AI 工具与 Agent 必读

过去 24 小时，28 个来源有 142 条相关新消息，这里是被报道最多的 8 条。

1. Anthropic Says Nobody Needs a Mental Model of Claude Code. Practitioners Are Testing That

   InfoQ · infoq.com · 另有 4 家在报

   Anthropic has published how the Claude Code team reviews code now that Claude writes roughly 90% of it, including a claim that nobody keeps a complete mental model of the codebase. The post documents a pull request that…

2. Google launches Gemini 4 Carbon with a ‘significant leap’ in cybersecurity

   The Verge · theverge.com · 另有 5 家在报

   Google has released its latest frontier model, Gemini 4 Carbon, offering "significant gains" in coding, agentic tasks, and cybersecurity. In an announcement on Wednesday, Google highlights Carbon's top spot on the…

3. Mistral’s Large 4 open weights arrive with a 1-point coding lead over GLM-5.3: “There’s nothing you can’t do with this model that you can do with competitors.”

   The New Stack · thenewstack.io · 另有 2 家在报

   On Thursday, Mistral released the weights for its frontier open model, Mistral Large 4 (ML4), on Hugging Face. Made available The post Mistral’s Large 4 open weights arrive with a 1-point coding lead over GLM-5.3…

4. Hacker stole personnel data about thousands of Homeland Security employees, including CBP and ICE agents

   TechCrunch · techcrunch.com · 另有 1 家在报

   A hacker obtained personally identifiable information about more than 40,000 Homeland Security staffers, which has since been published online.

5. Chrome’s AI-powered Autofill can now dig through your Gmail to fill out forms

   TechCrunch · techcrunch.com · 另有 1 家在报

   The autofill update lets Chrome pull flight confirmations, tracking info, order details, and other information from your inbox, with your permission.

6. I Put a Bunch of AI Agents in Charge of a Fantasy Soccer Team. They’re Not Better Than Humans—Yet

   Wired · wired.com · 另有 1 家在报

   To experiment with AI’s predictive skills, I let Codex loose on a fantasy team. It tried to bench Erling Haaland at first, but it soon learned the error of its ways.

7. Why Your AI Coding Agents Suffer from Amnesia (And How I Fixed It with SQLite + ONNX)

   Dev.to · dev.to · 另有 1 家在报

   If you use Claude Desktop, Cursor, or OpenCode, you know the daily struggle: every new session is a blank slate. Standard RAG solutions treat memory as flat vector chunks, which makes them fundamentally time-blind. To…

8. Why MCP servers should tell agents what to do next

   The New Stack · thenewstack.io

   Many MCP servers do one thing. An agent asks for data, and the server hands it back. AWS’s James Ward The post Why MCP servers should tell agents what to do next appeared first on The New Stack .