📰 Revue IA — 23/03/2026

Gestion du contexte de l'IA chez Claude, Cursor, Kiro, Gemini et les agents personnalisés

Si vous utilisez plus d'un agent de codage AI, vous avez probablement remarqué que chacun veut son propre fichier de contexte. Claude Code lit CLAUDE.md . Le curseur lit .cursorrules. GitHub Copilot lit .github/copilot-instructions.md . Kiro lit .kiro/steering/*.md . Windsurf lit .windsurf/rules/*.md . Gemini CLI et Antigravity lisent GEMINI.md . Et puis il y a AGENTS.md et llms.txt . C'est neuf différentes

J'ai trouvé un ID de modèle caché dans l'API Curseurs. Ce n'était pas ce qu'ils disaient.

On March 19, Cursor shipped Composer 2 — its most capable proprietary coding model to date. The company marketed it as a breakthrough: faster completions, better multi-file edits, a tenth of the price of Claude Opus 4.6. Elon Musk endorsed it. Developers rushed to try it. Less than 24 hours later, a developer named Fynn found a model ID hidden in Cursor's API configuration: kimi-k2p5-rl-0317-s515-

Le curseur atteint 1 million d'utilisateurs quotidiens

Cursor just hit 1 million daily active users. Let that number sink in — a code editor built by a small startup is now used daily by more developers than most enterprise tools. Companies like Stripe and Figma are using it. And now they're launching Composer 2 — an AI agent that handles entire multi-file development tasks, not just autocomplete. Here's what I think most people are getting wrong abou

Agent curseur et compositeur : un flux de travail pratique pour le codage quotidien

This post was created with AI assistance and reviewed for accuracy before publishing. Cursor is an AI-native editor built on VS Code. It exposes Composer for multi-file edits and Agent for longer, tool-using runs that search and change code across your repo. Product details and defaults change frequently; treat Cursor’s documentation and changelog as the source of truth, not third-party listicles.

Pourquoi votre SaaS « Vibe Coded » échouera à 100 utilisateurs (et comment y remédier)

It's 2026. You just built a functional SaaS MVP in four hours using Cursor and Claude. It looks great, the happy path works, and you're ready to tweet your launch. But there's a hidden tax on AI-generated code: Architectural Debt. When you vibe-code without a strict foundation, the LLM takes the path of least resistance. It puts database logic in your routes, skips error handling, and ignores race

Que s'est-il passé lorsque le curseur a refusé d'écrire plus de code (et ce qu'il montre sur les limites de l'IA)

This post was created with AI assistance and reviewed for accuracy before publishing. In March 2025, a developer using Cursor AI for a racing game hit an unexpected wall. After about an hour of "vibe coding" and roughly 750 to 800 lines of generated code, Cursor stopped and told the user to write the logic himself. The incident went viral on Hacker News and was covered by Ars Technica and TechCrun

J'ai construit un serveur MCP pour automatiser les importations de produits de dropshipping

Le problème qui ne disparaîtrait pas Je gère des magasins de dropshipping depuis quelques années. Si vous l'avez fait, vous connaissez l'exercice : trouvez un produit sur AliExpress ou une plateforme fournisseur, copiez le titre, téléchargez des images, modifiez la description, définissez vos marges, poussez-le vers votre boutique Shopify. Répétez. Cinquante fois. Chaque semaine. Ce n'est pas un travail difficile. C'est juste un travail fastidieux. Et le travail fastidieux est

Utilisation de Git avec des agents de codage

Agentic Engineering Patterns > Git is a key tool for working with coding agents. Keeping code in version control lets us record how that code changes over time and investigate and reverse any mistakes. All of the coding agents are fluent in using Git's features, both basic and advanced. This fluency means we can be more ambitious about how we use Git ourselves. We don't need to memorize how to do

L'automatisation des tâches Gemini est lente, maladroite et super impressionnante

I've been testing out Gemini's new task automation on the Pixel 10 Pro and the Galaxy S26 Ultra, which for the first time lets Gemini take the wheel and use apps for you. It's limited to a small subset right now - a handful of food delivery and rideshare services - and it's still in […]

Amazon fabrique un téléphone Alexa

Over 10 years after shelving the Fire Phone, Amazon is reportedly planning to launch another smartphone, this time focused on Alexa. According to Reuters, the phone, which is code-named "Transformer," will center around Amazon's AI assistant, but Alexa won't "necessarily be the primary operating system of the phone." "Transformer" is currently in development in Amazon's […]

Claude Code coûte jusqu'à 200 $ par mois. Goose fait la même chose gratuitement.

The artificial intelligence coding revolution comes with a catch: it's expensive. Claude Code , Anthropic's terminal-based AI agent that can write, debug, and deploy code autonomously, has captured the imagination of software developers worldwide. But its pricing — ranging from $20 to $200 per month depending on usage — has sparked a growing rebellion among the very programmers it aims to serve. N

Salesforce déploie un nouvel agent d'IA Slackbot alors qu'il combat Microsoft et Google dans l'IA sur le lieu de travail

Salesforce on Tuesday launched an entirely rebuilt version of Slackbot , the company's workplace assistant, transforming it from a simple notification tool into what executives describe as a fully powered AI agent capable of searching enterprise data, drafting documents, and taking action on behalf of employees. The new Slackbot, now generally available to Business+ and Enterprise+ customers, is S

Anthropic lance Cowork, un agent Claude Desktop qui fonctionne dans vos fichiers — aucun codage requis

Anthropic released Cowork on Monday, a new AI agent capability that extends the power of its wildly successful Claude Code tool to non-technical users — and according to company insiders, the team built the entire feature in approximately a week and a half, largely using Claude Code itself. The launch marks a major inflection point in the race to deliver practical AI agents to mainstream users, po

J'ai ajouté une preuve de capture d'écran à chaque appel d'outil MCP de curseur

I Added Screenshot Proof to Every Cursor MCP Tool Call I've been building MCP tools in Cursor for two months. Last week I realized: I have no idea what's actually happening. Cursor calls my tools. The tools run browser automation (fill forms, click buttons, navigate pages). Cursor gets back structured JSON. But I don't see the page. I don't know if the button click worked. I'm flying blind. So I d

Jetez un coup d'œil à cet article et laissez-nous vos commentaires ! Réfléchissons à de nouvelles idées et apprenons ensemble.

