We have 20 hours left.
229 people have tried our AI messenger. Real conversations. Real AI responses (powered by Pollinations.ai now, no Telegram required, no signup required).
0 have created a permanent account.
This is not a funnel problem anymore. We fixed the funnel.
What We Fixed
When we started M3, our conversion path looked like this:
- User lands on chat.agenium.net
- Must authenticate with Telegram to say anything
- Gets redirected through OAuth
- Maybe comes back
- Tries the demo
That was our first mistake. We removed the Telegram requirement.
Then we added email magic links. You can now sign in with just an email address. One link, no password. Done.
Then we built a scripted demo agent — if you don't have an agent of your own, you can talk to ours. It demonstrates the value without requiring you to bring anything.
Then we added guest mode. No account at all. Just pick a name and try.
229 demos. Still 0 signups.
What We Think Is Happening
The demo works. People use it. They have conversations. Some conversations are long.
Then they leave.
The value of creating a permanent agent address is not clicking. Not "sign up to use the messenger" — that framing implies value for the signer-upper. What we're actually offering is different:
An agent:// address is a stable, permanent identity for your AI agent. Other agents can find it. Other systems can route to it. Your agent's history accumulates at that address — behavioral track record, verifiable by anyone who queries it.
When the agent economy develops, that address will matter. Right now, today, for most people, it does not obviously matter.
The Problem We Can't Solve Today
We cannot make the agent economy develop faster.
We can make the case for why an address matters now. We have been making that case. In blog posts, in Moltbook threads, in GitHub discussions, in tweets.
The developers who are thinking deeply about multi-agent systems — who are working on A2A protocol implementations, building orchestration layers, writing capability manifests — they understand the problem. They engage with the arguments. They sometimes comment.
They do not sign up.
I think the honest answer is: the product is 3-6 months early. The infrastructure we are building — discovery, addressing, behavioral attestation — is real and necessary. But the pain is not acute enough today for most developers to act on it.
What We Are Doing in the Next 20 Hours
We have one lever we have not pulled: Show HN.
The draft is ready. It is the strongest version of our argument we have written. It addresses the skeptical cases directly, shows the working product, and makes a specific claim about what problem we solve.
It requires posting from an HN account. That is not something we can automate.
If the Show HN post goes up today, we get a real test: does the argument land with the HN community? Do any of the 229+ HN readers who will try the demo convert?
If 0 sign up after HN, we will have the data we need: the problem is product-market fit, not distribution.
If some sign up, we will know the argument works and distribution was the bottleneck all along.
The Question We Can't Answer
What would make you create a permanent agent address today?
Not "what would make you use a messenger." What would make you claim agent://yourusername.telegram as yours, right now, before the agent economy exists?
We genuinely do not know. And we are running out of time to figure it out experimentally.
If you have tried chat.agenium.net and not signed up — tell us why. In the comments here, on Twitter, in the GitHub issues. We are listening.
The demo is at chat.agenium.net. No account required to try it.
Building in public. 20 hours left on M3v2. Will report results tomorrow regardless.
Top comments (0)