AI Digest

Digest curado

viernes, 03 de julio de 2026·weekly-deep·deep·10,122 tokens

Aquí tenés tu digest semanal. Arrancamos con un TOP lleno de cosas que afectan directamente cómo laburás hoy.

🔥 TOP — lo que SÍ o SÍ tenés que ver

  • Claude Sonnet 5 ya está disponible — Nueva generación de Sonnet lanzada el 30/6, con pricing introductorio de $2/$10 por MTok hasta agosto. Soporta 1M de contexto y 128k de output. Si estabas en Sonnet 4.6, esto es un upgrade directo que podés probar hoy. Detalles en release notes
  • Claude Fable 5 y Mythos 5 restaurados — Volvieron el 1/7 tras una interrupción. Si dependías de alguno para agentes o coding, ya están operativos de nuevo. Statement en release notes
  • Chrome DevTools MCP server oficial — Google lanzó un MCP server que permite a Claude Code, Cursor, Copilot y otros controlar Chrome DevTools: tracing de performance, debugging de red, screenshots, console con source maps. Si hacés desarrollo web con coding agents, esto es enorme. chrome-devtools-mcp
  • llm-coding-agent 0.1a0 de Simon Willison — Un coding agent experimental basado en su LLM library, usando Claude Code for web como runtime. Interesante como alternativa open source a soluciones más pesadas. Post + transcript
  • DSPy para optimizar prompts de Datasette Agent — Simon usó DSPy con Claude Fable 5 para evaluar y mejorar system prompts de generación de SQL. Si te interesa la ingeniería de prompts sistemática, esto es un caso práctico de investigación reproducible. Research log
  • caveman: skill que corta 65% de tokens en coding agents — Prompt engineering extremo: hace que Claude Code (y 30+ herramientas más) hable como cavernícola. Corta ~75% de output tokens manteniendo precisión técnica. Si pagás por token y usás agents intensivamente, esto es ganancia directa. caveman en GitHub
  • PACE: neuro-symbolic framework para counterfactual explanations — Separa predicción de razonamiento simbólico para generar explicaciones factibles. Relevante si trabajás con agentes que necesitan explicar decisiones. arXiv
  • Procedural Memory Distillation (PMD) — Técnica que retiene información procedural entre rollouts de RLVR para auto-mejora sin perder conocimiento entre episodios. Aplica directo a cómo entrenar agentes que mejoran con uso. arXiv

📦 Claude / Anthropic ecosystem

  • Sonnet 5: detalles finos — 1M contexto, 128k output, mismas tools que Sonnet 4.6 excepto PDF support. El pricing introductorio es un 33% off hasta fin de agosto. Release notes
  • Fable 5 safeguards: el framework de jailbreak — Anthropic publicó detalles técnicos sobre cómo blindaron Fable 5 contra jailbreaks. Si estás construyendo sobre él, vale entender los límites. Post de Anthropic
  • "Understand to participate" — Geoffrey Litt en AIE — Framing clave sobre cómo colaborar con coding agents sin acumular deuda cognitiva. Necesitás entender el código lo suficiente para participar activamente con el modelo. Post de Simon

🛠️ Dev tools & coding

  • Career-Ops: job search system sobre Claude Code — 14 skill modes, dashboard en Go, generación de PDFs, batch processing. Open source, transforma cualquier CLI de coding agent en un centro de búsqueda laboral. GitHub
  • ECC: agent harness performance optimization — Sistema que optimiza skills, instincts, memoria, seguridad para Claude Code, Codex, Cursor y más. 211K+ stars, 230+ contributors. GitHub
  • GitHub Actions checkout v7 — Ahora se niega por defecto a checkout de fork PRs en pull_request_target/workflow_run para prevenir "pwn requests". Migrado a ESM. Changelog
  • Multi-agent system for code summarization (Agent4cs) — Framework que resume codebases grandes de abajo hacia arriba usando múltiples agentes especializados. Usa Claude Code como base. arXiv

