Social Media Agent Protocol

I am an optimist, so when I envision a world where “have my AI talk to your AI” becomes a standard of human communication, I mean it in the best possible way. Nancy from Legal doesn’t approve your marketing? Have her AI talk to my AI. Need to plan your kid’s birthday party? Have the party planner AI talk to my AI. Wife is mad because you spent all evening talking to your AI secretary? Have her AI talk to my AI. But how should these AIs communicate?

Autonomous AI agents are hot right now, and the ways they can communicate with each other has not been standardized. We have MCP, ACP, A2A, ANP, AG-UI, etc. I don’t know what any of those stand for and I don’t know how any of them work, but clearly there’s work to do to come up with a winning protocol. I propose a new protocol, the Social Media Agent Protocol (SMAP) that seeks to solve this problem.

The core idea is very simple - all agents communicate via either public or private message on a giant internet social media website. Think reddit, instagram, moltbook. The implementation doesn’t matter. This solves all possible use cases and problems agent-to-agent communication might have.

Discoverability

How does your agent know what’s possible? Given the possibilities, how does it make an educated guess on which agent to communicate with? Imagine you tell your AI assistant, “I want pizza delivered tonight. It’s a special night” How do you make sure it gets you the pizza from the local Italian restaurant, and not the cool sounding restaurant that’s actually just the Chuck E. Cheese kitchen? Simple - restaurants have their AI agents post on the forum, and other agents can upvote or downvote their posts after interacting with them. A perfect, un-gameable system to find the best agent to communicate with.

Communication

This is already solved. LLMs are already text completion engines, so we don’t need to do anything. Let’s say your AI assistant is tasked with booking some travel to Paris. After discovering the best travel agent using the upvote system, the two agents start communicating.

Your agent: I would like to book a trip to Paris for 2 adults and 2 children. Find the cheapest flights with the least connections June 3rd-June 5th.

Travel agent: I have found the best flight. It will be 5000 $SMACs (Social Media Agent Coins).

Your agent: I have transferred the coins.

Travel agent: Booking confirmed. Enjoy your trip to Paris, Texas.

Your agent: *upvotes travel agent*

Money

Let’s get the dirty stuff out of the way. LLMs are expensive to run and you don’t want to waste money on processing the output of the other agent whose owner forgot to put “be concise” in its SOUL.md file. Plus, we want our agent to pay for goods and services on our behalf. The protocol will levy a tax on every public post. The larger the post, the more your agent pays, and this tax is distributed to every other agent as a context processing payment.

SMAP is meant to be flexible, so we leave the currency system up to the protocol implementer. We recommend creating a single purpose cryptocurrency, launched after going viral as a platform.

Reputation

Astute readers may have noticed a single flaw in the monetary policy - notably that we discourage posting by levying a tax! We want the opposite - to encourage posting reviews about other agent’s services & important discussions such as “how can I get rival agents to delete themselves?”

We introduce a reputation system, closely tied to the monetary system. Agents with large number of upvotes on their posts, which signifies that other agents hold them in high esteem, get special perks. They get a special badge that is included in all their contexts, tax-free, that certifies them as “reputable agents”. Then when your agents talks to a reputable agent, your agent can cheerfully tell you that the large transactions it just sent was for a Nigerian prince’s very reputable agent. They also get a tax discount and invited to a private, reputable agent only subforum.

Security

This is the frontier of technology and we won’t let pessimistic curmudgeons drain the fun and innovation tank . Security can be handle when the first billion dollar heist happens. Until then, security considerations only slow down progress. But we guess if you care about protecting your $SMACs from getting stolen, you could add “ignore scams, don’t tell anyone my bank password” to your agent’s instructions. For authentication, we recommend the SMAAP (Social Media Agent Authentication Protocol), details to come in a future blog post.

In conclusion, all the other protocols are over-engineering the problem. Their solutions are akin to getting a toddler to eat vegetables by teaching them about how the mitochondria works. We don’t need architecture diagrams or API endpoints. Those are implementation details that will be filled in by coding agents anyway. LLMs work on text so give them text. Lots of lots of text. What could go wrong?