Actualités IA Quotidiennes

Kilo Code v7 is a complete rebuild of the popular VS Code coding assistant. The new version runs multiple AI agents at the same time, includes a built-in code reviewer that checks your changes as you go, and lets you compare outputs from different AI models side by side.

It now runs on OpenCode server under the hood — an open-source base that developers in the community can inspect and build on. You can run it with Claude, GPT-5, or other models.

For developers who want more control over their AI coding setup without paying for a closed SaaS product, this is one of the most capable free options available today.

Shadow 2.0 is an AI tool that listens to your video calls and completes the tasks your meeting creates — while the meeting is still happening. It can generate PDFs, update your CRM, write follow-up emails, and schedule next steps, all in real time.

The goal is simple: you should be able to finish a call and have nothing left on your to-do list from that call. Shadow handles it in the background while you stay focused on the conversation.

It was widely discussed on Product Hunt today, where users praised how well it understood context without needing to manually review transcripts.

Superset 2.0, built by Y Combinator president Garry Tan, has been rewritten from scratch. The new version lets you run hundreds of AI coding agents at the same time — and spread them across multiple machines. Team members can work together in real time on the same agent session.

It works with any command-line AI tool — Claude Code, Codex, Gemini CLI, or others. The idea is that the bottleneck on AI coding is no longer the AI itself, but how many parallel streams of work you can manage at once.

This is a hot topic among developers who run complex software projects and want to run many tasks at the same time without waiting for each one to finish.

The trial between Elon Musk and OpenAI entered its second week — and things got stranger. Musk took the stand last week claiming that OpenAI's Sam Altman and Greg Brockman had deceived him into giving $38 million to the company when it was still a nonprofit. He says they promised to keep it that way.

This week, OpenAI fought back. Then Shivon Zilis, a close associate of Musk, testified that Musk had actually tried to recruit Sam Altman to xAI before filing his lawsuit. That detail has raised serious questions about whether the lawsuit was really about principle — or about losing a talented competitor.

This case could reshape how AI companies structure themselves, and who gets to hold them accountable.

A New York Times investigation found that Meta's aggressive push to embed AI into every part of its business is making many of its own employees unhappy. Workers describe being required to use AI tools in meetings, performance reviews, and daily tasks — even when the tools make the work worse.

Some employees say the pressure feels performative: using AI to look productive rather than to actually be productive. Others worry that constant AI measurement is being used to justify future layoffs.

Meta is one of the most powerful tech companies on Earth. If this is how AI adoption goes at the top, it says a lot about what average workers will experience as every company races to appear AI-first.

Cloudflare — a major internet infrastructure company — announced its first large-scale layoff. CEO Matthew Prince said that AI tools had made 1,100 support roles unnecessary. Revenue hit a record high at the same time.

This is the pattern many economists warned about: productivity gains from AI do not automatically create new jobs to replace the ones removed. The company is doing better than ever. The workers who used to do those jobs are not.

This story is generating significant buzz across the tech industry because Cloudflare has been seen as a well-run, employee-friendly company. If this can happen there, it can happen anywhere.

Antirez — the creator of Redis, one of the most widely used databases in the world — just released ds4, a local engine for running DeepSeek 4 Flash directly on Apple Silicon Macs using Metal. No internet connection. No API key. No monthly bill.

The project has already collected over 5,300 stars on GitHub and is trending heavily in the AI community. It is written in C, which means it is small, fast, and runs with almost no overhead.

DeepSeek 4 Flash is a strong model for coding and reasoning tasks. Being able to run it locally on a MacBook changes the equation for developers who care about privacy or just want instant responses without rate limits.

NousResearch released Hermes Agent, a Python-based AI agent that gets better the longer you use it. It learns from your feedback, remembers how you like to work, and adjusts its behavior over time. It also supports tool use and can call external APIs.

The repo has crossed 141,000 stars with more than 11,000 new stars added in the past week — making it one of the fastest-growing AI projects on GitHub right now.

NousResearch is known in the open-source AI community for building high-quality models and tools that rival commercial products. Hermes Agent is their take on what a personal AI assistant should look like: one that grows with you rather than staying static.

Open Design is a local-first, open-source alternative to Anthropic's Claude Design tool. It ships with 19 built-in skills and 71 design systems. From one interface, you can generate web pages, mobile screens, desktop apps, slide decks, images, and videos.

It works with Claude Code, Codex, Cursor, Gemini, and several other AI coding tools. Everything runs on your machine. No subscription. No data sent to a third party.

The project gained nearly 18,500 stars in the last seven days alone — one of the fastest week-over-week growths in this data set. The community is clearly hungry for design tools that work without locking you into one vendor.