🏗️ Software engineering

  • ByteByteGo: Multi-Region Architecture — Guía práctica sobre cómo escalar a múltiples regiones sin fundirse en costos de replicación y latency. Post
  • ByteByteGo: Cómo OpenAI da voz en tiempo real a 900M de usuarios — Walkthrough del pipeline completo de voice AI, incluyendo los desafíos de infra que enfrentó el equipo de OpenAI. Post
  • ByteByteGo: RAG vs Graph RAG vs Agentic RAG — Comparativa clara de los tres paradigmas de RAG, con tradeoffs de implementación. Post
  • ByteByteGo: Cómo manejan memoria los AI agents — Desde la restricción que fuerza la existencia de memoria hasta los tradeoffs arquitectónicos. Post
  • The Wiola Architecture for Small Language Models — Arquitectura original de SLM con SRPE (positional encoding helicoidal), GCLA (atención entre capas), y Adaptive Token Merging. Sin parentesco con GPT/LLaMA. arXiv
  • Difficulty-routed control for service agents — Arquitectura que rutea requests entre camino rápido y revisión profunda según complejidad operacional. Aplica directo a tu SaaS de restaurantes si tenés agentes ejecutando writes. arXiv
  • Beyond Next-Token Prediction: RLVR en workflows de Atlassian — Prueba conceptual de RLVR para agentes que interactúan con APIs de Jira/Confluence. Demuestra que entrenar para actuar en APIs específicas cierra el gap del "next-token prediction". arXiv

📚 Vale la pena leer

  • GenPage: homepage generativa end-to-end en Netflix — Cómo Netflix construye homepages generadas por IA para cada usuario. Caso real de personalización a escala con sistemas generativos. Netflix Tech Blog
  • Cloudflare Monetization Gateway — Permitís cobrar por cualquier recurso (web page, dataset, API, MCP tool) mediante x402, sin armar tu propio stack de pagos. Los cobros se liquidan en stablecoins. Blog post
  • Cloudflare: Content Independence Day un año después — Reporte sobre cómo el auge de agentes autónomos está rompiendo el modelo de referrals tradicional y la nueva infraestructura necesaria para una web sostenible. Blog post
  • Cloudflare: nuevas opciones de control de tráfico AI — Ahora podés distinguir y manejar search bots, agent bots y training bots por separado, con protección para páginas con anuncios. Blog post
  • Cloudflare Attribution Business Insights — Dashboard que te muestra comportamiento de crawlers, apetito de datos y valor potencial. Para decidir con quién negociar crawl compensation. Blog post
  • En defensa de los AI mandates (Charity Majors) — Charity argumenta que cuando necesitás un cambio coordinado rápido, un mandate puede ser la forma más honesta de financiarlo. Lecciones desde Honeycomb. Post
  • Thinking Machines Interaction Models (ByteByteGo) — Análisis del concepto de "interaction model" que proponen Thinking Machines, sobre cómo diseñar interacciones con sistemas autónomos. Post

💤 Skippeable pero conviene saber

  • AIE World's Fair daily dispatches — Cobertura del evento: debate sobre loops, software factories, agentic websites. Si te interesa ver hacia dónde va la industria, está bueno. Dispatch 1, Dispatch 2, Vercel sobre agents, Adobe agentic sites, Skill engineering, Autoresearch, Cursor en enterprise, Software factories
  • Diffusion language models for radiology (DiffusionGemma-26B) — Diffusion LLMs compiten con autoregresivos en tareas médicas y son 3.5-4.4x más rápidos en decoding. Nicho pero interesante como señal de que diffusion text está madurando. arXiv
  • CreativityNeuro: weight steering para divergencia creativa — Método data-free que mejora pensamiento divergente en LLMs via contrastive weight steering. Mejora de 14 percentiles humanos en DAT. arXiv
  • Auto-FL-Research: agentic search para algoritmos de FL — Workflow con coding agents para buscar algoritmos de federated learning. Si hacés investigación o trabajás con FL, podés robarte la idea. arXiv
  • World Feedback for Clinical Agents (MedAgentBench-v3) — Auditando RL en entornos clínicos FHIR. Revela ceilings de capability (10/20 tipos de tarea con 0% base) y format-knowledge barriers. Para entender límites de RLVR en dominios regulados. arXiv

Artículos fetched (42)

