While GTC 2026 dominated this week’s headlines with hardware announcements and NemoClaw demos, two unrelated events on Tuesday told a quieter but arguably more consequential story: the agent communication protocol layer is crossing from concept to production.

First, IBM published its 2026 AI and tech trends forecast, in which IBM’s Kate Blair declared: “It’s been just a year since Anthropic launched MCP, alongside IBM’s ACP and Google’s A2A. If 2025 was the year of the agent, 2026 should be the year where all multi-agent systems move into production.” That’s a major enterprise incumbent saying the infrastructure layer for agent interoperability is ready — and that 2026 competition will be fought on systems, not models.

Hours later, airSlate SignNow shipped the SignNow MCP Server, a native Model Context Protocol integration that lets AI agents autonomously send, manage, and track e-signature requests from plain-language prompts. No human in the loop. An agent writes a prompt; a contract goes out for signature.

Three Protocols, One Direction

The agent protocol landscape has three contenders. Anthropic’s Model Context Protocol (MCP) defines how AI models interact with external tools and data sources. IBM’s Agent Communication Protocol (ACP) focuses on agent-to-agent messaging in enterprise environments. Google’s Agent-to-Agent Protocol (A2A) handles cross-platform agent coordination.

A year ago, all three were spec documents and reference implementations. Today, MCP has the largest ecosystem by developer adoption — it ships natively in Claude, has integrations with Cursor, Windsurf, and dozens of developer tools, and now has its first vertical SaaS integration in document signing. ACP has IBM’s enterprise distribution behind it. A2A has Google Cloud’s customer base.

IBM’s framing — all three are heading toward production simultaneously — suggests the question is no longer “will agent protocols matter?” but “will they converge or fragment?”

Why SignNow Matters More Than It Looks

airSlate SignNow is not a household name, but the product category is massive — electronic signatures are one of the most widely adopted enterprise SaaS categories. An MCP server for e-signatures means an AI agent can now complete one of the most common business workflows — “send this contract to the client for signature and tell me when they sign” — without a human opening a browser.

This is exactly the kind of integration that turns protocol adoption from a developer concern into a procurement decision. A VP of Legal doesn’t care about MCP. They care that their AI assistant can send NDAs. But the agent that sends those NDAs needs MCP to talk to SignNow’s API. The protocol becomes invisible infrastructure — which is when protocols win.

The Infrastructure Story Under the Hardware Headlines

GTC 2026 gave us NemoClaw, the Agent Toolkit, Nemotron models, and six enterprise integration partners. All of that runs on Nvidia GPUs. But the layer between the GPU and the application — the protocol that lets Agent A call Tool B and coordinate with Agent C — belongs to MCP, ACP, and A2A.

Nvidia’s Agent Toolkit already incorporates OpenShell for sandboxed execution, but it still needs a protocol layer for multi-agent coordination and external tool access. The company hasn’t publicly committed to MCP, ACP, or A2A. That decision — or the decision to build its own — will shape whether the agent infrastructure stack consolidates or splinters.

IBM’s Kate Blair may be right that 2026 is when these protocols hit production. The SignNow launch suggests MCP got there first.

Sources: IBM Think — AI Tech Trends 2026, airSlate SignNow MCP Server Launch