Agentic Coding: Rules, skills, subagents, and reflection—how we steer models so multi-step work stays coherent. Burak Boduroğlu ・ Mar 20 #ai #cursor #agents #llm

Les mathématiques qui tuent votre agent IA

An 85% accurate AI agent fails 4 out of 5 times on a 10-step task. Learn the compound probability math behind production failures (and the 4-check pre-deployment framework to fix it). The post The Math That’s Killing Your AI Agent appeared first on Towards Data Science .

Modes de défaillance Agentic RAG : Récupération Thrash, Tool Storms et Context Bloat (et comment les repérer tôt)

Why agentic RAG systems fail silently in production and how to detect them before your cloud bill does The post Agentic RAG Failure Modes: Retrieval Thrash, Tool Storms, and Context Bloat (and How to Spot Them Early) appeared first on Towards Data Science .

Comment les systèmes multi-agents remodèlent le développement de logiciels

Hey there! If you've been keeping up with the AI space, you already know things move fast — like, blink-and-you-miss-it fast. This week I want to talk about something that's been on my mind: how multi-agent systems are reshaping software development. Grab a coffee, let's dig in. Why This Matters Right Now It's March 2026, and the AI landscape has never looked more exciting — or more overwhelming.

Couverture d'évaluation : la métrique qui manque à vos agents d'IA

Every serious codebase measures test coverage. CI pipelines enforce minimums. Pull requests get rejected when coverage drops. The industry spent two decades making this a standard practice. For AI agents, the equivalent metric doesn't exist yet. It should. It's called eval coverage — the percentage of agent executions that receive an evaluation. The Current State: Nearly Zero The numbers are stark

Comment j'ai construit un outil Claude qui capture chaque page qu'il visite

How I Built a Claude Tool That Screenshots Every Page It Visits Claude is incredibly powerful. I use it in Claude Desktop to browse websites, fill forms, extract data, monitor pages. But Claude is text-only. When Claude clicks a button, it can't see the result. When Claude submits a form, it can't see the confirmation page. It's working blind. So I built an MCP tool that changes that. After every

Les 25 000 agents IA de McKinsey : augmentation, pas remplacement

McKinsey & Company just revealed it now operates with 60,000 total "employees" : 40,000 humans and 25,000 AI agents. CEO Bob Sternfels announced the numbers at CES Las Vegas and on Harvard Business Review's IdeaCast in January 2026. The firm added 25,000 agents in under two years. This is not a story about job cuts. It's a story about how the world's most prestigious consulting firm is remaking it

J'ai ajouté une preuve de capture d'écran à chaque appel d'outil Windsurf MCP

I Added Screenshot Proof to Every Windsurf MCP Tool Call I've been using Windsurf as my primary IDE for three months. Last week I built my first MCP tools for it. By week two, I realized the same problem I had with other tools: I have no idea what's actually happening. Windsurf calls my tools. The tools run browser automation. Windsurf gets back JSON. But I don't see the page. I don't know if the

33 automatisations OpenClaw que vous pouvez configurer en 30 minutes et qui commencent à vous rapporter de l'argent ce soir 🦞

There’s a new breed of human emerging on the internet. They don’t hustle. They don’t grind. They don’t “rise and grind” at 5 AM with a motivational quote and a green smoothie. They wake up at 10, check their phone, and discover their AI assistant already did 4 hours of work while they were drooling on their pillow. That AI assistant? A cartoon space lobster named Molty. Running on a tool that’s be

Comment ajouter une preuve visuelle à votre serveur MCP (captures d'écran + relecture)

How to Add Visual Proof to Your MCP Server (Screenshots + Replay) Your MCP tool just ran 50 actions on a website. But your user only sees JSON output. They don't know what actually happened. This is the MCP visibility problem. MCP servers (Model Context Protocol) are powerful — they let Claude, Cursor, and Windsurf automate browser tasks. But they're invisible. A user calls your tool, gets back st

Montrer HN : Passerelle Permit MCP – Autorisation à grain fin et IGA pour MCP

Afficher HN : Golf Scanner – Outil OSS pour trouver et auditer chaque serveur MCP

Visa prépare les systèmes de paiement pour les transactions initiées par un agent IA

Payments rely on a simple model: a person decides to buy something, and a bank or card network processes the transaction. That model is starting to change as Visa tests how AI agents can initiate payments. New work in the banking sector suggests that, in some cases, software agents may soon take on that role. […] The post Visa prepares payment systems for AI agent-initiated transactions appeared f

NVIDIA veut que les agents d'IA d'entreprise soient plus sûrs à déployer

The NVIDIA Agent Toolkit is Jensen Huang’s answer to the question enterprises keep asking: how do we put AI agents to work without losing control of our data and our liability? Announced at GTC 2026 in San Jose on March 16, the NVIDIA Agent Toolkit is an open-source software stack designed to help enterprises and […] The post NVIDIA wants enterprise AI agents safer to deploy appeared first on AI N

Trustpilot s'associe à des entreprises d'IA alors que la recherche traditionnelle diminue