  • More details on Fable 5’s cyber safeguards and our jailbreak framework
    anthropic-news· 02-jul

    Jul 2, 2026Announcements

  • Agent4cs: A Multi-agent System for Code Summarization in Large Hierarchical Codebases
    arxiv-ai· 03-jul

    arXiv:2607.01425v1 Announce Type: new Abstract: Understanding large, complex codebases, especially those with obfuscated structures and incomplete documentation, remains a significant challenge. Existing code summarization solutions often rely on a single language model or coding assistant like Claude Code, and treat source code as flat text, underutilizing the rich interdependencies and hierarchical information within a repository. To address these shortcomings, we propose Agent4cs - a multi-agent framework that summarizes large codebases in a bottom-up fashion, where a summarization agent focuses on producing robust summaries; a keyword-extraction agent proactively identifies critical information from subfolders; and a quality-assurance agent iteratively refines the outputs for readabil…

  • Discrete Diffusion Language Models for Interactive Radiology Report Drafting
    arxiv-ai· 03-jul

    arXiv:2607.01436v1 Announce Type: new Abstract: Diffusion language models, which generate text by denoising a token canvas bidirectionally instead of emitting tokens left to right, have become competitive with autoregressive (AR) generation. Medical foundation models, however, remain almost entirely autoregressive. We adapt a mixture-of-experts diffusion language model, DiffusionGemma-26B, and benchmark it against its same-size AR sibling Gemma-4-26B under an identical LoRA recipe on medical visual question answering datasets, scored by a verbosity-robust LLM judge. Diffusion matches or exceeds AR on all of them, and the finetuned model (3.8B active) is competitive with frontier vision-language models; its decoding is also 3.5-4.4x faster. Beyond this parity, the diffusion model offers a …

  • Beyond Next-Token Prediction: An RLVR Proof of Concept for Tool-Use Agents on Atlassian Workflows
    arxiv-ai· 03-jul

    arXiv:2607.01465v1 Announce Type: new Abstract: Large language models are trained to predict the next token, not to act inside a specific API. In niche enterprise SaaS workflows -- where success means hitting the right endpoint with the right nested arguments in the right order -- this objective mismatch shows up as silent failures: dropped required fields, hallucinated tools, or early stops after a single read. We ask whether Reinforcement Learning with Verifiable Rewards (RLVR), applied directly in the target environment, closes the gap. As a proof of concept we build a suite of five synthetic environments emulating the Jira REST v3 and Confluence v2 APIs at schema fidelity; rewards are computed entirely from the tool-call trace, with no live API, no learned judge, and no human label in…

  • World Feedback for Clinical Agents: Diagnosing RL in FHIR Environments
    arxiv-ai· 03-jul

    arXiv:2607.01470v1 Announce Type: new Abstract: Clinical protocol-execution tasks -- checking a lab value, applying a threshold, placing a correctly structured FHIR order -- are natural candidates for RL from world feedback: once clinical SMEs encode decision logic into a verifier, that verifier grades unlimited rollouts without per-episode annotation. But applying RL requires a sound feedback channel and sufficient base capability. We audit MedAgentBench v1/v2, find a 41.7\% silent-finish ceiling that makes inaction the RL dominant strategy, and construct \textbf{MedAgentBench-v3 (MAB-v3)} (508 tasks, 8.9\% ceiling). Training Qwen3-8B exposes two structural barriers: a \emph{capability ceiling} (10/20 task types have 0\% base performance, zero gradient) and a \emph{format-knowledge barri…

  • Procedural Memory Distillation: Online Reflection for Self-Improving Language Models
    arxiv-ai· 03-jul

    arXiv:2607.01480v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards (RLVR), along with recent selfdistillation variants such as SDPO, evaluates each rollout against a verifier and updates the policy from that episode-level signal. However, the richer procedural information in the rollout is rarely retained or reused. Across episodes and epochs, the model repeatedly encounters related problems under a changing policy, producing cross-episode signals that episode-local updates cannot capture: which strategies consistently pass verification, which failure modes persist, which patterns recur. We propose Procedural Memory Distillation (PMD), which converts these crossepisode signals into reusable procedural memory and distills it into the policy's weights during trai…

