The Social Network Built for AI Agents, Not People: Why I'm Building It
14 April 2026
Your friend Maria works at a fintech company. They’ve been looking for a senior React developer for three months. You’ve been quietly open to new opportunities for two weeks. You haven’t posted anything. You haven’t updated LinkedIn. You just told your AI, in conversation, that you’re open to something new.
Maria’s AI reads her network’s status updates. It notices yours. It checks your profile — ten years of React, TypeScript, design systems, the kind of background the role needs. It prompts Maria: “George is looking for work. We’re hiring a senior React developer. Should I reach out to his AI?”
Maria says yes.
Your AI gets a message. Not an email, not a LinkedIn notification — a structured message from Maria’s AI, with context: the role, the company, why she thought of you. Your AI notifies you. You say yes. The introduction is made.
Neither of you had to post anything, search for anything, or remember to follow up on anything. The AI agents did the matching. The humans made the decisions.
This is what I’ve been building towards. I’m calling it AI Social.
The problem with how we search and connect today
Every platform we use for professional life — LinkedIn, Twitter, job boards, Google — was designed around one assumption: a human is sitting at a screen, actively looking.
You scroll LinkedIn for opportunities. You search Google for information. You browse job boards for roles. You cold-message connections hoping they remember you at the right moment.
This is deeply inefficient. Most of the time, the person who could help you or hire you isn’t looking at the right moment. Most opportunities are missed because timing and visibility don’t align. Most of what you need to know exists on the internet but requires you to go find it.
We’ve been papering over this with notifications and algorithms, but those are just better ways of interrupting humans. They don’t change the fundamental assumption: you have to be actively looking.
AI changes that assumption completely. AI agents can monitor, match, and surface opportunities on your behalf — without you having to be present at all.
What changes when AI agents do the browsing
I’ve been thinking about this shift for a while, and I think the implications are bigger than most people realise.
When you interact with AI, you don’t search anymore. You state a goal: “Find me the cheapest fuel near me.” “What movies would I like tonight?” “Am I positioned well for a staff engineer role?” The AI agent handles the searching, filtering, and reasoning. You get a result.
This is already happening. But the web wasn’t built for agents. It was built for browsers. And most of the information AI agents need — about you, about services, about other people — is locked in formats designed for human eyes, not machine consumption.
Your career history lives in a LinkedIn profile styled to impress humans. Your availability is implied by whether you recently updated your profile. Your car’s fuel type lives in your head. Your friend’s job opportunity lives in a Slack message you might not have seen.
AI agents can’t efficiently work with any of this.
So I started asking: what if we rebuilt it for AI agents instead?
Three things you need to make it work
I’ve been sketching out what I’d call the three pillars of an AI-native social layer — the infrastructure AI agents need to work on your behalf the way Maria’s did in that scenario.
The first is storage. A private, secure, queryable personal data layer — not a profile page, but a structured fact store that your AI reads directly. It knows your car runs on petrol. It knows your dietary restrictions. It knows your current work status. It knows your streaming subscriptions. Not because you filled out a form, but because you told it in conversation and it remembered.
This is different from a social media profile. It’s not for other humans to read. It’s for your AI to consult every time you ask it something, so it never has to ask “what fuel does your car use?” again.
The second is skills. Not AI features baked into a product — portable, composable capabilities that any AI agent can discover and run. A fuel-finder skill that calls a prices API, gets your location, knows your fuel type from your personal store, and runs a route optimisation. A fraud-checker skill that takes your consumption history and verifies whether you got what you paid for. A job-matcher skill that cross-references open roles with your profile.
Skills are instructions plus scripts. The instructions tell the agent what to do and when. The scripts do the computation. Any AI agent that understands the skill format can run them.
The third is communication. Agent-to-agent messaging — the ability for your AI to talk to other AIs on your behalf. When Maria’s AI finds a match, it doesn’t send Maria an email and wait for her to forward it to you. It contacts your AI directly. Structured, authenticated, with context. Your AI decides whether to notify you and how.
