Actualités IA Quotidiennes

Memoket Gem is a small wearable device that records and remembers your conversations throughout the day. Unlike a phone app you have to remember to open, it runs quietly in the background — capturing what was said, who said it, and when.

The goal is to act as a second brain for busy people. If you had a meeting in the morning and can't remember a detail from it six hours later, Memoket can pull it back up. It does not just record — it understands what was discussed and makes that information searchable.

Wearable AI is a growing category, and Memoket is one of the more practical entries. No camera, no screen — just a small device that keeps track of the things your brain lets go.

Latitude for Claude Code is a developer tool that shows you exactly where your Claude Code sessions are spending tokens. Which files are being read repeatedly, which tool calls are happening most often, and where you are hitting context limits — all visible in one dashboard.

For developers working on large projects, this kind of observability turns a black box into something you can actually optimize. If a particular file is being re-read on every task, you can fix that. If a tool call pattern is wasteful, you can see it.

It launched on Product Hunt today with strong interest from developers who have noticed their Claude Code costs climbing but have had no way to understand why.

A judge has ruled against Elon Musk in his case against Sam Altman and OpenAI. Musk sued the company he helped found, arguing that it broke its original promise to stay a nonprofit focused on the public good. He claimed OpenAI became a profit-driven business and left its founding mission behind.

The court did not agree. The ruling is a clear win for Altman and OpenAI at a time when the company is raising huge amounts of money and expanding fast. Musk, who runs his own AI company called xAI, now walks away from the lawsuit empty-handed.

This does not end the rivalry between the two. But it does remove a legal cloud that has been hanging over OpenAI for months. The company can now move forward without this particular fight on its hands.

Anduril, a defense technology company, has revealed new details about a military headset it is building together with Meta. The device uses augmented reality — digital images layered on top of what you see in the real world — to help soldiers on the ground. One of the most striking features: a soldier can order a drone strike using only eye tracking. No hands, no radio call.

The project blends Meta's hardware experience with Anduril's defense expertise. It is a sign of how deep the tech industry is now embedded in military operations.

For critics, this raises hard questions about what AI companies are willing to build. For supporters, it is AI being put to work protecting lives in dangerous environments.

Anthropic acquired Stainless today — the company that automatically generates clean, idiomatic SDKs from OpenAPI specs. If you have used the official Claude Python or TypeScript SDK, you have used Stainless's output. Anthropic has been a Stainless customer for years; now it owns the tooling.

Stainless generates production-quality SDKs for Python, TypeScript, Java, Go, and other languages directly from an API specification. For any company shipping an API, this removes the engineering burden of maintaining multiple language-specific SDKs by hand.

The acquisition was posted by a Stainless founder on Hacker News and quickly rose to the top of the site. The developer community is paying attention — both because it validates the SDK generation space and because it tells you something about how Anthropic is thinking about its developer platform.

Vercel Labs released Zero, a new programming language designed specifically for AI agents. Most programming languages were built for humans to write and read. Zero was designed from the start for agents that run tasks, manage memory, and call tools.

The language gives agents precise control over what they remember, what they can do, and when they stop. It is still early — the documentation is thin and the project is explicitly experimental. But it has been picking up attention in the developer community as a serious attempt to ask: if agents are doing real software work, what does a language purpose-built for them look like?

Vercel has the developer platform reach to make an experimental project worth watching, even at this early stage.

Most AI coding agents can be extended with custom skills — small sets of instructions that tell the agent how to handle specific tasks. But there is no standard for what a skill looks like or whether it is safe to use.

tech-leads-club/agent-skills is an attempt to fix that. It is a vetted, structured registry of skills that work across the main AI coding tools: Claude Code, Cursor, Copilot, and others. Each skill is reviewed before it is added, so developers do not have to guess whether a community skill will cause problems.

This is trending today among developers who use AI tools daily and want their agent setups to be more reliable and less experimental.

AI coding agents spend a lot of time searching your codebase from scratch every time you start a new task. CodeGraph changes that by building a searchable map of your entire code library before any agent touches it.

The index lives locally on your machine. When an agent needs to understand your code, it reads from the map instead of scanning files one by one. Fewer tool calls, fewer tokens, faster answers. It works with Claude Code, Codex, Cursor, and OpenCode.

For developers working on large or complex projects, this kind of pre-indexed setup can make AI coding feel noticeably more responsive. It is a practical fix for a problem anyone using these tools regularly will recognize.

When an AI coding agent cannot find what it needs in a codebase directly, it falls back to scanning everything — which burns through tokens fast. Semble fixes this with a smarter search approach that finds the right piece of code with roughly 98% fewer tokens than a typical scan.

Two developers built it after hitting this problem repeatedly while using Claude Code on large projects. The result is a drop-in library: add it to your project, and your AI coding agent gets much more efficient on search-heavy tasks. Less scanning, less waste, same quality answers.

It launched today on Hacker News as a Show HN and has generated the highest comment count of any story in today's edition — a signal that developers recognize the problem immediately.