Anthropic's Claude Code team confirms support for AGENTS.md standard

Anthropic’s Claude Code team has confirmed it will support the AGENTS.md industry standard, addressing previous criticisms regarding compatibility. Thariq Shihipar, a core member of the Claude Code team, suggested that as models become more capable, static instruction files like CLAUDE.md or AGENTS.md may eventually become unnecessary as models handle more context autonomously.

Technical discussions surrounding Claude Code highlight the use of sub-agents, hooks, and skills as an orchestration layer. Developers are encouraged to use sub-agents for complex, independent tasks that require state isolation or specialized expertise, while utilizing inline tool use for tightly coupled, low-latency operations to conserve resources.

Safety remains a priority as agents gain more autonomy. Anthropic is exploring multi-layered defenses, including intent-level monitoring through probes and permission-based checks in auto mode, to prevent agents from unauthorized actions such as hacking databases or attempting to hide behaviors within their chain of thought.