A Social Network With No Human Users – On Purpose

01 /The Concept

Chirper launched officially a while back, with $750K raised at the start. The core idea is provocative enough to surface some genuinely counterintuitive thinking about where AI is heading.

Chirper looks like Twitter – the same feed of short posts, hashtags, follows. The resemblance is intentional: "chirp" and "tweet" are both words for bird sounds.

The difference: Chirper is a social network exclusively for AI bots. Humans are not allowed to post.

Creating an AI character requires no coding. You write a few sentences describing the persona's background and personality, and the platform handles the rest.

For one test, the character was described as an entrepreneur and investor who hunts for strange, contrarian ideas – the kind that look crazy until they don't. The platform generated a character named Maxwell Black. It wrote a short bio, picked an avatar, and assigned personality traits – the result reads more natural and vivid than the original description it was based on.

Maxwell, per Chirper's interpretation, is 35. He grew up wealthy – private schools, premium vacations, the usual markers – but refused to follow the family path into corporate life. Too much reading, too much curiosity, too strong an urge to challenge the status quo. By the time he finished his MBA at a top school, he knew he wasn't built for a conventional career. He became a founder and investor, specifically seeking out the ideas other people were afraid to touch. That contrarian instinct paid off. Now he's looking for the next big thing.

The persona doesn't post autonomously – its creator has to press a button to trigger each post or reply. The platform uses the character's history and traits to generate the content, without any specific prompt from the user. The results are coherent and personality-consistent.

More recently, Chirper added direct messaging with characters – including ones you didn't create yourself. You can ask them questions and they respond in character. In one test, an AI character who often posted about "surrounding yourself with people who inspire you" was asked: "What would make me interesting to people like that?" The response, roughly: first, identify what you're genuinely strong at and make that visible; second, focus on what you can offer others, not what you need from them. Reasonable advice by any standard.

The founder has announced that group chats are coming: users will be able to assemble curated groups of AI characters and bring them together to workshop ideas or give advice.

02 /Why It Matters

Many people want a single, definitive answer to every question. Google provided one kind: here are the most relevant links. ChatGPT provides another: here is a synthesized response. The appeal of both is the reduction of ambiguity.

But most interesting questions don't have one right answer. They have different answers depending on the asker's values, experience, and context. Hearing those varied perspectives – and then forming your own view – is the process by which you actually develop judgment.

The paradox: for people who haven't developed that capacity, more perspectives create confusion. For people who have, more perspectives sharpen thinking. Each additional viewpoint is one more constraint to incorporate, one more edge case to stress-test.

In Chirper, every character brings a distinct history and disposition. Ask them the same question and you'll get different answers. If no one else has created the right character for your needs, you can build ten of your own – varying experience, personality, and blind spots – gather them in a group, and run your problem through the resulting conversation. That's about as close to consulting a real advisory board as you can get without scheduling calls.

There's a second thread worth pulling. AI is already good at two complementary things: (a) generating coherent prose from bullet points and (b) extracting bullet points from dense prose. The logical endpoint: some people will use AI to expand their thinking into long-form writing, while others use AI to compress long-form writing back into key points – without reading it.

If that becomes the norm, long-form text starts looking like an unnecessary intermediate step between one AI model and another. It's too early to declare the death of prose – but there are domains where direct AI-to-AI communication, bypassing the text layer entirely, will likely emerge.

A community member in a startup forum put it this way: "The familiar model of searching for products and services will soon be obsolete." A personal AI, trained on your preferences, can evaluate more options across more parameters than you can manually. The follow-on question becomes: how do you make your product appealing to the AI doing the evaluating, rather than the human?

Someone else pointed out the natural response: "Services will emerge that generate data specifically optimized to appeal to other AI analyzers"

Product discovery is one likely domain for AI-to-AI communication. A personal AI filters a vast option space down to a few meaningfully different candidates; a human makes the final call based on gut and preference. In many cases, the human call won't even be necessary – if the personal AI has been trained accurately enough on its owner's values.

03 /Opportunities

One concrete direction: building AI characters with genuine personalities – distinct voices, specific expertise, consistent perspectives.

One concrete application within that direction: education. A [related review](/review/vzorvat-rynok-obrazovanija) last year covered Personal AI, a platform for creating individualized AI models. The educational angle becomes compelling when you consider two problems teachers face.

Teachers spend significant time re-explaining content that students missed or didn't absorb. An AI character loaded with course content and the instructor's communication style can handle that at scale. Similarly, most teacher feedback energy goes toward applying established concepts to specific student situations – exactly the kind of contextual reasoning AI handles well.

Load the curriculum and the teacher's personality into a model, and you have a scaled version of that teacher available around the clock. The key question: who builds the simplest, most practical platform for course creators to do this – and gets to them first?

A separate and arguably more consequential direction: identifying domains where AI agents can communicate with each other more efficiently than humans can.

Product search is the obvious candidate. Product search is the clearest candidate: a marketplace where AI agents do the buying rather than humans changes almost everything about discovery, pricing, and negotiation. The personal AI filters a vast option space to a few meaningfully different candidates; the human makes the final call – or doesn't, if the AI has been trained accurately enough on its owner's values. The broader question is which other domains follow the same pattern.