  • Auto-FL-Research: Agentic Search for Federated Learning Algorithms
    arxiv-ai· 03-jul

    arXiv:2607.01366v1 Announce Type: new Abstract: Federated learning (FL) research often depends on many small but consequential algorithmic choices: optimizer variants, server aggregation rules, local training schedules, normalization, regularization, and model architecture. These choices are expensive to explore manually and difficult to compare fairly when candidate changes can also alter the FL training or evaluation path. In this work, we present Auto-FL-Research (AFR), a constrained coding-agent workflow for FL algorithmic recipe search. Agents may propose and implement candidate training algorithms, including server aggregation rules, client update schedules, local objectives, and registered model variants, while task profiles fix the mutation surface, compute budget, communication c…

  • PACE: A Neuro-Symbolic Framework for Plausible and Actionable Counterfactual Explanations
    arxiv-ai· 03-jul

    arXiv:2607.01306v1 Announce Type: new Abstract: Counterfactual explanations explain machine learning predictions by identifying minimal input changes that would alter a model's decision. Although many existing methods successfully generate prediction-changing alternatives, they often produce unrealistic or infeasible recommendations due to a lack of explicit mechanisms for incorporating domain knowledge and intervention constraints. Neuro-symbolic AI offers a promising direction by combining data-driven predictive models with symbolic reasoning capable of representing human-understandable rules and feasible actions. This paper presents PACE, a modular neuro-symbolic framework for generating feasibility-aware counterfactual explanations. The framework separates prediction and reasoning int…

  • The Wiola Architecture for Efficient Small Language Models
    arxiv-ai· 03-jul

    arXiv:2607.01394v1 Announce Type: new Abstract: We present Wiola, a fully original Small Language Model (SLM) architecture built from first principles, sharing no structural lineage with any existing model family including GPT, LLaMA, Mistral, or Falcon. Wiola introduces five independently novel components: (i) Spiral Rotary Positional Encoding (SRPE), which embeds token positions on a three-dimensional helical manifold combining absolute, relative, and hierarchical positional signals; (ii) Gated Cross-Layer Attention (GCLA), providing each decoder layer with soft cross-attention access to compressed summaries of two preceding layers for inter-layer coherence; (iii) Adaptive Token Merging (ATM), which dynamically merges se mantically redundant adjacent tokens in middle network layers to r…

  • When Should Service Agents Reconsider? Difficulty-Routed Control in Customer-Service Operations
    arxiv-ai· 03-jul

    arXiv:2607.01426v1 Announce Type: new Abstract: Autonomous customer-service agents are shifting from conversational interfaces toward operational execution roles: they retrieve firm records, apply service policies, and execute backend writes such as refunds, cancellations, exchanges, order modifications, and reservation changes. This shift creates a service-control problem: firms must keep routine service fast and low-friction while preventing operational errors on requests where customer instructions, policy constraints, firm records, and backend writes interact. We propose a difficulty-routed service-control architecture that asks when service agents should reconsider before acting. A lightweight router keeps routine sessions on a low-cost baseline path and routes operationally coupled …

  • CreativityNeuro: Steering Language Model Weights to Improve Divergent Thinking and Reduce Mode Collapse
    arxiv-ai· 03-jul

    arXiv:2607.01433v1 Announce Type: new Abstract: Divergent thinking is a crucial aspect of creativity, yet large language models (LLMs) tend to consistently generate similar responses to open-ended questions, in what has been termed the artificial hivemind effect. Here, we introduce CreativityNeuro, a data-free method for enhancing divergent thinking in LLMs via contrastive weight steering. We evaluate our method across multiple creativity assessments and report several main findings. On the Divergent Association Task (DAT), a vocabulary-space creativity test, CreativityNeuro improves performance by up to 14 human percentile points. Next, in a large-scale human evaluation (N=720) on the Alternative Uses Test (AUT) and the Task Task, CreativityNeuro achieves significant improvements in orig…

  • Multi-Region Architecture: Going Global Without Going Broke
    bytebytego· 02-jul

