Anthropic Publishes Official Best Practices for Claude Code

On June 8, Anthropic published best practices for Claude Code, its agentic coding environment. Unlike a traditional chatbot, Claude Code can autonomously read files, run commands, and make edits. Rather than writing the code yourself and asking for a review, you describe the outcome you want and Claude decides how to implement it.

The article distills practical guidance for using Claude Code effectively — context management, how to write prompts, and how to break down tasks — packed with insights on optimizing collaboration with agentic AI. It's well worth a read for developers who already use AI coding in their day-to-day work.

Claude Cowork Now Generally Available on All Paid Plans

Also on the Anthropic front: Claude Cowork, the AI-agent-powered work assistant, is now generally available across all paid plans. Claude Cowork breaks away from the old "AI = asking questions" paradigm, taking an agentic approach that supports the work process itself.

Administrative capabilities for organization-wide rollout have been expanded as well, putting the pieces in place for full-scale enterprise adoption. For companies that had begun to feel the limits of chat-based AI, agentic work assistance looks set to become a compelling new option.

Claude Temporarily Unselectable in Notion AI Amid Performance Degradation Incident

Meanwhile, a notable Claude-related incident was also reported. Notion temporarily disabled the ability to select Claude models in Notion AI following performance degradation in Claude Opus 4.7 and 4.8. According to Notion, the measure was a response to a rising Claude failure rate, with the cause attributed to a "temporary service incident." The issue was resolved as of June 8, but it highlighted just how much fluctuations in AI model quality can affect third-party products.

Gemma 4 31B FP8 Shows Performance on Par with Sonnet 4.6

Local LLMs have taken yet another step forward. In Reddit's LocalLLaMA community, benchmark results were shared showing that the FP8-quantized version of Google's open-source model Gemma 4 31B performs nearly on par with Anthropic's Sonnet 4.6 (medium).

There are also reports that Google's official QAT (Quantization-Aware Training) Q4_0 quantization delivers higher accuracy than Unsloth's Q4_K_XL, throwing the rapidly improving quality of locally runnable models into sharp relief.

SDSU Installed 1,300 AI Cameras Including in Student Dorms, Revealed Without Prior Notice

Drawing attention on Hacker News: San Diego State University (SDSU) had installed roughly 1,300 AI-equipped cameras across campus, including in student dorms. It emerged that students were never informed beforehand, and privacy concerns are mounting.

While the spread of AI surveillance cameras brings security benefits, transparency and consent are essential — especially for deployment in living spaces. How the university responds from here will be closely watched.

IBM Proposes "AI Operating Model," a New Framework for Enterprise AX

IBM has put forward an "AI Operating Model" as a new framework for enterprise AI Transformation (AX). The company argues that businesses need to "update the very premises of their work," advocating a rethink of how organizations operate and of business processes themselves, rather than mere tool adoption.

With enterprises pursuing AI now being called on to design their organizations as well as their technology, the framework is drawing attention as a source of practical guidance.

MIT CSAIL Announces New Architecture for "Self-Repairing AI"

MIT's CSAIL has announced an architecture for "Self-Repairing AI," in which LLMs detect, explain, and correct their own reasoning errors. This metacognitive approach points to a new direction for making reasoning more reliable.

With LLM hallucinations and reasoning mistakes remaining practical challenges, mechanisms that let models recognize and fix their own errors could have a major influence on future AI system design.

Developers Voice Frustration Over No Claude Desktop for Linux

Frustration is mounting in the developer community over the fact that Anthropic's Claude Desktop supports only Windows and macOS, with no Linux version available. Requests have piled up on GitHub, and a related article has gathered 407 upvotes on Hacker News.

The contradiction of touting "AI for developers" while sidelining Linux — the home turf of many developers — has also sparked debate from an AI democratization perspective. Features unique to the desktop app, such as integration with MCP servers, remain unavailable to Linux users.

---