Does this mean I no longer need to worry about vendor lock-in?

It lowers the risk considerably. When the connection layer is an open standard rather than one vendor's proprietary format, the integrations you build stay portable if you switch model providers later. You still choose a provider and a partner, but the plumbing - how the agent reaches your systems - is no longer a bet on one company.

Do open standards change the safety rules for deploying an agent?

No. Standards make integration cheaper and more portable, but they do not scope your process, wire up your data or decide what an agent is allowed to do. The safety rules still stand: read-only access by default, human approval on any write, everything running inside your environment, and a full audit trail.

The AI agent standards war just ended

Q: What is the Agentic AI Foundation?

The Agentic AI Foundation (AAIF) is a neutral body hosted by the Linux Foundation to steward the open standards that let AI agents connect to tools and to each other. It was formed on 2025-12-09 with founding contributions from Anthropic, Block and OpenAI - the Model Context Protocol (MCP), goose and AGENTS.md - and grew from 49 members to more than 250 in under a year. On 2026-08-20 Google's A2A protocol joined it, putting the two main agent standards under one roof.

Q: What is the difference between MCP and A2A?

MCP (Model Context Protocol) is vertical: it standardizes how one agent reaches your tools and data - the ERP, the CRM, a database, a search index. A2A (Agent-to-Agent) is horizontal: it standardizes how two agents talk to each other, negotiate a task and pass credentials across organizations. MCP connects an agent to your systems; A2A connects agents to other agents.

An AI agent standard is the shared language an agent uses to reach your systems and to coordinate with other agents. For most of 2025 there were competing, vendor-owned versions of that language, and choosing one felt like a bet. As of this month that bet is largely settled: on 2026-08-20 Google's A2A protocol joined the Linux Foundation's Agentic AI Foundation, where it now sits beside Anthropic's Model Context Protocol. Here is what the consolidation means in practice - specifically for a small or traditional business, not for a hyperscaler.

Key takeaways. The two main agent standards - MCP (agent-to-tool) and A2A (agent-to-agent) - now live under one neutral, open foundation backed by AWS, Anthropic, Google, Microsoft, OpenAI and others. That makes the integration layer between an agent and your ERP, CRM and inbox cheaper, more portable and less of a vendor bet. It does not scope your process, wire up your data or decide what the agent may touch. The safety rules are unchanged: read-only by default, human approval on writes, everything in your environment, full audit trail.

What actually happened

The plain facts first, because they are easy to overstate. The Agentic AI Foundation was formed on 2025-12-09 by the Linux Foundation, with founding contributions from Anthropic, Block and OpenAI - the Model Context Protocol, the goose agent framework, and the AGENTS.md convention. It grew from 49 members to more than 250 in under a year, with AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI all among its top-tier backers. Then, on 2026-08-20, Google donated its A2A protocol to the same body. In other words, the two companies whose formats mattered most now steward them jointly in neutral, open governance rather than competing to own the standard. That is the news, and for once the practical reading is simpler than the headline.

The two standards, in plain terms

You do not need to implement either protocol yourself, but the distinction is worth thirty seconds because it maps cleanly onto how automation reaches your business. MCP is vertical: it standardizes how a single agent reaches a tool or a data source - your ERP, your CRM, a database, a shared drive. A2A is horizontal: it standardizes how two agents talk to each other, hand off a task and pass credentials, even across two different organizations. Think of MCP as the wiring between an agent and your systems, and A2A as the wiring between agents. We covered the first in depth in what MCP means for connecting AI to your business systems; the August news is that its counterpart for agent-to-agent coordination now lives in the same, openly governed place.

Why this matters more to you than to a tech giant

A hyperscaler can afford to back the wrong standard and rewrite later; a mid-market manufacturer or a small firm cannot. That is exactly why consolidation helps the smaller company most. When the connection layer is an open standard rather than one vendor's proprietary format, the integrations you pay to build stay portable if you later switch model providers - the agent that reads your orders today is not welded to whichever lab was fashionable this quarter. Consequently the practical fear that stalls so many first projects, "what if we pick wrong and have to start over," loses most of its force. It also nudges the ecosystem away from the fragmented mess we described in AI agent sprawl, because shared standards make it far easier to consolidate on one governed layer instead of ten disconnected tools.

What does not change

Here is the part the announcements will not tell you, and it is the part that decides whether your project works. Standards make integration cheaper and more portable; they do not do the actual job. A protocol will not choose which bounded process to automate, it will not clean and wire up your data, and it will not decide what the agent is allowed to touch. Those remain human decisions, and they are still where most pilots fail rather than in the plumbing. The security posture is likewise untouched: as we argue in deploying AI without data leaving your perimeter, an agent should get read-only access by default, every action should be logged, and anything customer-facing or financial should wait for human approval. Open standards make the wiring portable. They do not make governance optional.

What to do this quarter

Do not wait for the "final" standard. The consolidation is the signal to start, not another reason to delay. The bet you were afraid of has mostly been made for you.
Pick one bounded process. Clear start and finish, measured in hours saved - the same first step in our implementation guide. The standard is irrelevant until you have a process to point it at.
Ask vendors about MCP and A2A support. Any partner you evaluate should build on the open standards, so your integrations stay portable. Treat proprietary-only connectors as lock-in.
Keep the guardrails. Read-only defaults, human approval on writes, in-perimeter deployment and a full audit trail - decided before anyone writes code, as in our back-office automation guide.

If you are still weighing whether your first automation should be a bounded agent at all or an off-the-shelf copilot, our decision framework for custom agent versus copilot is the right place to start. The standards question is now the easy one. The hard, valuable work - choosing the process, wiring the data, keeping a human at the gate - is exactly where it always was.

Frequently asked questions

What is the Agentic AI Foundation?

A neutral body hosted by the Linux Foundation, formed on 2025-12-09, that stewards the open standards behind AI agents. It grew from 49 members to over 250 in under a year, and on 2026-08-20 Google's A2A protocol joined Anthropic's MCP inside it.

What is the difference between MCP and A2A?

MCP connects an agent to your tools and data (vertical); A2A connects agents to other agents (horizontal). One is the wiring to your systems, the other is the wiring between agents.

Does this end vendor lock-in?

It reduces it sharply. An open connection layer keeps your integrations portable across model providers, so you are no longer betting the plumbing on one company.

Do standards change the safety rules?

No. Read-only by default, human approval on writes, in-perimeter deployment and a full audit trail all still apply. Standards move the wiring, not the governance.

The AI agent standards war just ended: what it means for your business