Five Tech Giants Unite Against Anthropic's AI Agent Standard

30-Second Summary

Google, Microsoft, Salesforce, Snowflake, and ServiceNow have agreed to back a shared AI agent connectivity standard built on Google's Agent-to-Agent (A2A) protocol. The alliance is a direct response to Anthropic's Model Context Protocol becoming the de facto default for AI agent integration over the past 18 months. For operators, this means the platforms holding your business data are about to become the battleground for which standard controls how AI agents access and act on that data.

At a Glance

• Topic: Agent Systems, Enterprise AI Standards
• Companies: Google, Microsoft, Salesforce, Snowflake, ServiceNow, Anthropic (incumbent)
• Date: 13 July 2026
• Announcement: Five enterprise technology vendors agree to back a shared standard for connecting AI agents to business software, countering Anthropic's Model Context Protocol
• What Changed: Anthropic's MCP has had near-monopoly status as the default agent connectivity protocol since early 2025. Five of the largest enterprise software vendors now have a coordinated alternative backed by Google's A2A standard
• Why It Matters: The companies in this alliance collectively own the software platforms, Salesforce CRM, ServiceNow IT service management, Snowflake data warehousing, Azure infrastructure, and Google Cloud, where most enterprise business data already lives. Whoever controls the connectivity standard controls how AI agents access and act on that data
• Who Should Care: Any operator using AI agents that connect to SaaS platforms or enterprise software; any business evaluating AI integration investment over the next 12 to 24 months

Key Facts

• Anthropic's Model Context Protocol (MCP) became the dominant AI agent connectivity standard over approximately 18 months from early 2025, adopted across most major AI tools including Claude, ChatGPT plugins, and developer frameworks
• Google's A2A (Agent-to-Agent) protocol is the technical foundation for the new alliance. It is designed specifically for multi-agent orchestration, allowing a Salesforce agent to hand off tasks to a Google agent, which can then query a ServiceNow agent, all through a single standard
• The five companies backing the new standard, Google, Microsoft, Salesforce, Snowflake, and ServiceNow, collectively serve the vast majority of enterprise software deployments globally
• All five companies, along with OpenAI and Anthropic, are simultaneously members of a Linux Foundation working group building open AI agent standards. The alliance does not appear to conflict with that participation
• The announcement coincides with a period of rapid agent deployment across enterprise: the share of organisations using generative AI in at least one business function reached approximately 71% in 2024 and has continued growing through 2026

What Happened

On 13 July 2026, Google, Microsoft, Salesforce, Snowflake, and ServiceNow announced they would jointly back a shared standard for connecting AI agents to enterprise business software. The standard is built on Google's existing A2A protocol, which the company developed as a specification for orchestrating multiple AI agents from different vendors on a single task.

The backdrop is Anthropic's Model Context Protocol. Released in late 2024, MCP became the dominant integration layer for AI agents within roughly 18 months, adopted as a standard connection method across a wide range of AI tools and developer frameworks. Its growth came primarily from the bottom up: developers adopted it because it was open, well-documented, and supported by Anthropic's Claude, which had become a widely deployed enterprise AI.

The five vendors in this new alliance make the bulk of the software where enterprise data lives. Salesforce manages customer relationships and sales pipelines. ServiceNow manages IT workflows and operational processes. Snowflake holds structured data that powers reporting and analytics. Microsoft runs Azure infrastructure and the Office 365 productivity layer. Google operates Cloud infrastructure and Workspace. Their concern is that Anthropic's protocol, embedded in their platforms, gives a competitor structural control over how AI agents connect to the tools these five companies sell and maintain.

A2A is positioned as the standard for orchestration between agents from different vendors. Its practical implication is interoperability without forcing any single AI provider to be the connectivity layer. A Salesforce Agentforce agent could hand a task to a Google Vertex AI agent, which could retrieve data from a ServiceNow agent, all through A2A, without requiring Anthropic's MCP as the connector.

Why It Matters

The connectivity layer is becoming as important as the model itself. The AI model your business uses will continue to commoditise. The integration layer, the protocol through which agents access your CRM, your databases, and your workflows, is becoming the more durable competitive advantage. Whoever controls that standard shapes the AI agent market for years.

Enterprise platforms are not neutral in this competition. If you use Salesforce, ServiceNow, or Snowflake, the vendors you pay for have now formally taken a position. Their native agent capabilities will be optimised for A2A. That does not mean MCP will stop working inside those platforms, but it does mean future native features, deeper integrations, and certified agent workflows will be built with A2A as the priority.

Anthropic's MCP has momentum that will not disappear overnight. MCP has 18 months of ecosystem adoption, an enormous developer community, and is embedded in Claude, ChatGPT plugins, and most of the major AI development frameworks. Displacing it requires the A2A alliance to ship high-quality tooling, documentation, and platform integrations at pace. Standards wars in enterprise software typically take two to four years to resolve.