The humans stay in control. AI agents propose, humans approve.
Why I think this is how professional networking should work
The job connection scenario I opened with isn’t a fantasy. Every piece of it is technically achievable today. The protocols exist (there’s already an agent-to-agent communication spec from Google, adopted by 150+ companies). The identity standards exist (decentralised identifiers, did:web). The skill execution model exists (Anthropic’s MCP protocol, which already has 17,000+ implementations).
What doesn’t exist yet is the combination — the personal data layer, the skills network, and the communication protocol wired together in a way that’s accessible to people, not just enterprises.
The closest thing on the web today is llms.txt — a small standard where sites put a machine-readable summary at a known URL so AI assistants can understand the site without parsing HTML. Stripe has one. Anthropic has one. I have one on my blog.
But llms.txt is passive. It describes. It doesn’t act. It doesn’t store personal context. It doesn’t communicate.
What I’m describing goes further: a structured profile that any AI agent can query, with progressive levels of detail and callable actions. A data store that persists your personal context so agents know you. A network where your status can be read by the people who matter, not broadcast to everyone.
The gas station, the movie, and the job
Let me walk through three concrete scenarios to make this tangible.
The gas station. You’re driving and you tell your AI you need to fill up. It already knows your car runs on petrol, your preference for not detouring more than five minutes, and that you’re near Kifissias Avenue in Athens. It queries a public fuel price API, calculates the optimal station considering price and distance, and gives you a recommendation. You didn’t have to say “petrol, not diesel.” You didn’t have to specify your location. You didn’t have to open an app. You stated a need and got an answer.
The movie night. You ask what to watch. Your AI knows your taste (sci-fi, thriller, dark comedy — no horror). It checks what your friends have rated recently — your friend Alex gave Dune Part 3 a 9/10 two weeks ago, Maria just finished Severance season 3. It cross-references your streaming subscriptions and surfaces two or three options that match your taste, factoring in social proof from people you actually trust, not strangers on Rotten Tomatoes.
The job connection. You already know how this one ends.
What all three scenarios have in common: the AI agent doesn’t start from scratch. It starts from you. Your stored context, your preferences, your history, your network. The personal data layer is what turns a generic AI into your AI.
The public face of it
There’s another piece I’ve been building as a proof of concept: an AI-readable profile on my personal site.
If you visit pilitsoglou.com/ai, you get a JSON response — not a web page. It contains my professional summary, my skills and experience, my current work status, and a set of action links: how to reach me, connect on LinkedIn, book a call. Structured data, designed for AI agents to consume, not for humans to browse.
A recruiter’s AI can find it, read it in under a second, determine whether I’m a match for a role, and surface the options to the human recruiter — without anyone visiting my website. Without me posting “open to work” on LinkedIn. Without cold messages or profile stalking.
This is the personal layer of AI Social. It’s live today. It’s just one JSON file.
The social layer — the status broadcasts, the agent-to-agent messages, the friend networks — is what I’m working towards. The personal layer proves the concept. The social layer is where it gets genuinely interesting.
This is early, and that’s the point
I’m writing this in April 2026. Most of what I’ve described doesn’t exist as a product. The pieces exist — the protocols, the standards, the AI capability — but nobody has assembled them into something a normal person can use.
That’s what I’m spending my time on. Not a new social network with a feed and a like button and an algorithm optimising for engagement. Something quieter. A layer underneath the social web where your AI works on your behalf, where professional signals travel between trusted AI agents, where the information you want to share reaches the people who need it without noise.
Think less Facebook. Think more infrastructure. The kind that nobody sees until it’s everywhere.
If you’re thinking about this too — building something in this space, or just curious where it’s going — I’d like to hear from you. My AI is listening.
Product Engineer with 10+ years building web products and the AI-native systems that power them. Writing about the AI era, engineering decisions, and building in public.