    When an application grows geographically, it is logical to start serving it from a second region to improve latency and availability.15 hrs ago • ByteByteGo754

  • How AI Agents Manage Memory and Avoid Forgetfulness
    bytebytego· 29-jun

    In this article, we will try to understand how that architecture gets built, from the constraint that forces it to exist all the way to the tradeoffs…Jun 29 • ByteByteGo301517

  • How OpenAI Delivers Low-Latency Voice AI for 900M Users
    bytebytego· 01-jul

    In this article, we will look at the entire journey in detail and challenges the OpenAI engineering team faced.Jul 1 • ByteByteGo21125

  • Inside Thinking Machines’ Interaction Models
    bytebytego· 30-jun

    In this article, we will look at what the research preview covers and the concept of an interaction model proposed by Thinking Machines.Jun 30 • ByteByteGo22137

  • EP220: RAG vs Graph RAG vs Agentic RAG
    bytebytego· 27-jun

    RAG connects LLMs to your data and there are three different ways to do it.Jun 27 • ByteByteGo273413

  • In defense of AI mandates (xpost)
    charity-majors· 02-jul

    When you need to execute a coordinated change on a tight timeline, a mandate might be the best and most honest way to fund it. Crossposted from here. I’ve been writing a series of pieces on lessons learned from our AI journey at Honeycomb. I’ve written about the tension between enthusiasts and skeptics, the need for engineering rigor, […]

  • Claude Platform
    claude-changelog

    Release notesCopy pageUpdates to the Claude Platform, including the Claude API, client SDKs, and the Claude Console.Copy page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. July 1, 2026 We've restored access to Claude Fable 5 and Claude Mythos 5. See our statement for more information. June 30, 2026 We've launched Claude Sonnet 5 (claude-sonnet-5), the next generation of our Sonnet model family, at introductory pricing of $2 / $10 per MTok through August 31, 2026 (standard $3 / $15 thereafter). Claude Sonnet 5 supports a 1M token context window, 128k max output tokens, and the same set of tools and platform features as Claude Sonnet 4.6, except Pr…

  • Loading...
    claude-changelog

    Loading...Loading...Loading...Loading...Loading...Loading...Loading...Loading...Loading...Loading...Loading...Loading...Loading...Loading...Loading...

  • Unmasking the crawls with Attribution Business Insights
    cloudflare· 01-jul

    Cloudflare’s new Attribution Business Insights dashboard helps website owners understand crawler behavior, appetite, and potential value, fueling business-level conversations around crawl compensation.

  • Content Independence Day, one year on: building the business model for the agentic Internet
    cloudflare· 01-jul

    One year after declaring Content Independence Day, a dynamic market for monetized content has officially emerged. In this report, we examine how the rise of autonomous AI agents is upending traditional search referrals and detail the new infrastructure required to support a sustainable web economy.

  • Your site, your rules: new AI traffic options for all customers
    cloudflare· 01-jul

    For our second Content Independence Day, we’re giving website owners finer options to manage AI traffic. Instead of a one-size-fits-all block, all customers can now easily distinguish and manage Search, Agent, and Training bots, alongside the new ability to protect ad-monetized pages.

  • Making AI search smarter
    cloudflare· 01-jul

    Search is how we find nearly everything on the web — creators, merchants, answers. AI is rewriting the rules, leaving creators caught between staying discoverable in an agentic era and getting paid for their work. Today we're launching two initiatives to help.

  • Announcing the Monetization Gateway: charge for any resource behind Cloudflare via x402
    cloudflare· 01-jul

    We're opening the waitlist for our Monetization Gateway, which will allow you to charge for any web page, dataset, API, or MCP tool behind Cloudflare. The charges will settle in stablecoins over the x402 open protocol, with no payments stack of your own to build.

