Le meilleur de l'écosystème IA et MCP, sélectionné chaque jour.
Sources
GGML and llama.cpp — the foundational libraries behind local LLM inference — are joining Hugging Face to ensure their long-term development and maintenance. This brings the most widely-used local inference stack under HF's open-source umbrella, with commitments to continued compatibility and community governance. A significant consolidation in the local AI ecosystem.
IBM Research and UC Berkeley released IT-Bench and MAST — a benchmark suite and failure taxonomy for enterprise AI agents in real IT environments. The research identifies the root causes of enterprise agent failures across categories like tool misuse, context loss, and ambiguity handling. Practical reference for teams debugging why their production agents underperform in enterprise settings.
Hugging Face demonstrates using Codex and Claude agent skills to generate custom CUDA kernels — making GPU-level optimisation accessible to developers without deep CUDA expertise. The post shows how agent skills can encode specialist knowledge (GPU programming patterns) and apply it on demand. A compelling use case for agent skills beyond typical software engineering tasks.
OpenEnv is a new evaluation framework that tests tool-using agents in real-world execution environments rather than sandboxed simulations — catching failure modes that controlled benchmarks miss. The post covers the framework design, evaluation methodology, and results across several agent models. Useful for any team building agents that interact with real APIs or filesystems.
OpenAI shares practical patterns for building agents that do real sustained work — combining agent skills, hosted shell execution, and server-side context compaction in the Responses API. Covers how to keep long-running agents from hitting context limits and how compaction lets them work across sessions. Useful for anyone building beyond simple single-turn agents.
Transformers.js v4 is now available on NPM, bringing Hugging Face's in-browser and Node.js inference library to a major new release with improved model support, better performance, and a cleaner API. Developers can now run transformer models directly in JavaScript without a Python backend. A big upgrade for web developers building AI-powered browser applications.
Anthropic engineers built a working C compiler by running multiple Claude instances in parallel, each responsible for different compiler stages, with a coordinator orchestrating the workflow. The experiment demonstrates real multi-agent software engineering at a non-trivial scale and surfaces practical lessons about task decomposition, inter-agent communication, and error recovery. A compelling proof-of-concept for parallel agent coding systems.
Hugging Face launched Community Evals — an open framework for community-run model evaluations that anyone can contribute to, challenging the opacity of existing leaderboards. The goal is to build a more trustworthy, diverse set of model benchmarks driven by real use cases rather than optimised for leaderboard gaming. Important infrastructure for the open-source AI evaluation ecosystem.
OpenAI distills 15 hard-won lessons from building ChatGPT Apps — covering UX patterns, capability design, handling ambiguity, and how they packaged the learnings into a Codex Skill to bootstrap new apps 10x faster. Practical and opinionated guide for anyone building on the ChatGPT Apps SDK.
Hugging Face reflects on one year since DeepSeek disrupted the AI landscape, examining how open-source AI has evolved in response — from the proliferation of capable open-weight models to new training approaches and the emerging AI+ ecosystem. The third in a series, focusing on where the global open-source AI movement goes next.
LinkedIn's AI team shares a practical retrospective on training GPT-OSS with agentic reinforcement learning — covering the infrastructure challenges, reward design decisions, and lessons learned from scaling RL training to production-grade agent models. One of the more candid engineering accounts of what it actually takes to do agentic RL at scale.
MCP Apps launched as the first official MCP extension, enabling tools to return interactive UI components — dashboards, forms, visualisations, multi-step workflows — that render directly in the conversation. Built as an open standard in collaboration with OpenAI and the MCP-UI community. A landmark moment for the MCP ecosystem that expands what MCP servers can deliver beyond plain text.
The MCP Core Maintainer team announces changes for 2026: Inna Harper and Basil Hosmer are stepping back, with new maintainers joining to keep the project moving. Also covers what's ahead for MCP governance as the protocol moves under the Agentic AI Foundation. Good context for anyone tracking MCP's organisational trajectory.
OpenAI explains how to build a systematic eval harness for agent skills — turning skills into testable, scoreable artefacts you can improve over time. The post covers designing eval cases, running them via the Codex evals framework, and iterating on skill quality based on results. A must-read for anyone shipping agent skills to production.
Anthropic shares what they learned from three iterations of a performance engineering take-home assessment that Claude kept solving correctly — forcing them to redesign it to remain useful for hiring. The post is a practical guide to crafting technical evals that test genuine human reasoning rather than tasks AI can trivially handle. Useful for any team still using take-home coding challenges.
Hugging Face marks one year since DeepSeek-R1 shocked the AI world by matching frontier performance at a fraction of the cost, fundamentally changing assumptions about what open-source models can achieve. The post reflects on how that moment catalysed a wave of open-weight model development and reshaped the competitive landscape between open and closed AI.
Hugging Face launched Open Responses — an open-source implementation of the Responses API spec, letting developers use the same agent-friendly API pattern against any open-weight model instead of being locked to OpenAI's hosted version. Compatible with OpenAI's SDK, making migration straightforward. Important infrastructure for the open-source agent ecosystem.
Skyscanner integrated Codex CLI with JetBrains IDEs via MCP, enabling in-IDE AI assistance for debugging, test generation, and development workflows without switching to a separate tool. The post details their MCP server setup, the JetBrains plugin integration, and productivity gains observed across the engineering team. A solid real-world MCP adoption case study.
A practical breakdown of how to design, run, and interpret evaluations specifically for agentic systems — covering trajectory evals, outcome evals, LLM-as-judge approaches, and the unique challenges agents pose vs. static model benchmarks. Anthropic draws on their experience evaluating Claude Code and research agents. One of the most useful practical guides on agent evals published to date.
OpenAI's year-end retrospective for developers covering the biggest model, API, and platform milestones of 2025 — from the Responses API launch to Codex, GPT-5.4, Realtime API, and the shift toward production-grade agentic systems. A useful reference for understanding where the platform stands heading into 2026.
