Claude Code Usage Climbs at Anthropic; New Eval Tool and Agent APIs Emerge
AI agent capabilities expand across platforms with new APIs and evaluation tools, while security concerns and adoption trends shape the evolving skills landscape.
New here? The 60-second primer
Skills are packaged expertise you install into an AI assistant: Claude Code has skills and plugins (from Anthropic's directory and open git marketplaces), ChatGPT has custom GPTs (the GPT Store), and Google's Gemini has Gems (shareable custom experts). Every rank below comes from captured data — official install counts, GitHub star growth, and the GPT Store's own trending list — not from opinion.
The week
This week saw significant movement in the AI skills landscape, particularly around agentic capabilities and their adoption. Anthropic's Claude now leads 26% of the company's internal research and development work, a substantial increase from just 1% in February.
This growing reliance on Claude within Anthropic itself highlights its utility for complex tasks. Supporting developers, Claude Code introduced a new `claude plugin eval` command, enabling authors to score their plugins against a baseline for performance evaluation.
Additionally, Claude Code now supports `AGENTS.md` files as a built-in plugin, offering more flexible project instructions. On the broader agent front, OpenAI announced a new Agents API, signaling a shift towards more autonomous agent designs.
Meta also unveiled a new AI agent capable of interacting across applications for tasks such as sending emails and arranging travel. Meanwhile, Salesforce is deepening its integration of AI agents into daily tools like Slack, following its acquisition of Fin for AgentForce.
Google is also making Claude available to its engineers, further expanding its reach. These developments collectively point to a rapid expansion of agentic AI, with a growing emphasis on practical application and measurable performance.
Pattern watch
A clear pattern emerging this week is the maturation of agentic AI, moving from simple prompt-response to more autonomous, orchestrated workflows. Anthropic's internal adoption of Claude for 26% of its R&D and the introduction of `claude plugin eval` underscore a focus on verifiable performance and reliability.
This push for robust, evaluable agents is critical as platforms like OpenAI, Meta, and Salesforce introduce new agent APIs and integrations, highlighting the need for developers to ensure skill quality and security. The increasing complexity and independence of these agents necessitate better evaluation tools and governance, a trend also noted with Red Hat focusing on governance for AI coding agents.
Reporting + analyst voices: grounded via Google Search at publish time.