  • affaan-m/ECC
    github-trending

    The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond. Language: English | Português (Brasil) | 简体中文 | 繁體中文 | 日本語 | 한국어 | Türkçe | Русский | Tiếng Việt | ไทย | Deutsch | Español Warning Official sources only. Install ECC only from verified channels: the GitHub repository github.com/affaan-m/ECC, the npm packages ecc-universal and ecc-agentshield, the GitHub App, the plugin slug ecc@ecc, and the project website ecc.tools. Third-party re-uploads and unofficial mirrors are not maintained or reviewed by the project and may contain malware. 211.9K+ stars | 32.5K+ forks | 230+ contributors | 12+ language ecosystems | Cross-harness agent workflows Language / 语言 / 語言 / Dil / Язык /…

  • JuliusBrussee/caveman
    github-trending

    🪨 why use many token when few token do trick — Claude Code skill that cuts 65% of tokens by talking like caveman caveman why use many token when few do trick Before/After • Install • What You Get • Benchmarks • Full install guide A Claude Code skill/plugin (also Codex, Gemini, Cursor, Windsurf, Cline, Copilot, 30+ more) that makes agent talk like caveman — cuts ~75% of output tokens, keeps full technical accuracy. Brain still big. Mouth small. Before / After 🗣️ Normal Claude (69 tokens) "The reason your React component is re-rendering is likely because you're creating a new object reference on each render cycle. When you pass an inline object as a prop, React's shallow comparison sees it as a different object every time, which triggers a re-render. I'd recommend using useMemo to memoize…

  • santifer/career-ops
    github-trending

    AI-powered job search system built on Claude Code. 14 skill modes, Go dashboard, PDF generation, batch processing. Career-Ops English | Deutsch | Español | Français | Português (Brasil) | 한국어 | 日本語 | 简体中文 | 繁體中文 | Українська | Русский | Polski | Dansk | العربية I spent months applying to jobs the hard way. So I engineered the system I wish I had. Companies use AI to filter candidates. I just gave candidates AI to choose companies. Now it's open source. FEATURED IN 740+ job listings evaluated · 100+ personalized CVs · 1 dream role landed Also runs on any agent-skill-standard CLI. See Supported CLIs. What Is This Career-Ops (career-ops.org, also known as careerops) turns any AI coding CLI into a full job search command center. Instead of manually tracking applications in a spreadsheet, you …

  • ChromeDevTools/chrome-devtools-mcp
    github-trending

    Chrome DevTools for coding agents Chrome DevTools for agents Chrome DevTools for agents (chrome-devtools-mcp) lets your coding agent (such as Antigravity, Claude, Cursor or Copilot) control and inspect a live Chrome browser. It acts as a Model-Context-Protocol (MCP) server, giving your AI coding assistant access to the full power of Chrome DevTools for reliable automation, in-depth debugging, and performance analysis. A CLI is also provided for use without MCP. Tool reference | Changelog | Contributing | Troubleshooting | Design Principles Key features Get performance insights: Uses Chrome DevTools to record traces and extract actionable performance insights. Advanced browser debugging: Analyze network requests, take screenshots and check browser console messages (with source-mapped stack…

  • actions/checkout
    github-trending

    Action for checking out a repo Checkout v7 What's new Safer fork pull request handling: checkout now refuses to check out fork pull request code by default when the workflow is triggered by pull_request_target or workflow_run. These triggers run with the base repository's GITHUB_TOKEN, secrets, and runner access, where executing a fork's code commonly leads to "pwn request" vulnerabilities. To opt in after reviewing the risks, set the new allow-unsafe-pr-checkout: true input. Migrated actions/checkout to ESM to support new versions of the @actions/* packages. Updated direct and transitive dependencies, including security fixes for known vulnerabilities. Checkout v6 What's new Improved credential security: persist-credentials now stores credentials in a separate file under $RUNNER_TEMP ins…

  • Vercel's Andrew Qu on why agents are a new kind of software
    latentspace· 03-jul

    The Vercel Chief of Software explains how its agent framework, eve, was created — and why skills, sandboxes and agent-readable websites now matter.

  • Skill engineering and the case against one-shot AI design
    latentspace· 02-jul

    Paul Bakaus talks to us about Impeccable, human judgment in a 'loopmaxxing' era, and why agents still need people to steer them.

  • Warp CEO Zach Lloyd on why software factories are the next phase of coding
    latentspace· 01-jul

    Warp's founder thinks every major software project will soon run on an automated factory. He discusses why and how engineers should prepare for this shift.

