On September 18, 2026, Anthropic released version 2.1.277 of its developer tool Claude Code. The primary technical update is native support for the AGENTS.md configuration file as an automated fallback. Whenever the system does not locate a model-specific CLAUDE.md in the root directory of a project, Claude Code now parses AGENTS.md instead. The change was announced by Anthropic engineer Thariq Shihipar and directly responds to the growing configuration overhead in modern multi-agent development environments.
The update addresses widespread frustration across the software engineering community. In recent months, industry leaders such as Shopify CEO Tobi Lütke had openly criticized the proliferation of fragmented tool settings. Development teams were frequently forced to maintain redundant configuration files with identical project guidelines across different platforms, including Cursor, OpenAI Codex, and Claude. By recognizing AGENTS.md, Anthropic aligns with an emerging cross-platform standard that consolidates coding styles, architectural rules, and tooling instructions.
From an architectural standpoint, the support was not implemented directly into the weights of the foundation model. Anthropic instead integrated the feature through the new internal Mods system inside the Claude Code harness. This modular architecture allows declarative rule sets to be injected dynamically into the agent runtime without retraining or fine-tuning the underlying model. Consequently, software teams can establish consistent project rules that remain portable across diverse agent toolchains.
Alongside the terminal tool update, Anthropic initiated the beta rollout of Claude Projects v2 for Pro and Max tier subscribers. While the initial generation of Projects functioned primarily as a static file repository and context store, version 2 transforms the service into an active orchestration environment. Complex tasks can now be distributed across concurrent execution threads running inside isolated Cloud Sessions, enabling developers to run multiple workloads simultaneously.
Within Projects v2, intermediate artifacts generated by various sub-agents are consolidated into a shared, persistent workspace memory. This architecture allows specialized helper agents to perform asynchronous refactoring or unit testing without cluttering the primary conversational context. The system breaks larger development assignments into structured sub-tasks and synthesizes the outputs before any code modifications are submitted back into the project repository.
The scale of agentic automation at Anthropic itself was highlighted in a recent company progress report. According to the document, Claude now autonomously carries out 26 percent of Anthropic's internal research and development workloads. The organization routinely orchestrates tens of thousands of internal agents simultaneously to maintain software repositories, monitor training runs, and validate experiment results. The updates in Claude Code and Projects v2 represent a direct commercial application of these internally tested scaling practices.