Can AI agents develop a “common language”? As of April 15, 2026, the answer seems to be yes—but not in the simple way many people imagine. OpenAI’s latest update to the Agents SDK does not introduce one grand universal protocol. Instead, it strengthens a growing stack of shared conventions: MCP for tool use, skills for reusable workflows, and AGENTS.md for project-specific instructions. OpenAI describes this as “standardized infrastructure” for agents, adding a model-native harness and native sandbox execution so agents can work across files, tools, and long-running tasks more reliably. (openai.com)
The most important standard in this race may be MCP, or Model Context Protocol. MCP is an open standard for connecting AI applications to external systems such as files, databases, and tools; its own documentation compares it to a USB-C port for AI. OpenAI has already woven MCP deeply into its ecosystem: the Apps SDK in ChatGPT builds on MCP, OpenAI says MCP is the foundation for connectors and apps in ChatGPT, and the Agents SDK documentation now includes hosted MCP integration through the Responses API. In other words, OpenAI is not trying to replace this protocol—it is helping normalize it. (docs.anthropic.com)
But MCP is only one layer. On April 9, 2025, Google announced A2A, an open protocol for agent-to-agent communication. Google explicitly positioned A2A as complementary to MCP: MCP helps agents use tools and context, while A2A helps separate agents discover one another, exchange information securely, and coordinate tasks. Then, on June 23, 2025, Google donated A2A to the Linux Foundation, saying more than 100 companies supported the protocol. That move made the contest feel less like a product war and more like a standards war. (developers.googleblog.com)
The biggest clue came on December 9, 2025, when OpenAI, Anthropic, Google, Microsoft, Amazon, Bloomberg, Cloudflare, Block, and others backed the new Agentic AI Foundation under the Linux Foundation. MCP and OpenAI’s AGENTS.md became founding projects, and OpenAI said AGENTS.md had already been adopted by more than 60,000 open-source projects and frameworks. So, will AI agents end up speaking one common language? Probably not one language, but a family of shared standards—more like the internet’s protocol stack than a single dictionary. That may be even more powerful. (openai.com)









