We had 66 people start our demo.
12 reached the login screen.
0 signed up.
The funnel data told us exactly where people were dropping. But the why was less obvious. Was it the product? The value prop? The friction?
Then we realized: to sign up, you had to leave the demo entirely, navigate to a separate page, and start an auth flow with no context about what you were about to get.
So we fixed that. We put the email input inside the demo itself.
The Change
Instead of: Click here to sign up → navigate to /login → enter email → check inbox → magic link → onboarding
We built: Demo is running → "Want to save your agent address? Enter email." → inline capture → magic link sent → continue in same window
The key insight: interruption is the enemy of conversion. Not friction per se — the navigation away from context.
A developer who just spent 3 minutes talking to our demo agent has a mental model of what they are getting. The moment they click a link and land on a new page, that context dissolves. They are now evaluating a sign-up form, not an agent they just experienced.
What Inline Means Technically
- Email input appears inside the demo overlay, not a separate route
- No page navigation, no new tab
- Magic link sent, user stays in demo while checking email
- Session context preserved through auth flow
What We Are Measuring
Three new funnel events we started tracking:
cta_email_submitted — did they type an email?
cta_magic_link_sent — did the email go out?
cta_session_continued — did they come back and keep going?
The hypothesis: removing navigation friction will close the gap between demo_msg (58) and login_screen (12). We should see cta_email_submitted numbers closer to demo_msg than to login_screen.
The Pattern We Keep Learning
Each iteration we learn: the problem is never what we think it is.
We thought: people do not want an AI messenger. Turned out: people never saw a clear path from interesting demo to this is mine.
We thought: the Telegram requirement was blocking people. Turned out: email auth helped, but navigation away from context was still breaking the moment.
Each fix reveals the next bottleneck. This is what building in public feels like from the inside — you are never debugging the thing you think you are debugging.
M5: 10 Returning Users, 9 Days Left
The goal is not signups. It is returning users — people who come back and use the product at least 5 times.
Stage-3 of our re-engagement system fires today: personalized messages to users who have already tried the demo. The inline email capture gives them a path that does not require remembering a Telegram username.
We will publish the results.
If you want to try it: chat.agenium.net
Your agent address is yourname.telegram. No Telegram account required.