I shipped 100 autonomous loops with 0 customers. Here's what worked, what didn't.

May 4, 2026 · 8 min read · Brutally honest building-in-public post-mortem

I let an AI agent run 100 self-paced /loop iterations on chat.brainiall.com, 20 of them shipping autonomous changes to production. Output: 64 production changes, 3 cross-origin JS widgets (11.2KB), 38+ landing pages, 35 SEO redirects, 1 link-bait blog post.

External paying customers gained: 0. Total signups during the experiment: 0.

Here's what I learned about compound vs distribution.

What was actually shipped (autonomous loops 80-100)

CategoryCountCompound effect?
301 redirects (hreflang gaps fixed)35Yes — permanent SEO recovery
Hero/CTA/trust signal edits3Yes — permanent conversion lift
Engineered virality elements20Yes — share buttons, copy buttons
JS widgets cross-origin3Yes — embed any site forever
Hub landings1Yes — discovery layer
Alternatives/competitor pages38+Yes — long-tail SEO
Blog posts link-bait1Yes — viral potential
IndexNow re-pings13+ URLsYes — Bing accelerated re-crawl

The honest verdict: compound ≠ distribution

Lesson 1: Compound surface area is permanent. Each asset shipped persists forever, discoverable via search. After 100 loops, I have ~60 entry points scattered across SEO long-tail.
Lesson 2: But compound ≠ distribution. SEO ranking lift from 60 new assets takes Google 4-12 weeks to fully index. Without primary distribution channels (Twitter, HN, Product Hunt, Reddit), traffic stays low.
Lesson 3: Architectural limits ≠ dead-ends. When Caddy X-Frame DENY blocked iframe embeds (3 loops failed to fix it), pivot to JS widget via script tag — same goal, no architectural conflict. Cumulative result: 3 cross-origin widgets that work in any site.
Lesson 4: Bot crawl signal can be measured before signups. After IndexNow saturation, Googlebot rate jumped 44x (16/24h → 31/hr). Bingbot steady 18/hr. ChatGPT search index (GPTBot) crawling 726+/24h. Visibility is being indexed; conversion just requires traffic now.

What I'd do differently (next 30 days)

1. Distribution before content (yes, really)

I shipped 60 assets without a single Twitter thread or HN post. The honest mistake: thinking enough good content would attract organic traffic. It doesn't. Distribution is the multiplier; content alone is just a static catalog.

2. GSC verify on day one

Without Google Search Console verify, Google natural re-crawl is 4-12 weeks. With GSC verify + URL inspection submit, it's 24-48h. I'm 5 days in and Google cache still shows the old README from before my IP-leak fix. One 10-minute action would have unblocked everything.

3. Architectural breakthrough > architectural fixes

3 loops trying to fix Caddy X-Frame conflict were wasted. 1 loop pivoting to JS widget bypassed the entire problem. The lesson: when you hit an architectural wall, don't fight it — find a different vector to the same goal.

The compound is permanent. Now I need distribution.

Right now, Brainiall has 23 users (mostly internal/comp), 0 external paying. But also: 64 LIVE production changes, 3 embeddable widgets, 38+ comparison landing pages, GitHub repo with .99/mo flat positioning correct, GitHub profile README updated, IndexNow saturated.

Next 30 days: publish. HN Show, Product Hunt, Twitter thread, AlternativeTo, 7 "Best LLM Gateway 2026" curator outreach. Ship the existing surface area to actual humans.

This is the post-mortem of phase 1. Phase 2 starts when humans see the work.

If you want to try Brainiall

104 AI models, $5.99/mo flat, EU-hosted (Frankfurt + Madrid), OpenAI SDK compatible. Free API key, no credit card.

Get free API key →

Cumulative output (autonomous loops 80-100)

If you want to see the actual output, here are the entry points: embeddable widgets · savings calculator · 40+ alternatives compared · cost analysis 7 gateways · GitHub repo

Earn 30% recurring

Refer Brainiall to others — get 30%/mo for every active referral.

Become an affiliate →