The Linux Foundation parallel signals that the two camps are not irreconcilable. All the major players are simultaneously working on open shared standards through the Linux Foundation. The A2A alliance may function as a vendor coalition applying competitive pressure rather than an attempt to fragment the market permanently. The end state could be a unified standard that incorporates elements of both approaches.

This is a risk signal for bespoke integration work. Any organisation that has spent engineering budget building custom AI agent connectors on either MCP or A2A is now sitting on a potential technical liability. The protocol layer is in active contest. Building on it now is building on sand.

Cost and capability implications will follow competition. The last time a major tech standards war played out, between app store models, browser engines, and cloud APIs, competition between camps accelerated feature development and drove down costs. The same dynamic is likely here. Operators who wait for the dust to settle will inherit better, cheaper, more interoperable tools than those who commit to one side prematurely.

The David and Goliath View

The phrase "shared standard" is always branding. What these five companies are actually doing is defending the value of their own platforms by refusing to let a competitor's protocol become the default plumbing for enterprise AI. That is a rational commercial decision, not a charitable one. But it matters for operators because the intent behind the standard does not change its practical value. If A2A delivers reliable, well-supported agent interoperability across Salesforce, ServiceNow, and Snowflake, the business case for adopting it is real, regardless of the politics that created it.

For operators building AI capabilities in 2026, the most important insight from this announcement is not which standard wins. It is that the pace of enterprise AI maturity is now fast enough that five of the largest software companies in the world feel the need to act defensively. The window for building AI into business operations before competitors do is narrowing more quickly than most operators recognise. The companies moving now, even on imperfect tooling, will have 12 to 24 months of compounding advantage over those waiting for the standards war to resolve.

David and Goliath's position has always been that the AI models and protocols matter less than the workflows they enable. The standards war reinforces this. Build the workflow. Adjust the plumbing later.

Where This Fits in the AI Stack

The AI agent stack has three layers. At the top is the model, the intelligence that reasons and generates. In the middle is the orchestration layer, which manages how agents are coordinated and when they hand off tasks. At the bottom is the connectivity layer, which determines how agents access tools, data, and business software.

MCP operates primarily at the connectivity layer. A2A operates primarily at the orchestration layer. They are complementary in function but competing in the market because control of either layer confers influence over the full stack. The five-vendor alliance is trying to control orchestration while keeping connectivity open. Anthropic controls connectivity while building orchestration through Claude's multi-agent capabilities. Both camps are trying to own the layer that becomes the default surface for enterprise AI.

Questions Operators Are Asking

Do I need to do anything right now? No immediate action is required. If you have AI agent integrations running on MCP, they will continue to work. The A2A standard is not yet widely deployed at the platform level across Salesforce, ServiceNow, and Snowflake. Watch for product announcements from your existing vendors, not for a protocol switch.

Should I stop building MCP integrations? For light integration work, MCP remains the better-supported option with more documentation and tooling. For heavy investment in custom agent pipelines, the risk of building on either standard is elevated until the competitive landscape settles. Consider phasing substantial integration investment over the next 6 to 12 months.

Is this going to affect my Salesforce or ServiceNow costs? Not directly in the short term. The standards alliance is an infrastructure decision that will eventually influence which agent capabilities are native to these platforms. Native capabilities reduce the need for third-party AI tools, which could have cost implications over a 12 to 18 month horizon.

What does this mean for Anthropic and Claude? Anthropic's MCP remains the dominant agent connectivity protocol today. The company has invested heavily in enterprise distribution, and Claude is embedded in many of the same platforms that are now backing A2A. The alliance creates long-term competitive pressure but does not destabilise Anthropic's market position in the near term.

Which standard should we design our AI agent strategy around? Design your strategy around the workflows your business needs, not the protocols. Select vendors that support both standards where possible, and avoid architectural decisions that create deep dependency on either MCP or A2A as a proprietary lock-in point.

Citable Summary

On 13 July 2026, Google, Microsoft, Salesforce, Snowflake, and ServiceNow announced a joint commitment to back a shared AI agent connectivity standard built on Google's A2A protocol. The announcement is a direct response to Anthropic's Model Context Protocol establishing de facto dominance in AI agent integration over the previous 18 months. The five companies collectively own the enterprise software platforms where most business data resides. A2A is designed to enable multi-vendor agent orchestration across these platforms without requiring Anthropic's protocol as the connector layer. MCP retains 18 months of ecosystem momentum and is not immediately displaced. The standards competition is expected to shape enterprise AI agent infrastructure over the next 12 to 24 months.