/
Published on 06.01.2026
TLDR: The AI industry has crystallized around a layered protocol stack for agents - MCP for tool/data access, Function Calling for external software, A2A for cross-vendor agent coordination, and AP2 for autonomous payments. Understanding these protocols is now a basic requirement for building with AI agents.
If 2025 was the year AI agents became real, 2026 is about the connective tissue that makes them useful. The bottleneck for agentic AI isn't model capability - it's execution infrastructure. Current AI systems can generate detailed instructions, analyze options, make recommendations, and create step-by-step plans. But they cannot access the tools needed to execute those plans, conduct transactions across systems, or maintain security while acting autonomously.
This is a standardization moment analogous to TCP/IP for the internet, HTTP for the web, or REST APIs for cloud computing. The December 2025 formation of the Agentic AI Foundation (AAIF) formalized this standardization, and the protocol landscape has crystallized into distinct layers:
Model Context Protocol (MCP) sits between AI models and external resources, standardizing how agents connect to local or remote data sources like Google Drive, Slack, or databases. It uses a client-server architecture with local/remote flexibility and has a growing open-source ecosystem. Anthropic's strategic bet here: commoditize integrations, compete on model quality.
OpenAI Function Calling standardizes the JSON format for agents to trigger external software tools - the mechanism by which agents actually do things in the world.
agents.json is a discovery file that tells agents which capabilities an API or website offers - essentially making services machine-readable for autonomous discovery.
Agent-to-Agent (A2A) allows agents from different vendors (Google, Microsoft, etc.) to communicate and share tasks. This is critical for the multi-agent futures people are building toward.
Agent Communication Protocol (ACP) manages memory and state for long-running agent tasks - solving the problem of agents that need to maintain context across extended operations.
Open Agentic Schema Framework (OASF) defines an agent's "resume" - its skills and costs - so other agents can discover and hire them. Think of it as a registry for agent capabilities.
Agent Protocol provides a universal API (POST /task) so developers can swap agent frameworks without changing code - portability across the emerging ecosystem.
AP2 (Agent Payments) enables agents to pay each other autonomously using digital wallets. Built on Google's existing ecosystem and enterprise payment infrastructure.
For architects and technical leaders, the strategic implications are significant. Protocol choices determine who captures value in the agent economy, which companies become ecosystem gatekeepers, how much interoperability exists, and whether we get open ecosystems or walled gardens.
The defensive play is preventing lock-in to competitors' ecosystems. The offensive play is creating ecosystems that generate network effects and switching costs. For enterprises, it's about future-proofing AI investments while maintaining security and control. For developers, it's building on infrastructure that won't be obsoleted by platform changes.
Three trends make this standardization urgent: enterprise demand for agentic AI that works with existing systems without security compromises, regulatory scrutiny requiring audit trails for autonomous systems, and economic scale requiring standardized transaction protocols for network effects.
Value capture has shifted from controlling the protocol itself (those are commoditizing) to building specialized agent skills and proprietary orchestrators that run on top of shared standards. The protocol layer is becoming plumbing; the differentiation happens above it.
Link: The Agent Protocol Handbook
The content above is AI-generated based on newsletter sources. While I strive for accuracy, please verify critical information from original sources.