompt for one new…

  • liquidslr/system-design-notes
    github-trending

    Notes of the book System Desgin Interview - An Insider's Guide https://pagefy-blush.vercel.app/system-design/system-design-interview-by-alex-xu System Design Interview - An Insider's Guide (Vol 1 and 2) These notes are based on the System Design Interview books - Vol 1 and Vol 2 2nd Ed Check the notes here: https://pagefy.io/system-design/system-design-interview-by-alex-xu Note: These notes are a work in progress. Chapter 1 - Scale From Zero To Millions Of Users Chapter 2 - Back-of-the-envelope Estimation Chapter 3 - A Framework For System Design Interviews Chapter 4 - Design A Rate Limiter Chapter 5 - Design Consistent Hashing Chapter 6 - Design A Key-Value Store Chapter 7 - Design A Unique ID Generator In Distributed Systems Chapter 8 - Design A URL Shortener Chapter 9 - Design A Web Cr…

  • Tencent/teamai-cli
    github-trending

    Make Every Team AI Native TeamAI — Make Every Team AI Native English | 简体中文 TeamAI manages your team's skills, rules, MCP, and knowledge across Claude Code, Codex, CodeBuddy, WorkBuddy, OpenCode, Cursor, and other AI agents. Contributors Thanks to everyone who has contributed to TeamAI! Made with contrib.rocks. Quick Start Install npm install -g teamai-cli Team admin / solo user Create a shared-experience repo on your git host (GitHub, GitLab, GitCode, CNB, TGit, or a private Git service), grant write access to team members, then run teamai init https://github.com/yourorg/yourrepo. No team repo yet? Start from a template pre-loaded with production-ready skills, rules, and review agents. Browse the teamai-hub org, click Use this template, then teamai init against your new repo. Team member…

  • [AINews] not much happened today
    latentspace· 10-sept

    a quiet day

  • [AINews] Collusion.wiki: A second undisclosed OpenAI agent swarm incident...
    latentspace· 05-sept

    AI News for 9/2/2026-9/3/2026.

  • Fragments: September 8
    martin-fowler· 08-sept

    Christian Catalini says we’re in a situation where we are vastly reducing the cost of generating things, but not the cost of verifying them:. This explains why the first major AI products appeared in chat, image generation, and code assistance. Not because these were the hardest human problems, but because their outputs were relatively easy to inspect. A user can judge the tone of a message, look at an image, or run a test on a piece of code. […] The old automation boundary was routine versus non-routine work. The new boundary is increasingly measurable versus non-measurable work. The issue is then over how well you can measure something. In our profession, we know there’s a big difference between how many lines of code we write and how productive we are, and we’ve seen a regular failure …

  • Social Media Engagement: summer 2026
    martin-fowler· 09-sept

    A quick survey of recent engagement of my posts on social media, indicating which service has by far the most engagement, and which service has seen a precipitous decline since early 2025. more…

  • Do you even need a presentation?
    martin-fowler· 08-sept

    Like me, Sumeet Gayathri Moghe is tired of poor presentations with bad slide decks. He's started to write a series of posts on how to avoid these calamities, beginning with a post that questions whether a presentation is needed at all. more…

  • Any Nix package, live in your browser
    simonw· 10-sept

    <p><strong><a href="https://fzakaria.com/2026/09/04/any-nix-package-live-in-your-browser">Any Nix package, live in your browser</a></strong></p> Farid Zakaria calls this his "<em>magnum opus</em> of Nix work", and I can see why.</p> <p><a href="https://trynix.dev">trynix.dev</a> provides a <a href="https://github.com/ktock/qemu-wasm">qemu-wasm</a> powered x86_64 Linux virtual machine running entirely in your browser through WebAssembly. That VM can then be booted with <em>any Nix package</em> from the past 13 years. They are URL addressable, so you can navigate to this page:</p> <p><a href="https://trynix.dev/?pkg=python3%403.6.2">https://trynix.dev/?pkg=python3%403.6.2</a></p> <p>Then click "Load" and get an interactive shell against a virtual machine running Python 3.6.2 from 2017.</p> …

  • Datasette 1.0a39 and 0.65.4 security releases
    simonw· 11-sept

    <p><strong><a href="https://datasette.io/blog/2026/september-security-releases/">Datasette 1.0a39 and 0.65.4 security releases</a></strong></p> Today we're releasing two new security patch versions of Datasette: <a href="https://docs.datasette.io/en/latest/changelog.html#v1-0-a39">1.0a39</a> and <a href="https://docs.datasette.io/en/stable/changelog.html#v0-65-4">0.65.4</a> - one for the current alpha series and one for the stable 0.65.x family.</p> <p>These are security fixes which you should apply if you are running a Datasette instance on the public web - in particular if that instance mixes both public and private tables.</p> <p>Following issues reported by <a href="https://github.com/jankesec">Sevban Dönmez</a>, <a href="https://alexgarcia.xyz">Alex Garcia</a> and I ran an extensive …

  • .blend URL Viewer
    simonw· 09-sept

    <p><strong>Tool:</strong> <a href="https://tools.simonwillison.net/blender-viewer">.blend URL Viewer</a></p> <p>I'm continuing to have a lot of fun with GPT-6 Astra and Blender (see <a href="https://til.simonwillison.net/llms/blender-coding-agents-macos">my TIL</a>).</p> <p>As a big fan of the <a href="https://en.wikipedia.org/wiki/Faberg%C3%A9_egg">Imperial Fabergé Easter eggs</a>, I've always thought it would be fun to make some new ones that celebrate popular culture.</p> <p>Yesterday I decided to try out the new <a href="https://simonwillison.net/2026/Sep/8/introducing-chatgpt-images-25/">ChatGPT Images 2.5</a> by <a href="https://chatgpt.com/share/6aa1f6d2-a7d8-83ea-92ec-0daeb8422617">running this prompt</a>:</p> <blockquote> <p><code>Generate a photo of a faberge egg that's themed a…

  • Quoting Calif Research
    simonw· 10-sept

    <blockquote cite="https://calif.io/research/weworm"><p>Today, we're releasing a demo of WeWorm, the first zero-click worm to spread through WeChat calls across iOS and Android. [...]</p> <p>The victim does not need to answer the call, or interact with their phone at all. Even if they do answer, they hear nothing, and the exploit still succeeds. [...]</p> <p>Working with AI, our team found the bug and wrote the first remote code execution (RCE) exploit in about two days. Building the worm took one more week.</p> <p>A worm at this scale used to be the kind of thing that took a larger team months. AI can already do most of the work here. Our team provided the judgment about what to target and how to test it safely.</p></blockquote> <p class="cite">&mdash; <a href="https://calif.io/research/w…

  • Native is now the future of mobile at Shopify
    simonw· 10-sept

    <p><strong><a href="https://shopify.engineering/back-to-native">Native is now the future of mobile at Shopify</a></strong></p> Shopify are moving from React Native back to separate Swift and Kotlin codebases for their native apps, for the exact reason you would expect:</p> <blockquote> <p>We decided to switch from native to React Native in 2020 for three reasons:</p> <ul> <li>Stop building the same features twice</li> <li>Allow developers to work across the stack</li> <li>Spend less time chasing feature parity and more time shipping value</li> </ul> <p>[...]</p> <p>Native still means building and maintaining software on two platforms, that cost has not disappeared. What changed is that agents can now do enough of the implementation, translation, testing, and review work that it’s no longe…