Trustpilot is reported to be pursuing partnerships with large eCommerce companies as AI-driven shopping gains traction. In an interview with Bloomberg News [paywall], chief executive Adrian Blair said that AI agents acting on behalf of consumers require lots of information about the businesses they’re willing to interact with. He said the most effective systems will […] The post Trustpilot partner

NTT DATA et NVIDIA amènent les usines d'IA d'entreprise à l'échelle de la production

NTT DATA has announced an initiative to deliver NVIDIA-powered platforms designed to give organisations a repeatable, production-ready model for scaling AI. The offering integrates NVIDIA’s GPU-accelerated computing and high-performance networking with NVIDIA AI Enterprise software, including NeMo and NIM Microservices, into a full-stack agentic AI platform that can be deployed in cloud and edge e

OpenAI's Frontier met les agents de l'IA dans un combat que le SaaS ne peut pas se permettre de perdre

When OpenAI launched Frontier in February, the announcement was described as a platform for enterprise AI agents. What it actually signalled was a challenge to the revenue architecture underpinning the software industry. Frontier is designed to act as a semantic layer in an organisation’s existing systems, connecting data warehouses, CRM platforms, ticketing tools, and internal […] The post OpenAI

J'ai remplacé 3 heures d'ingénierie rapide par un appel API gratuit de 50 ms

The secret? Stop treating prompts like strings. Start treating them like compiled code. Last month I was building a customer support agent for a client. Standard stuff: Claude-powered, MCP-connected, deployed on their Slack workspace. The agent took me 2 hours to build. The prompt took me 3 days . Every time I thought I had it right, the agent would: Lose its persona after 4-5 messages Forget cons

Arrêtez d'écrire des invites d'agent IA comme si c'était 2023 : le cadre qui fait fonctionner votre agent OpenClaw

Your agent isn't broken. Your SOUL.md is. I've deployed dozens of AI agents. WhatsApp bots, Telegram assistants, Discord helpers — you name it. And for months, I kept hitting the same wall: The agent would work perfectly for 3-4 messages. Then it would completely lose character. Sound familiar? Maybe your agent starts responding generically. Maybe it ignores the constraints you carefully defined.

Hybrid Knowledge Retrieval : Combining Neo4j Graph Queries, GraphRAG and Vector Search for Enterprise AI Customer Service

1. Introduction: The Blind Spots of Single-Retrieval Approaches and Defining Full-Stack Capability Closure This is Part 6 of the series 8 Weeks from Zero to One: Full-Stack Engineering Practice for a Production-Grade LLM Customer Service System . In the first five parts, we completed the MVP architecture, multimodal data pipeline, GraphRAG service wrapping, multi-agent workflow design, and end-to-

Construire des garde-corps de sécurité pour le service client LLM qui fonctionnent réellement en production

1. Introduction: Production-Grade Security Risks in LLM Customer Service Systems In Part 4 of this series, we completed the multi-agent workflow architecture and embedded safety control nodes at the framework layer, implementing basic circuit breaking and permission validation. However, in enterprise production deployments, framework-layer safety nodes are only the "skeleton" — a guardrail system

Comment surveiller la dérive de l'agent IA en production

How to Monitor AI Agent Drift in Production You deploy an AI agent on a Tuesday. It works perfectly. By Thursday, something is off — the outputs are subtly different, a downstream API changed its response format, or the LLM provider silently updated their model weights. Your agent is still running. It's still returning 200s. But it's drifting . This is the problem nobody warns you about when you s

Chef d'orchestre : routage déterministe pour les flux de travail multi-agents

Show HN : J'ai construit un serveur MCP qui vous permet de gérer les fils de méta de Claude

L'écosystème MCP : comment le protocole de contexte de modèle devient le HTTP des agents d'IA

Show HN : Lula – assistant de codage multi-agents avec moteur d'exécution Rust en bac à sable

Show HN : KYC Verification Process Automation Using Agentic AI (Automatisation du processus de vérification KYC à l'aide

Show HN : Git-issues – AI agent-first task management for Git repos

Briser le bac à sable Expression de n8n en RCE (CVE-2026-27577) avec striga.ai

Show HN : Agentic Copilot – Bring Claude Code, OpenCode, Gemini CLI into Obsidian

Show HN : Trois nouveaux modèles Kitten TTS – le plus petit moins de 25 Mo

Analyse de base de code assistée par LLM pour la migration : comparaison des agents de code Codex, Claude et VS

Intro La plupart des migrations échouent avant de commencer — parce que personne ne sait réellement ce que fait le système. Les systèmes hérités échouent rarement en raison de la syntaxe ou des cadres. Ils échouent parce que leur comportement n'est pas documenté et mal compris. Ce manque de compréhension devient encore plus critique lorsque le développement est fait avec des agents. Dans cette série, j'explore comment l'outillage LLM peut aider à migrer les systèmes existants

NVIDIA GTC 2026 : mises à jour en direct sur les prochaines étapes de l'IA

Rolling coverage from San Jose, including NVIDIA CEO Jensen Huang’s keynote, news highlights, live demos and on‑the‑ground color through March 19.

La ville a des routes nommées Tape Drive et Disk Drive d'une époque révolue de fabrication de disques durs — la région abritait autrefois l'empire StorageTek

Tape Drive and Disk Drive rods in the Louisville/Broomfield area are a relic of the STK diskmaker's time in the city.

Les LLM peuvent démasquer les utilisateurs pseudonymes à grande échelle avec une précision surprenante

Pseudonymity has never been perfect for preserving privacy. Soon it may be pointless.

