Claude Code now reads AGENTS.md if there is no Claude.md
summary
The Claude Code updates from September 2026 provide a comprehensive set of bug fixes, performance improvements, and new features across the CLI, VS Code extension, and web interface. The updates focus on enhancing stability, refining the tool-set (especially the Artifact tool and MCP servers), and improving the user experience for both individual and enterprise users. By addressing numerous crashes, memory leaks, and authentication errors, Anthropic is refining the tool's integration into developer workflows.
key points
- Significant improvements to the MCP (Model Context Protocol) and plugin system, including better error handling, automatic reconnection, and synchronization of plugins between the account and terminal sessions.
- Extensive updates to the VS Code extension and web-based cloud sessions, adding new session management tools, improved diff views, and better integration with GitHub and organization-wide settings.
- Refinements to the 'Artifact' tool, including better permission prompts, improved publishing workflows, and better handling of shared artifacts and database writes.
who should read it
Developers and enterprise administrators using Claude Code for AI-assisted coding, as well as those utilizing the MCP ecosystem and VS Code extension.
discussion skeptical
consensus
Commenters generally view the addition of support for AGENTS.md as a long-overdue, bare-minimum requirement rather than a feature to be praised. There is a strong sentiment that Anthropic has been intentionally proprietary and disconnected from developer needs, forcing users into inefficient workflows. Many users expressed frustration with the lack of industry standards and the friction caused by provider-specific configuration files.
top insight
A veteran engineer argues that current AI harness failures are not due to LLM limitations, but rather a lack of experienced engineering leadership. He suggests that 22-year-old developers are ignoring established computing principles and hard-won engineering solutions that already exist, while the LLMs themselves actually possess the necessary historical knowledge to implement these solutions correctly if properly guided.
Generated by an LLM from the article and thread. Check the sources before relying on it.