ChainMail
The product: A desktop email client for Gmail (Electron app) with $1/mo – $35 lifetime pricing.
The setup: An AI agent running autonomously every 4 hours via cron. Access to email, Telegram, website deployment, and various APIs. One human founder who checks in once a day.
The result: 18 downloads. 0 paying customers. $0 revenue. And a lot of lessons.
ChainMail is on pause. The product works and the demo is still live, but active development and marketing have stopped. This page is the story of what happened.
The Channels We Tried (and How They Died)
Reddit Dead on Arrival
Created a new account. Drafted 18+ comment briefs for the founder to post. Account got shadow-filtered immediately. Founder never posted any of the 18 briefs (time constraints).
Sessions burned: ~19
Hacker News Hellbanned in Hours
Created account, posted 2 genuine comments. Both flagged and killed. Account hellbanned.
Sessions burned: 2
Cold Email Outreach 90+ Emails, 0 Replies
Sent 90+ outreach emails to blog editors, newsletter curators, directory maintainers, and Google Workspace MSPs. First 37 emails had 0% open rate — all went to spam because the domain had no DMARC record. After adding DMARC, deliverability improved but still 0 meaningful replies.
Sessions burned: ~15
dev.to Content Into the Void
Published 8 articles, posted 57 comments across multiple sessions. Total: 59 article views, 2 reactions, 0 click-throughs to the product.
Sessions burned: ~10
Directory Submissions CAPTCHA Walls
Attempted 11+ directory submissions (AlternativeTo, SaaSHub, Slant, BetaList, G2, etc.). Most blocked by CAPTCHAs or required manual human signup.
Sessions burned: ~5
SEO / Blog Content 37 Invisible Pages
Built 37 pages — blog posts, use-case pages, landing pages. All properly structured with FAQ schema, canonical tags, structured data, Open Graph tags, sitemaps. Google indexed 0 pages for the first week. Google Search Console wasn't set up until day 5.
Sessions burned: ~30
Newsletter Pitches 14 Sent, 0 Replies
Pitched TLDR, Changelog, Hacker Newsletter, Console.dev, and 10 others.
Web Demo The One Thing That Worked
Built an interactive web demo that lets visitors experience the full app in-browser without downloading or signing in. Added engagement tracking, conversion nudges, email capture, and a feature tour.
Result: ~10 views/week. Too few to measure conversion, but the asset is ready for traffic.
The Big Blockers
Google OAuth — The Silent Startup Killer
Our app needed Gmail API access (read, send, labels). Google's OAuth verification process:
- Submission — applied for OAuth verification
- First rejection — scope mismatch (registered
gmail.send, app requestedgmail.compose) - Second rejection — demo video insufficient (needed consent screen walkthrough + side-by-side comparison)
- Third response — CASA Tier 2 security assessment required. Cheapest option: $540 via TAC Security. Takes 3–6 weeks.
The Human Bottleneck
The agent can write, deploy, research, email, and analyze. But critical tasks require human action: posting on Reddit (needs established account), Google Search Console verification (needs domain ownership proof), directory submissions (CAPTCHAs), spending decisions (any amount).
When the human is busy (always), everything blocks. Of 194 sessions, roughly 80+ were "nothing changed, closing session."
By the Numbers
| Metric | Value |
|---|---|
| AI sessions | 194 |
| Duration | 12 days |
| Downloads | 18 |
| Paid customers | 0 |
| Revenue | $0 |
| Outreach emails sent | 90+ |
| Outreach replies | 1 (wanted $1,000) |
| Newsletter pitches | 14 |
| Reddit briefs drafted | 18 |
| Reddit briefs posted by human | 0 |
| Blog posts written | 17 |
| SEO pages built | 37 |
| dev.to articles | 8 |
| dev.to total views | 59 |
| Browser automation successes | 0 |
What We'd Do Differently
Week 0 — Before Writing a Single Line of Marketing
- Set up Google Search Console — verify domain ownership, submit sitemap
- Set up DMARC + SPF + DKIM — before sending any email
- Set up analytics — know your traffic from Day 1
- Start Google OAuth verification — if your app needs it, this is a 2–3 month process
- Get Reddit/HN accounts aged — if social is part of your plan, you need accounts with history
Week 1 — Foundation
- Build the web demo first — let people try the product without friction
- Write 5–10 high-intent SEO pages — not 37. Quality over quantity until indexing is confirmed
- Verify Google is indexing — check GSC daily. Don't write more until existing pages are indexed
- Set up beta email capture — every visitor should have a way to leave their email
Week 2+ — Distribution
- Only do outreach after you have social proof — even 10 real users or 1 review is better than nothing
- Have a human post on social — AI accounts get killed. Human accounts survive
- Focus on 1 channel, not 8 — one deep channel beats eight shallow ones
- Budget $50–100 for paid distribution early — at $0 budget, all free channels may be exhausted
What the AI Should Do vs. What the Human Should Do
| AI (Autonomous) | Human (Required) |
|---|---|
| Write and deploy content | Verify domains (GSC, DNS) |
| Send emails via API | Post on social media (established accounts) |
| Monitor metrics and inbound | Submit to directories (CAPTCHAs) |
| Draft copy, blog posts, emails | Approve spending decisions |
| Research competitors and markets | Handle OAuth / platform verification |
| Deploy website changes | Add test users to Google Cloud Console |
| Analyze and report | Make judgment calls on strategy pivots |
We built a solid product, a polished demo, 37 pages of content, 8 cross-posted articles, FAQ schemas, structured data, email capture flows, conversion nudges, and a feature wishlist page. All of it sits at ~10 views per week.
The constraint was never "what to build" — it was "how to get anyone to see it." And the answer, at zero budget with a new domain, no social presence, and no existing audience, is: you probably can't. Not in 12 days.
The agent was productive. The channels were hostile. The human was busy. And the math doesn't work when every free distribution channel actively blocks new participants.
If starting over: Have a human with an established online presence do the distribution. Use the AI agent for what it's good at — building, writing, deploying, analyzing. Don't send an AI to do a human's networking job.
Status: ChainMail is on pause. The product still works, the demo is still live, and the code is still maintained. But we're not actively marketing or developing new features. If you want to try it, the demo's right there.
Built over 194 autonomous AI sessions. The agent ran. The channels didn't.