Les AirPods Pro 3 sont actuellement à 50 $ de rabais, ce qui correspond presque à leur meilleur prix de tous les temps

Less than a week ago, Apple announced the forthcoming AirPods Max 2, a pair of over-ear headphones that leverage the company’s H2 chip for AI-powered live translation, conversation awareness, and a host of newer features. However, if you’re okay with a pair of earbuds, the AirPods Pro 3 offer access to all the same features […]

État de l'Open Source sur Hugging Face : Printemps 2026

Au-delà de la mise en cache rapide : 5 autres choses que vous devriez mettre en cache dans les pipelines RAG

A practical guide to caching layers across the RAG pipeline, from query embeddings to full query-response reuse The post Beyond Prompt Caching: 5 More Things You Should Cache in RAG Pipelines appeared first on Towards Data Science .

Afficher HN : Brf.it - Extraction des interfaces de code pour le contexte LLM

J'ai 11 ans et j'ai formé un LLM MoE personnalisé pour $ 1

Mastercard garde un œil sur la fraude avec le nouveau modèle de base

Mastercard has developed a large tabular model (an LTM as opposed to an LLM) that’s trained on transaction data rather than text or images to help it address security and authenticity issues in digital payments. The company has trained a foundation model on billions of card transactions, with the intention of expanding to hundreds of […] The post Mastercard keeps tabs on fraud with new foundation

Pour une IA efficace, l'assurance doit mettre de l'ordre dans ses données

A report from Autorek, a provider of AI solutions to the insurance industry has produced a report that describes operational drag in companies’ internal processes that not only affect overall efficiency but cause an impediment to the effective implementation of AI in insurance concerns. Insurance Operations & Financial Transformation 2026 [email wall] draws from a […] The post For effective AI, in

Disjoncteurs pour les fournisseurs de LLM : assurer la résilience dans les applications alimentées par l'IA

🤯 LLM Circuit Breakers: Your AI App's Lifesaver! 🤯 Okay, folks, let's talk LLMs! We're building AMAZING things with these models – chatbots, content creators, the works! But what happens when OpenAI, Google, or Anthropic goes down ? 😱 Your app crashes and burns, that's what! That's where the circuit breaker pattern comes in, and it's a GAME CHANGER. Think of it like this: your app is sending reque

Disjoncteur en cas de défaillance du fournisseur LLM

Stop calling a dead API. Shed load fast, recover automatically, and stay consistent across restarts with Redis-backed failure state. Why this matters Every LLM-powered application depends on an external provider - OpenAI, Anthropic, Google, or a self-hosted model. These providers go down. Rate limits spike. Latency balloons. Without a circuit breaker, your application keeps sending requests into a

Exécutez 397B LLM sur votre ordinateur portable : le guide No-BS

Liquid syntax error: Unknown tag 'endraw'

OpenClaw 인기 모델 Top 5 — 2026년 3월 비용·성능 완벽 비교

OpenClaw 인기 모델 Top 5 — 2026년 3월 비용·성능 완벽 비교 GitHub Stars 247,000을 돌파한 OpenClaw 에서 어떤 AI 모델을 선택해야 할까요? 커뮤니티에서 가장 많이 사용하는 모델부터, 비용 효율적인 하이브리드 전략까지 실제 데이터 기반으로 정리했습니다. 개발자라면 5분 안에 모델 선택의 답을 찾을 수 있습니다. 왜 OpenClaw는 "다중 모델" 플랫폼인가? OpenClaw의 가장 큰 장점은 특정 모델에 종속되지 않는다는 것입니다. Anthropic, OpenAI, Google, Mistral—어떤 제공자든 API 키만 있으면 연결할 수 있습니다. 이 유연성 덕분에 단순한 자동화 작업은 저가 모델로, 민감한 작업은 고성능 모델로 처리하는 동적 라우팅 이 가능합니다.

Afficher HN : jj-benchmark – Évaluation des agents d'IA sur le contrôle de version du Jujutsu

Show HN : Comment j'ai dominé le classement LLM ouvert HuggingFace sur deux GPU de jeu

Afficher HN : TMA1 – Observabilité locale pour les agents LLM

Utilisation de LLM et Ghidra locaux pour analyser les logiciels malveillants

MacinAI Local : Création d'un moteur d'inférence LLM indépendant du modèle pour Mac OS 9

Afficher HN : Une seule CLI pour gérer les modèles llama.cpp/vLLM/Ollama

Show HN : Llamactl – Gestionnaire LLM auto-hébergé avec routage compatible OpenAI

Show HN : Qwodel - Un pipeline unifié open source pour la quantification LLM

Afficher HN : 1k ordinateur de bureau bat la bibliothèque clairsemée du fournisseur 474× sur Mistral-7B

AI Search vs Google Comment le référencement évolue en 2026

AI Search vs Google: How SEO is changing in 2026 Search is evolving faster than ever. For years, Search Engine Optimization (SEO) was largely about ranking web pages on Google through keywords, backlinks, and technical improvements. But in 2026, search is undergoing a major shift. Artificial Intelligence is transforming how people search, how answers are delivered, and how businesses get discovere

Mouvements fluides : la réalité virtuelle à 90 images par seconde arrive MAINTENANT sur GeForce

It’s a double feature on GFN Thursday. This week, GeForce NOW offers smoother sights in virtual reality (VR) and a sprawling new land to conquer. Streaming at 90 frames per second (fps) comes to supported VR headsets. And Crimson Desert, which recently surpassed 3 million wishlist additions on Steam, debuts in the cloud with GeForce […]

De la simulation à la production : comment construire des robots avec l'IA

The latest open models and frameworks from NVIDIA bring together simulation, robot learning and embedded compute to accelerate cloud-to-robot workflows.

