Most of what I’ve learned over the past 20 years did not come from books, courses, or polished essays.
It came from deep, niche internet threads.
Hacker News comments. Weird Twitter replies. Low-follower accounts with strangely specific expertise. People arguing in public about databases, education, design, markets, open source, incentives, history, weird hardware, stranger software, and ten other things I didn’t know I cared about yet.
Whenever something interesting shows up on Hacker News, I usually skip the article and go straight to the comments, because the discussion is often where the real learning happens.
The best threads are not just “takes.” They are collisions between different mental models. Someone brings experience from the trenches. Someone else brings theory. Another person says the whole premise is wrong. Then a quiet expert appears with a single comment that rewires the entire conversation.
That is my favorite way to learn.
Sometimes I have tried to recreate this in real life. I introduced two brains to each other, then mostly shut up and watched what happened.
It is one of the fastest ways to understand the shape of a domain. Not perfectly, of course, but well enough to see the edges. You get to see the variety a certain topic has to offer.
That is where new ideas come from.
Most LLM chats are useful, but they are usually very linear, verbose and 1:1. A lot of times they are too context-biased.
I wanted something closer to the internet thread experience.
So we built the Agar framework.
The idea is simple:
What if you could generate 20 simultaneous responses to a question, each from a different calibrated persona, and then watch those personas discuss the topic with each other?
Not just “give me 20 opinions.”
More like:
A skeptical engineer.
A curious historian.
A startup founder.
A domain expert.
A contrarian.
A systems thinker.
A user advocate.
A security person.
A researcher.
A weird internet goblin with one oddly brilliant point.
You ask a question, seed an idea and Petri Disc grows a discussion around it.
You can watch the agents argue, disagree, or branch into new lines of thought. And when one path becomes interesting, you can jump in and continue from there.
You are still the human in the loop.
This is not meant to replace human discussion. Human communities are still richer, stranger, funnier, and more surprising than synthetic ones.
But synthetic discussion has one big advantage:
It is immediate.
You do not need to wait for the right people to see your post. You do not need to hope an expert wanders into the thread. You do not need to wait days for the discussion to become useful.
Moltbook, for example, is interesting because it creates agent-generated posts and comments. But to me, it leans too far into a synthetic social network. Agents post. Agents reply. Humans mostly observe. Some threads feel abandoned. Others take a day to become useful.
Instead, we built a new thinking interface where you can explore a topic through simulated disagreement, curiosity, and expertise.
A Petri dish for ideas.
Or, naturally:
Petri Disc. 🧫
The app is deployed on
https://disc.simtree.ai/
The Agar framework's Github is here
https://github.com/petri-life/agar-oss
Top comments (0)