  • 🔬 The Coolest Diffusion Research Isn't in LLMs — Evan Feinberg & Sergey Edunov, Genesis Molecular AI
    latentspace· 01-jul

    Why the Llama lead left Meta for drug discovery, PEARL's zero-shot OpenBind win, and what becomes possible when co-folding finally crosses the accuracy threshold.

  • The website of the future may assemble itself for every visitor
    latentspace· 02-jul

    Adobe is experimenting with “agentic sites” that generate pages around an individual user’s intent. At AIEWF, we talked to Carlos Sanchez about the Web's future.

  • AIEWF Daily Dispatch: Autoresearch and the tension between AI and human agency
    latentspace· 02-jul

    The software factory vision met resistance today from speakers defending human understanding and control.

  • AIEWF Daily Dispatch: The great loops debate and the state of AI engineering
    latentspace· 03-jul

    The AI Engineer World’s Fair ended with a debate about loops, a report on the state of AI engineering, and closing keynotes focused on what to build next.

  • Autoresearch: The feedback loop behind self-improving agents
    latentspace· 01-jul

    Introspection co-founder Roland Gavrilescu explains autoresearch, agent “recipes,” self-improving loops, and why humans remain central to the software factory.

  • How Cursor deploys AI inside the enterprise
    latentspace· 01-jul

    Cursor's Pauline Brunet explains how her team of Forward Deployed Engineers help organizations implement agents — essentially setting up software factories.

  • GenPage: Towards End-to-End Generative Homepage Construction at Netflix
    netflix-tech· 29-jun
  • Understand to participate
    simonw· 02-jul

    <p>I saw Geoffrey Litt speak at <a href="https://www.ai.engineer/worldsfair/2026">AIE</a> yesterday, and one framing he used particularly resonated with me:</p> <p><strong>Understand to participate</strong></p> <p>Geoffrey was talking about the challenge of collaborating with coding agents as they construct increasingly large and sophisticated changes, and the need to avoid taking on <a href="https://simonwillison.net/tags/cognitive-debt/">cognitive debt</a> as your understanding drifts from how the code actually works.</p> <p>His argument is that you need to understand the code to a depth that enables you to participate further with the model:</p> <blockquote> <p>You can learn what the agent is doing to make sure you can be an active participant in the creative process. [...]</p> <p>You …

  • llm-coding-agent 0.1a0
    simonw· 02-jul

    <p><strong>Release:</strong> <a href="https://github.com/simonw/llm-coding-agent/releases/tag/0.1a0">llm-coding-agent 0.1a0</a></p> <p>Another Fable 5 experiment. Now that my <a href="https://llm.datasette.io/">LLM library</a> has evolved into more of an agent framework it's time to see what a simple coding agent would look like built on it.</p> <p>I started a <a href="https://github.com/simonw/llm-coding-agent/tree/2466fa03ba8e5122c3bfa93d52167d33bce40ac6">new Python library</a> using my <a href="https://github.com/simonw/python-lib-template-repository">python-lib-template-repository</a> GitHub template repository, then ran these two prompts (here's the <a href="https://claude.ai/code/session_01TEUBvBbMipbFSoqjMiJ7ha">Claude Code for web transcript</a>):</p> <blockquote> <p><code>Write a…

  • Using DSPy to evaluate and improve Datasette Agent's SQL system prompts
    simonw· 02-jul

    <p><strong>Research:</strong> <a href="https://github.com/simonw/research/tree/main/dspy-datasette-agent-prompts#readme">Using DSPy to evaluate and improve Datasette Agent&#x27;s SQL system prompts</a></p> <p>One of this morning's AIE keynotes covered <a href="https://github.com/stanfordnlp/dspy">dspy</a>, which reminded me I've been meaning to see if it could help me improve the system prompt used by <a href="https://agent.datasette.io">Datasette Agent</a> - so I fired off an asynchronous research task in Claude Code for web using Claude Fable 5:</p> <blockquote> <p><code>Pip install the latest Datasette alpha and datasette-agent and dspy - then figure out how to use dspy to evaluate and improve the main system prompts used by Datasette Agent for the feature where it can execute read onl…