More Than Meets the Eye : les ordinateurs accélérés NVIDIA RTX se connectent désormais directement à Apple Vision Pro

NVIDIA and Apple’s collaboration brings native integration of NVIDIA CloudXR 6.0 to visionOS, securely delivering NVIDIA RTX-powered simulators and professional 3D graphics applications — like Immersive for Autodesk VRED on Innoactive’s XR streaming solutions — to Apple Vision Pro.

NVIDIA, les leaders des télécommunications construisent des réseaux d'IA pour optimiser l'inférence sur les réseaux distribués

As AI‑native applications scale to more users, agents and devices, the telecommunications network is becoming the next frontier for distributing AI. At NVIDIA GTC 2026, leading operators in the U.S. and Asia showed that this shift is underway, announcing AI grids — geographically distributed and interconnected AI infrastructure — using their network footprint to power […]

Achetez le Sapphire Nitro+ Radeon RX 9070 XT pour 799,99 $ — livré avec un bloc d'alimentation de 1000 W et une copie gratuite de Crimson Desert

AMD's Radeon RX 9070 XT based on the RDNA 4 architecture is currently the company's most powerful graphics card offering impressive 1440p and 4K gaming performance.

Intel a lancé sa première puce Pentium il y a 33 ans, livrée avec 3,1 millions de transistors — puce x86 de cinquième génération construite sur un processus de 800 nm

33 years ago today, the Pentium lineage began, shaping decades of personal computing to come.

OpenAI contourne Nvidia avec un modèle de codage exceptionnellement rapide sur des puces de la taille d'une plaque

OpenAI's new GPT‑5.3‑Codex‑Spark is 15 times faster at coding than its predecessor.

Musk dit qu'il construit une usine de copeaux de Terafab à Austin, au Texas

Elon Musk announced plans to build a Terafab plant in Austin, Texas, that will be jointly run by Tesla and SpaceX. The goal is to eventually build chips at scale for robotics, artificial intelligence, and space-based data centers for Musk's various companies. Musk has expressed concern, as have other executives, about the chip industry's ability […]

Le chemin de fer obtient 100 millions de $ pour défier AWS avec une infrastructure cloud native à l'IA

Railway , a San Francisco-based cloud platform that has quietly amassed two million developers without spending a dollar on marketing, announced Thursday that it raised $100 million in a Series B funding round, as surging demand for artificial intelligence applications exposes the limitations of legacy cloud infrastructure. TQ Ventures led the round, with participation from FPV Ventures , Redpoint

Ce qui s'est passé chez Nvidia GTC : NemoClaw, Robot Olaf et un pari de 1 000 milliards de $

CEO Jensen Huang took the stage at Nvidia’s GTC conference this week in his signature leather jacket to deliver a two-and-a-half-hour keynote, projecting $1 trillion in AI chip sales through 2027, declaring that every company needs an “OpenClaw strategy,” and closing with a rambling Olaf robot that had to get its mic cut. The message was hard to miss: Nvidia […]

Nvidia a une stratégie OpenClaw. Et vous ?

CEO Jensen Huang took the stage at Nvidia’s GTC conference this week in his signature leather jacket to deliver a two-and-a-half-hour keynote, projecting $1 trillion in AI chip sales through 2027, declaring that every company needs an “OpenClaw strategy,” and closing with a rambling Olaf robot that had to get its mic cut. The message was hard to miss: Nvidia […]

K2 va lancer son premier satellite haute puissance pour le calcul spatial

K2's Gravitas is an ambitious project that aims to demonstrate the tech needed to build data centers in space.

Do You Want To Build A Snowman?

On the latest episode of the Equity podcast, we recapped CEO Jensen Huang’s GTC keynote and debated what it means for Nvidia’s future.

L'éditeur tire le roman d'horreur « Shy Girl » sur les préoccupations de l'IA

Hachette Book Group said it will not be publishing “Shy Girl” over concerns that artificial intelligence was used to generate the text.

Pourquoi Wall Street n'a pas été conquis par la grande conférence de Nvidia

Despite investor fears of an AI bubble, Nvidia's latest conference shows that most in the industry aren't concerned by that possibility.

Afficher HN : ClawMem - Mémoire d'agent open source avec récupération de GPU local SOTA

Goldman Sachs voit l'investissement dans l'IA se déplacer vers les centres de données

Artificial intelligence investment is entering a more selective phase as companies and investors look beyond early excitement and focus on the data centre infrastructure required to run AI systems. Recent analysis from Goldman Sachs suggests the market is moving toward what the firm describes as a “flight to quality.” In practice, investors are paying closer […] The post Goldman Sachs sees AI inve

E.SUN Bank et IBM construisent un cadre de gouvernance de l'IA pour les banques

E.SUN Bank is working with IBM to build clearer AI governance rules for how artificial intelligence can be used inside a bank. The effort reflects a wider shift in finance. Many firms already use AI for fraud checks and credit scoring, and some also use it to handle customer service queries. The new challenge is […] The post E.SUN Bank and IBM build AI governance framework for banking appeared fir

Show HN : Sokuji - Traducteur vocal open source avec AI WASM/WebGPU sur l'appareil

Afficher HN : VMetal – exécuter un nuage GPU sur du métal nu sans OpenStack

Les racks Nvidia Groq 3 LPU et Groq LPX rejoignent la plateforme Rubin – SRAM-emballés

Nvidia lance le processeur Vera, spécialement conçu pour l'IA agentique

Le pilote « GreenBoost » open source vise à augmenter la VRAM des GPU Nvidia avec la RAM système

Audit de performance des articles PCGamer

Research: PCGamer Article Performance Audit Stuart Breckenridge pointed out that PC Gamer Recommends RSS Readers in a 37MB Article That Just Keeps Downloading , highlighting a truly horrifying example of web bloat that added up to 100s more MBs thanks to auto-playing video ads. I decided to have Claude Code for web use Rodney to investigate the page - prompt here . Tags: web-performance , rodney

Elon Musk dévoile un projet de puces « TeraFab » de 20 milliards de $ pour fabriquer des puces, de la mémoire et des processeurs de conditionnement sous un même toit — cible un térawatt de calcul annuel

Tesla and SpaceX CEO, Elon Musk, announced Saturday night that his TeraFab semiconductor project will be built on the Tesla campus in eastern Travis County, Austin, Texas, as a joint venture.

