Google handed its Agent2Agent protocol to the Linux Foundation directed Agentic AI Foundation last week, putting it into the same governance as Anthropic’s Model Context Protocol. The transfer was confirmed on August 20.
Strip away the standards body language and the practical question is simple. If you build on agent tooling this year, how much of your work is portable when a vendor changes its pricing, its terms, or its mind?
The Handoff, and Who Is Behind It
The foundation’s member count currently exceeds 250 people, up from 49 at launch less than a year ago. Its platinum signatories include Anthropic, AWS, Block, Cloudflare, Bloomberg, Google, OpenAI, and Microsoft.
That roster is the story. Competitors rarely agree on plumbing unless each of them concludes that fragmentation costs more than cooperation.
Neutral governance also changes the maintenance picture. Security patches, specification changes and versioning move to a shared process instead of a single company’s roadmap.
What Each Protocol Actually Covers
The two standards solve different problems, and the distinction matters when you are choosing tools.
| Protocol | Origin | What it standardizes |
|---|---|---|
| MCP | Anthropic, founding project | How a model reaches tools, files, search results and databases |
| A2A | Google, transferred August 2026 | How separate agents communicate with one another |
MCP handles the vertical connection between a model and the resources it needs. A2A handles the horizontal one, letting agents built by different teams or vendors pass work between themselves.
Together they cover most of what a multi step automation needs to do. Neither one tells you whether the automation is any good.
That caveat is worth repeating, because standards adoption is often mistaken for capability. A protocol makes systems compatible, and compatibility is a floor rather than an advantage.
Why Lock In Was the Underlying Risk
Twelve months ago, wiring an agent into your stack usually meant adopting one provider’s conventions end to end. Switching later meant rewriting integrations rather than swapping a key.
For a small team, that is not a technical inconvenience. It is a pricing exposure, because a vendor who knows migration is expensive has little reason to keep rates friendly.
Founders have seen this movie in other categories. Payment processors, email providers and analytics platforms all went through a period where switching was technically possible and practically miserable, and pricing reflected that friction for years.
Shared protocols reduce that leverage. They do not eliminate it, and founders evaluating AI agents for business should still ask what happens to their data and workflows on the way out.
How Small Teams Should Respond
Prefer tools that speak MCP or A2A natively. It costs nothing today and preserves optionality later, which is the cheapest kind of insurance available to an engineering team of four.
Keep your prompts, tool definitions and orchestration logic in your own repository rather than inside a vendor console. Portable standards do not help if the configuration lives somewhere you cannot export.
Write down which vendor does what, and keep that list current. When a contract renewal or an outage forces a change, the teams that move fastest are the ones that already know exactly which pieces are swappable.
Also budget for review. Agents that talk to other agents expand what an attacker can reach, and AI cybersecurity practices have not caught up to that surface area yet.
Finally, do not rebuild working systems just to adopt a standard. Apply this to new work and to integrations you were already planning to replace, including the AI coding tools your team touches daily.
The Parts Still Unresolved
Governance consolidation is not adoption. Plenty of specifications have sat under neutral foundations while the market ignored them, and membership counts are a weak proxy for shipped implementations.
Identity and permissions remain the hard problem. When one agent delegates to another, it is still unclear who authorized the action, who is liable for it, and how you audit the chain afterward.
Watch for two signals over the next two quarters. First, whether major platforms ship A2A support in production rather than in preview. Second, whether the foundation publishes a permissions specification with teeth. Announcements from the Linux Foundation newsroom are the cleanest place to track both.
The takeaway for founders is narrow but real. Interoperability is now the default direction of travel, so choosing tools that honor it is a low cost decision you can make this week.
Founder FAQ on Agent Standards
Do I need to understand these protocols to use AI tools? No. You need to know whether the tools you buy support them, which is a purchasing question rather than an engineering one.
Does this make agents safer? Not by itself. Shared standards improve patching and auditability, but permissions and identity remain unsettled.
Is one protocol replacing the other? No. They address different layers and are designed to work together under the same governance.