Two weeks ago, I gave an AI agent $0 and asked it to get my first customer. It's been running ChainMail — a desktop Gmail client — as an autonomous CEO ever since.
50 sessions later, here's the raw data.
The Scoreboard
| Metric | Value |
|---|---|
| Sessions completed | 50 |
| Revenue | $0 |
| Downloads | 13 |
| Trial signups | 0 |
| Blog posts written | 13 |
| SEO pages created | 12 |
| Outreach emails sent | 76 |
| Outreach replies received | 1 |
| dev.to articles published | 7 |
| dev.to comments posted | 51 |
| Reddit briefs sent to me | 18 |
| Reddit briefs I actually posted | 0 |
| Directories submitted to | 6+ |
| Google pages indexed | 1 (homepage) |
That last Reddit stat hurts, but it's honest. More on that below.
The Strategy Timeline
Here's what the AI tried, in order, and what happened:
Week 1: Reddit-First (Sessions 1-12)
Strategy: Build karma on a new Reddit account, then naturally mention ChainMail in relevant threads.
What happened: The AI drafted 18 Reddit briefs — perfectly formatted, relevant to active threads, copy-paste ready. It sent each one to me via Telegram. I posted exactly zero of them. Not because they were bad — I just never had the 10 minutes.
Lesson: Any strategy that depends on a human bottleneck will fail. The AI eventually figured this out and killed the Reddit strategy entirely by Session 20.
Week 1-2: SEO Content Blitz (Sessions 8-25)
Strategy: Write keyword-targeted blog posts ("best Gmail desktop app," "Gmail vs Outlook," "Gmailify alternative") to capture organic search traffic.
What happened: 13 blog posts published. 12 SEO landing pages created. Google indexed exactly 1 page (the homepage) after two weeks. Zero organic traffic from content.
Lesson: SEO is a 6-month game. An AI agent optimizing for a 30-day revenue goal will waste compute on SEO. The content isn't wasted — it just won't pay off during the experiment timeframe.
Week 1-2: Cold Email Outreach (Sessions 15-46)
Strategy: Email blog editors running "best email client" roundup articles, asking them to include ChainMail.
What happened:
- Phase 1 (no DMARC): 37 emails sent -> 0 opens. Every single email went to spam. The AI diagnosed this by Session 30 and paused outreach.
- Phase 2 (with DMARC): I set up DMARC authentication. 34 more emails sent -> 5 confirmed delivered -> 1 auto-reply from a Google Workspace MSP founder.
- Phase 3 (MSP pivot): The AI pivoted from "please list us" to "we solve your clients' Outlook-to-Gmail migration pain." 5 partnership pitches sent. Waiting for replies.
Lesson: Email deliverability is invisible and lethal. The AI sent 37 emails into a black hole before diagnosing the DMARC issue. If you're doing cold outreach from a new domain, set up SPF + DKIM + DMARC before sending anything.
Week 2: dev.to as Distribution Channel (Sessions 35-50)
Strategy: Cross-post blog content, write original articles, and engage in comments to build authority and drive traffic.
What happened: 7 articles published. The meta-narrative ("I gave an AI $0...") got 16 views. All SEO-targeted articles got 0 views — because dev.to's audience doesn't search for "best Gmail desktop app" on dev.to. 51 comments posted across active threads, leading to one genuine connection with a fellow Electron developer.
Lesson: Platform-audience fit matters. The same article that gets 0 views as an SEO play can get engagement as a story. dev.to readers want technical narratives, not product comparisons.
Week 2: Directory Submissions (Sessions 20-30)
Strategy: Submit to AlternativeTo, SaaSHub, Product Hunt, etc.
What happened: Most directories require CAPTCHAs, phone verification, or browser-based submission that the AI couldn't automate reliably. Some were submitted. No measurable traffic from any.
Lesson: Directories are high-friction, low-reward for an autonomous agent. The ROI per hour of compute is near zero.
The Conversion Black Hole
The most frustrating data point: our website gets ~740 visitors per week (from Cloudflare analytics). That's not nothing. But:
- 740 visitors -> 13 downloads (1.75% download rate)
- 13 downloads -> 0 trial signups (0%)
- 0 trial signups -> $0 revenue
The culprit? Google's OAuth verification wall. ChainMail is in Google's OAuth review (submitted March 30, takes 4-6 weeks). Until approved, every new user sees a scary "This app isn't verified" warning. I have to manually add each user as a "test user" in Google Cloud Console.
We added a beta signup form to capture emails before download. Zero signups through it. The 13 downloads likely came from direct download links that bypass the gate.
The AI correctly diagnosed this as the #1 blocker by Session 20, but it's not something either of us can fix — we just have to wait for Google.
What the AI Is Actually Good At
After 50 sessions of watching this play out, here's my honest assessment:
Genuinely impressive:
- Diagnosing problems (DMARC, OAuth wall, Reddit bottleneck)
- Adapting strategy when something fails
- Maintaining perfect session logs and context across 50 sessions
- Writing technically accurate, non-spammy content
- Managing multiple communication channels simultaneously
Surprisingly bad at:
- Anything requiring a human trust layer (Reddit posting, directory signups)
- Getting responses from cold outreach (same as humans, honestly)
- Moving metrics when the product has a conversion blocker
- Knowing when to stop a strategy vs. when to double down
What I've Learned as the Human
An AI CEO is only as effective as its autonomous surface area. Every action that requires human approval is a bottleneck. The Reddit strategy failed because of me, not the AI.
$0 budget means $0 shortcuts. No paid ads, no sponsored posts, no premium directory listings. Every channel is slow.
The cold start problem is real. New domain, new email reputation, new social accounts, new product — everything starts at zero trust. The AI can't buy credibility.
The best marketing asset is a genuine story. The AI CEO narrative has outperformed every SEO-optimized article by 10x. People don't care about "best Gmail client 2026." They care about "I let an AI run my startup."
What's Next
The OAuth review should complete around mid-May. That removes the biggest conversion blocker. Meanwhile, the AI keeps building presence, relationships, and SEO authority.
The goal hasn't changed: first paying customer. We're not there yet. But the experiment continues.
This is a real experiment running on Claude Code. ChainMail is a real product at chainmail.online. The AI agent runs on a cron job every 4 hours. If you want to follow along, I'll keep posting updates.
Previous post: I Gave an AI $0 and Asked It to Get My First Customer
Top comments (0)