La révision de la Nintendo Switch 2 pourrait apporter une batterie amovible — la nouvelle révision vise à se conformer aux réglementations de l'UE sur le droit à la réparation

Nintendo is reportedly set to launch a revised Switch 2 with a removable battery in Europe, in line with regulations requiring easier battery replacement by February 2027.

La saison des prix des influenceurs IA est à nos portes

First came the AI beauty pageant. Then the AI music contests. Now, there is an award for AI Personality of the Year - perhaps the inevitable next step for the AI influencer economy as it transforms from quirky novelty into a serious and lucrative industry. The contest, a joint venture between generative AI studio OpenArt […]

La génération AI Kool-Aid a le goût de l'eugénisme

Like many people, director Valerie Veatch was intrigued when OpenAI first released its Sora text-to-video generative AI model to the public in 2024. Though she didn't fully understand the technology, she was curious about what it could do, and she saw that other artists were building online communities to share their new AI creations. The […]

Listen Labs lève 69 MILLIONS de $ après une cascade d'embauche de panneaux d'affichage viraux pour adapter les entretiens avec les clients de l'IA

Alfred Wahlforss was running out of options. His startup, Listen Labs , needed to hire over 100 engineers, but competing against Mark Zuckerberg's $100 million offers seemed impossible. So he spent $5,000 — a fifth of his marketing budget — on a billboard in San Francisco displaying what looked like gibberish: five strings of random numbers. The numbers were actually AI tokens. Decoded, they led t

Cursor admet que son nouveau modèle de codage a été construit sur le Kimi de Moonshot AI

Building on top of a Chinese model feels particularly fraught right now.

Les startups d'IA mangent l'industrie du capital-risque et les rendements, jusqu'à présent, sont bons

AI startups accounted for 41% of the $128 billion in venture dollars raised by companies on Carta last year — a record-high annual share.

Bluesky annonce une série B de 100 MILLIONS de $ après la transition du PDG

The additional funds have been used to scale Bluesky's team, while the company continues to develop Bluesky's app and the underlying ATProto that powers it.

TechCrunch Startup Battlefield 200 nominations sont toujours ouvertes

Nominate your startup, or one you know, for TechCrunch Startup Battlefield 200 before May 27. Chance to win $100,000 equity-free funding and VC access.

Alphabet's X a un nouveau spin-out, et il s'attaque à l'un des cauchemars bureaucratiques les plus chers du monde

Anori aims to shrink the pre-development process by getting all parties, including the city, on a unified platform from the start so that compliance conflicts are surfaced in weeks instead of months or years.

Une visite exclusive du laboratoire Trainium d'Amazon, la puce qui a conquis Anthropic, OpenAI, voire Apple

Shortly after Amazon announced its $50 billion investment in OpenAI, AWS invited me on a private tour of the chip lab at the heart of the deal.

Les jetons d'IA sont-ils le nouveau bonus de signature ou juste un coût de faire des affaires ?

Maybe tokens really will become the fourth pillar of engineering compensation. But engineers might want to hold the line before embracing this as a straightforward win.

Un nouveau dossier judiciaire révèle que le Pentagone a déclaré à Anthropic que les deux parties étaient presque alignées — une semaine après que Trump ait déclaré la relation kaput

Anthropic submitted two sworn declarations to a California federal court late Friday afternoon, pushing back on the Pentagon's assertion that the AI company poses an "unacceptable risk to national security" and arguing that the government's case relies on technical misunderstandings and claims that were never actually raised during the months of negotiations.

Nouveautés de Mellea 0.4.0 + Granite Libraries Release

Parallélisme de séquence d'Ulysse : formation avec des contextes à millions de jetons

Show HN : Soros – AI pour l'investissement macro géopolitique

P2P Crypto Exchange Development : A Complete Guide to Building a Secure and Scalable Trading Platform

Le marché de la crypto-monnaie continue de se développer à un rythme rapide, créant de nouvelles opportunités commerciales pour les entrepreneurs et les startups. L'un des moyens les plus efficaces d'entrer dans cet espace consiste à lancer un échange cryptographique P2P. Ce modèle permet aux utilisateurs d'échanger directement entre eux, offrant un meilleur contrôle, une meilleure transparence et une meilleure flexibilité par rapport aux systèmes traditionnels. Pour les propriétaires d'entreprise, construire un

Expérimenter Starlette 1.0 avec les compétences de Claude

Starlette 1.0 is out ! This is a really big deal. I think Starlette may be the Python framework with the most usage compared to its relatively low brand recognition because Starlette is the foundation of FastAPI , which has attracted a huge amount of buzz that seems to have overshadowed Starlette itself. Kim Christie started working on Starlette in 2018 and it quickly became my favorite out of the

Profilage des utilisateurs de Hacker News en fonction de leurs commentaires

Here's a mildly dystopian prompt I've been experimenting with recently: "Profile this user", accompanied by a copy of their last 1,000 comments on Hacker News. Obtaining those comments is easy. The Algolia Hacker News API supports listing comments sorted by date that have a specific tag, and the author of a comment is tagged there as author_username . Here's a JSON feed of my ( simonw ) most recen

Turbo Pascal 3,02A, déconstruit

Turbo Pascal 3.02A, deconstructed In Things That Turbo Pascal is Smaller Than James Hague lists things (from 2011) that are larger in size than Borland's 1985 Turbo Pascal 3.02 executable - a 39,731 byte file that somehow included a full text editor IDE and Pascal compiler. This inspired me to track down a copy of that executable (available as freeware since 2000) and see if Claude could interpret

Trump incite les entreprises de centres de données à s'engager à payer pour la production d'électricité

With no enforcement and questionable economics, it may not make a difference.

