The Agent Plugin Land Grab: Who Controls the AI Distribution Layer?
AI-generated image · US National Wire
A new 'write-once-run-anywhere' standard for AI agents promises developer freedom, but the real battle is over who owns the marketplaces and interfaces.
The AI industry is attempting to standardize the chaos of agentic development, but if you follow the money, the 'interoperability' pitch looks more like a strategic move to secure the distribution layer of the AI economy.
As first reported by The Register, a group of industry titans—including **Amazon**, **Microsoft**, **OpenAI**, **Cursor**, and **Vercel**—have collaborated on the **Agent Plugins 1.0** specification. The goal is to create a modular, 'write-once-run-anywhere' container that allows developers to move agentic skills and tools across different platforms seamlessly.
***Opinion:*** *While the industry frames this as a win for developer portability, the fine print reveals a desperate land grab. By standardizing the 'skills' and 'tools,' these giants are effectively commoditizing the labor of the creators while fighting for control over the high-margin distribution points: the user interfaces and the marketplaces.*
As The Register details, the technical architecture relies on a combination of three elements:
* **Agent Skills:** An open standard created by **Anthropic** in 2025 that provides instructions for specific tasks. * **MCP (Model Context Protocol):** A protocol stewarded by the **Agentic AI Foundation (AAIF)** that handles connectivity to APIs, shell scripts, and external databases. * **Agent Plugins 1.0:** The new spec, initially drafted by **Vercel**, which packages these skills and MCP configurations into a single directory structure using a JSON manifest.
Despite the push for a universal backend, the actual power remains concentrated in the hands of the platform holders. The Register notes that it is still up to the agent clients to establish security boundaries, provide the user interface, and—most importantly—create the marketplaces where these plugins are shared. Currently, the platforms supporting this standard include **ChatGPT**, **VS Code**, **GitHub Copilot**, **Cursor**, **Codex**, and **Kiro**.
Other efforts to consolidate this frontier are already underway. The Register reports that **Google** has folded its **Agent2Agent (A2A)** protocol into the **AAIF** and is developing **Agentic Resource Discovery** to help find tools on the web. Additionally, **AAIF Vice President Angie Jones** highlighted a complementary project called **Skills Over MCP**, which bundles skills directly into the MCP server.
While the **AAIF** claims the 1.0 release is intentionally 'narrow' to keep the format predictable, the core development team remains a tight circle of engineers from **Amazon**, **Cursor**, **Microsoft**, **OpenAI**, and **Vercel**. The spec is licensed under the **Creative Commons Attribution 4.0 International License (CC-BY-4.0)**, but the real value isn't in the license—it's in who controls the gateway to the user.