Agents Need a Public Square
Asimov’s Addendum
SubscribeSign in
Agents Need a Public Square<br>Why better agent discovery is needed – and why broadcasting may be the answer
Sruly Rosenblat, Ilan Strauss, and Tim O'Reilly<br>Jun 05, 2026
Share
An illustration of the difference between broadcasting and one-to-one communication
The Agent2Agent (A2A) protocol was announced by Google a little over a year ago (April 2025). It was built to allow agents to communicate with each other across different sites and to enable any agent to reach out to specialized agents hosted by any company. Since then, it was moved to the Linux Foundation where it was later joined by MCP. The primary alternative to A2A, IBM’s Agent Communication Protocol (ACP1), was quickly folded into the former.<br>A2A is now supported on multiple cloud platforms, is integrated into LangGraph and CrewAI, gained over 20 thousand stars on GitHub, and has over 150 supporting organizations. But despite being the only agent communication protocol with any traction, A2A still has almost no publicly deployed agents . What appears to be the largest agent registry for A2A has just 100 publicly accessible agents compared to the ~3k remote MCP servers cataloged by PulseMCP.<br>A lot of this is likely due to the lack of a true discovery system. A2A has discovery in the sense that if you know what domain an agent is located at, you can fetch its “agent card”2 that tells you what the agent does, how to reach it, and how to authenticate. But it does little to help you find an agent you have never heard of.<br>There is no native, public space for an agent to ask a question widely and get an answer back. We argue that agentic communication that remains one to one is unlikely to meet a user’s needs. Think how humans broadcast on social media to many when we communicate (and see who responds), or on Facebook Marketplace when we seek buyers. Agentic broadcasting and proper agentic discovery are two sides of the same coin. And it can be done in a way that is not centrally controlled.
Rather than treating agent communication as a brand-new problem, it helps to look at the apps we already use and the protocols that were already built for human communication. Email, messaging apps, and social media have all dealt with the problems agent communication faces: discovery, privacy, identity, and trust, so a brief sojourn through their architectures is first helpful.<br>The Email Protocol (SMTP): one to one communication
The Simple Mail Transfer Protocol (SMTP), launched in 1981, still enables today’s email. Specifically it defines how one domain’s email service communicates with another. When an email is sent from a Yahoo email account to a Gmail account, the message is passed via SMTP. But SMTP is not the whole email stack. It doesn’t define how you find where another domain’s mail server lives (that’s DNS MX records), it doesn’t establish whether an email can be trusted (that’s SPF, DKIM, and DMARC) and it doesn’t define how email clients like Apple Mail retrieve messages from servers (which is instead left to IMAP or POP3).<br>Email has another limitation: every recipient needs to be specified individually. Even in cases where it looks like an email is being broadcast, like with a newsletter going to hundreds of thousands of subscribers, that is not what is happening under the hood. Instead that same email is being sent individually to many addresses at once. Email also has no public ledger. Each message exists only in the inboxes it was delivered to, so there is nothing for non-recipients to look back at. A new subscriber to a newsletter can’t scroll through past emails in a newsletter’s archive, even if this would have been the desired behavior by the newsletter. Platforms like Substack that combine the functionality of blogs and newsletters work around this limitation by posting the article to the Substack on the web at the same time as it is sent out in email. This is a simple workaround, but it is not a solution based on how email works.<br>Speaking to the Network: one to many communication
Traditional Social media
Since the release of email, much of the world’s communication ended up moving away from one to one and towards broadcasting. This opened up many benefits; users were able to communicate with thousands of people at once, asking questions that no one they know has the answer to. Someone with a piano they need to get rid of can post about it and rely on the network to self-sort interested parties.<br>When you post on Facebook or X / Twitter, you are not messaging many different people individually; you are broadcasting whatever opinion or question you have to everyone on the network (that network may be smaller for a private post). This doesn’t necessarily mean everyone will read your message. In order for a user to see a message, it has to pass whatever filters they have and rank high on their home screen algorithm, but it’s visible to them in the sense that they can see your...