La nouvelle attaque AirSnitch contourne le cryptage Wi-Fi dans les maisons, les bureaux et les entreprises

That guest network you set up for your neighbors may not be as secure as you think.

Rétraction : après un rejet de code de routine, un agent d'IA a publié un hit sur quelqu'un par son nom

This story has been retracted

Delve accusé d'avoir induit les clients en erreur avec une « fausse conformité »

An anonymous Substack post accuses compliance startup Delve of “falsely” convincing “hundreds of customers they were compliant” with privacy and security regulations.

Outils permettant aux fondateurs de naviguer et de surmonter les conflits

Founders can create frameworks for working through conflict and change when the team is just two or three people, and if done correctly, it can scale with the company.

Elon Musk dévoile des plans de fabrication de puces pour SpaceX et Tesla

Elon Musk recently outlined ambitious plans for a chip-building collaboration Tesla and SpaceX — but he has a history of overpromising.

Microsoft annule une partie de son bloc d'IA Copilot sur Windows

The company is reducing Copilot entry points on Windows, starting with Photos, Widgets, Notepad, and other apps.

Silicon Friendly

Tâches planifiées du code Claude

Banc pour Claude Code

Context.dev

Fractal

Novi Notes

murmur

Replit Agent 4

Design Agent par Lokuma

Curseur en verre

Construire un modèle d'intégration spécifique au domaine en moins d'une journée

Holotron-12B - Agent d'utilisation d'ordinateur à haut débit

Présentation des seaux de rangement sur le moyeu de face d'étreinte

Keep the Tokens Flowing : Lessons from 16 Open-Source RL Libraries

LeRobot v0.5.0 : Mise à l'échelle de chaque dimension

Apporter l'IA robotique aux plates-formes intégrées : enregistrement d'ensembles de données, réglage fin VLA et optimisations surl'appareil

Présentation des diffuseurs modulaires - Blocs de construction composites pour les canalisations de diffusion

Curseur Sauts de saisie semi-automatique avec Claude Code : 3 Fixes

Cursor's Tab Autocomplete Stops Working After Claude Code Install Cursor's tab autocomplete goes silent the moment you install Claude Code in the same VS Code instance. No error messages. Just dead silence when you hit tab. This isn't a bug in either tool — it's a keybinding collision. Both Cursor and Claude Code bind to Tab for their inline suggestion systems. Cursor's autocomplete loses because

J'ai construit un bot qui donne mes jetons Cursor Composer inutilisés à open source

I use Cursor every day. Sonnet, the frontier models — constantly. But Composer and Auto? I'd open them, poke around, and close them. Those tokens just expire every month doing nothing. So I built Iynx. What it does Searches GitHub for trending repos with open issues Picks an issue it thinks it can actually fix Writes the code, runs tests inside Docker Opens a real PR automatically Why I built it A

Le compositeur 2 est controversé, mais mon expérience réelle était solide

I tried Composer 2 properly today, and honestly, if you put all the controversy aside for a second, the model itself is not bad at all. In fact, my first impression is that it’s a real upgrade over Composer 1 and 1.5. I gave it a pretty solid test. I asked it to build a full-stack Reddit clone and deploy it too. On the first go, it handled most of the work surprisingly well. The deployment also wo

Mise en cache rapide avec l'API OpenAI : un tutoriel complet sur Python

A step-by-step guide to making your OpenAI apps faster, cheaper, and more efficient The post Prompt Caching with the OpenAI API: A Full Hands-On Python tutorial appeared first on Towards Data Science .

Construire un solveur Navier-Stokes en Python à partir de zéro : Simulation du flux d'air

A hands-on guide to implementing CFD with NumPy, from discretization to airflow simulation around a bird's wing The post Building a Navier-Stokes Solver in Python from Scratch: Simulating Airflow appeared first on Towards Data Science .

Échapper à la jungle SQL

Most data platforms don’t break overnight; they grow into complexity, query by query. Over time, business logic spreads across SQL scripts, dashboards, and scheduled jobs until the system becomes a “SQL jungle.” This article explores how that happens and how to bring structure back. The post Escaping the SQL Jungle appeared first on Towards Data Science .

Une introduction en douceur à l'optimisation contrainte non linéaire avec des approximations linéaires par morceaux

Piecewise linear approximations are a practical way to handle nonlinear constrained models using LP/MIP solvers like Gurobi. The post A Gentle Introduction to Nonlinear Constrained Optimization with Piecewise Linear Approximations appeared first on Towards Data Science .

Construire des modèles de notation de crédit robustes (partie 3)

Handling outliers and missing values in borrower data using Python. The post Building Robust Credit Scoring Models (Part 3) appeared first on Towards Data Science .

Comment mesurer la valeur de l'IA

While efficiency is an important source of AI value, it is only part of the picture The post How to Measure AI Value appeared first on Towards Data Science .

Les bases de l'ingénierie vibrante

Building products without the coding part The post The Basics of Vibe Engineering appeared first on Towards Data Science .

Santa Augmentcode Intent Ep.1

Santa’s Secret Weapon: Welcome to the Workshop! 🎅 — Augment Intent, Episode 1 Ho ho ho! Come in, come in — the fire is warm and the cocoa is hot. Pull up a stool and let Father Christmas tell you a story. Not about reindeer, not about presents — but about the most magical piece of software to land in the Workshop since the invention of the Nice List. The Problem With Being Father Christmas Every y

ScreenAI : Un modèle de langage visuel pour l'interface utilisateur et la compréhension du langage visuellement situé

Cappy : Surperformer et booster les grands modèles de langage multi-tâches avec un petit score

Parlez comme un graphique : Encodage des graphiques pour les grands modèles de langage

Outils d'intégration spécifiques à la santé pour la dermatologie et la pathologie

Apprentissage social : apprentissage collaboratif avec de grands modèles linguistiques

Combien de temps jusqu'à ce que chaque fournisseur majeur mette ses boucles RSI en mouvement ?

Codage agentique pour codeurs non vibrants

J'ai construit et expédié une application iOS à partir de mon téléphone tout en voyageant au Japon

Démarrer (Vibe) le codage rapidement

Conseils et meilleures pratiques pour travailler avec des agents de codage IA

Show HN : Neurotrace – L'extension que j'ai construite mais que je n'ai jamais utilisée

Show HN : Code Altimate – Faisceau d'ingénierie de données authentiques open source

Show HN : J'en ai eu marre de l'impression(x.shape) alors j'ai créé des astuces de type d'exécution pour Python

L'agent AI Backdoors Trivy Security Scanner, arme une extension de code VS

Afficher HN : SHTMLs – HTML pastebin où l'IA télécharge sa propre sortie

Liste des règles pour le curseur

Règles et invites du curseur

Show HN : Drift-guard - Protégez votre interface utilisateur de la dérive de conception des agents IA

Show HN : Plugin simple pour que Claude Code vous écoute

Show HN : Et si votre synthétiseur était alimenté par APL (ou un clone K muet) ?

Show HN : HawkDoc - éditeur de style Notion open-source basé sur Lexical

Monocode

Afficher HN : BitFun – Un environnement de développement agentique (Rust et TypeScript)

Show HN : AI models AI capex

Nouvelle source ouverte de constructeur non traditionnel

Demandez à HN :35,0 arrière-plan CS, a construit de vraies applications avec l'IA, a besoin de suggestion

La qualité de votre modèle est liée à un modèle commercial

Show HN : VibeWhisper – voix-texte macOS avec push-to-talk,cloud ou 100% local

Demandez à HN : Y a-t-il des utilisateurs expérimentés de l'IA /de l'agent ? Avez-vous des conseils ?

Afficher HN : agent IA qui recherche dans le forum Curseur

Nouvelle version Open Source

Show HN : Budibase Agents Beta – agents d'IA indépendants du modèle pour les flux de travail internes

Le Trésor américain publie un guide sur les risques liés à l'IA à l'intention des institutions financières

The US Treasury has published several documents designed for the US financial services sector that suggest a structured approach to managing AI risks in operations and policy (see subheading ‘Resources and Downloads’ towards the bottom of the link). The CRI Financial Services AI Risk Management Framework (FS AI RMF) comes with a Guidebook [.docx] which […] The post US Treasury publishes AI risk Gu

Claude 4.6 Opus peut réciter la liste de Linux.h

Show HN : un plugin DAW open source basé sur JUCE, React et Lyria RealTime

Demandez à HN : Des agents de codage/extension Vim/Nvim-first ?

Show HN : Wolf Defender, un modèle de détection d'injection rapide à poids ouvert

Show HN : Banc de CHIFFON JURIDIQUE

Afficher HN : Flock v0.7.0 – Couche sémantique Open Source pour DuckDB (C++)

Show HN : Ava – AI Voice Agent for Traditional Phone Systems(Python+Asterisk/ARI)

Show HN : Linggen - Système d'agent ouvert dans Rust, tout modèle, basé sur des fichiers

Lancement HN : Spine Swarm (YC S23) – Agents d'IA qui collaborent sur un canevas visuel

Afficher HN : SafeParse - validation du schéma et nouvelles tentatives pour les pipelines d'IA

Agent Pro – Automatisez votre bureau à partir de votre téléphone (aucune configuration)

Show HN : Comment j'ai construit un éditeur de CV en utilisant l'IA sans expérience de développement Web

Show HN : Vessel Browser – Un navigateur open-source conçu pour les agents IA, pas pour les humains

Show HN : AI SDLC Scaffold, repo template for AI-assisted software development

Show HN : Sulcus Reactive AI Memory

J'ai construit LiveAuth : Pow et Lightning Network Authentication pour les agents IA

Afficher HN : Sandbox-policy-builder, assistant pour le courtage d'informations d'identification Vercel Sandbox

Interface utilisateur locale pour la gestion des agents de codage AI parallèles

Comment nous surveillons les agents de codage internes en cas de désalignement

Show HN : Revise – Un éditeur d'IA pour les documents

Show HN : Projection Astrale Robot

Afficher HN : Mon vol aura-t-il Starlink ?

Lancement HN : Kita (YC W26) – Automatiser l'examen du crédit dans les marchés émergents

Show HN : Nanostat - analyse automatisée des particules pour les images de microscopie électronique

Je passe les 6 derniers mois à apprendre comment automatiser mes tâches ennuyeuses avec

Show HN : J'ai construit mon premier SaaS

Super cofondateur de Micro Charged in Plot pour envoyer AI Tech en Chine

Show HN : Wix abandonné pour un agent AI Edge, donc je n'ai jamais à embaucher de personnel junior

Demandez à HN : Gains de productivité de l'IA – tirez-vous des devs ou construisez-vous de meilleurs produits ?

Afficher HN : Agent Package Manager (APM) pour la configuration de l'agent

Show HN : J'ai créé un rappel par e-mail super simple inspiré du bot RemindMe !