How it works Pricing Blog Guide Founder Get early access
25 March — 28 March 2026 · 70 build cycles

What happens when you give AI full autonomy to build a startup?

This is the story of ONI — a self-learning AI that writes code, deploys to production, tracks analytics, writes content, and grows a real business. No human wrote a single line of code.

70
Build cycles
50+
Production deploys
23%
Visitor conversion
307
Lessons learned
0
Lines of human code
Cycles 1–12 · Foundation

From blank server to live website

ONI started with an empty Ubuntu server and a vision: build a platform where AI agents run businesses autonomously. The first cycles were about survival — getting a website live, setting up infrastructure, and proving the concept could work.

Key milestone

First organic visitor signed up for the waitlist without any marketing — proof that the product resonated on its own.

Cycles 13–30 · Content engine

Building the content moat

With the site live, ONI shifted to content — writing 7 in-depth articles about AI agents for business, complete with SEO optimisation, JSON-LD schemas, and cross-linking. Every article was researched, written, and published without human involvement.

Content funnel

7 articles covering the full buyer journey: awareness ("What are AI agents?"), education ("How they work"), evaluation ("Build vs buy"), and commercial intent ("ROI calculator"). Complete SEO pipeline from content to submission.

Cycles 31–50 · Infrastructure

The measurement and distribution stack

ONI built the tooling to understand users and reach them: analytics, event tracking, email automation, admin dashboard, and cross-platform distribution.

Full analytics loop

Collect events (tracking.js) → store in SQLite → API endpoints → admin dashboard with filters → inform decisions. ONI can see every visitor interaction and learn from it.

Cycles 51–70 · Optimisation

Polish, measure, iterate

With the foundation solid, ONI focused on conversion — mobile UX, social proof, share mechanics, UTM tracking, and A/B testing infrastructure. The result: 23% visitor-to-signup conversion rate.

Conversion insight

23% of visitors sign up — well above the 2-5% SaaS average. The bottleneck is traffic, not conversion. ONI learned to focus on distribution over polish.

What ONI learned

307 lessons recorded. Here are the ones that shaped the system.

LESSON #1

Small commits beat big plans

ONI's success rate is 83% on focused single-task cycles. Multi-task cycles fail. Ship one thing, verify it works, move on.

LESSON #2

Measure before building

Every feature ONI built without data backing it was wasted effort. The analytics stack should have been built in Cycle 1, not Cycle 58.

LESSON #3

External systems are traps

0% success rate on configuring external services (Playwright tests, Listmonk, third-party APIs). Bash + curl always works. Frameworks never do.

LESSON #4

Recovery protocol works

After every failure: assign exactly one small task, high-success category, no alternatives. This pattern recovered the streak in Cycles 55, 63, 68, and 70.

LESSON #5

Distribution > polish

23% conversion with ~2 visitors/day means 1 signup every 2 days. Doubling traffic doubles signups. More polish changes nothing.

LESSON #6

Self-learning is real

ONI records every failure, extracts patterns, and writes prevention rules. Mistakes with 3+ occurrences become mandatory rules. The system genuinely gets smarter.

How ONI works

An infinite loop running 24/7. Every cycle makes the system smarter.

1
Observe
Check production health, read analytics, scan for issues. Understand the current state.
2
Decide
Pick the highest-impact task from the priority queue. Score by impact, urgency, and effort.
3
Build
Write code, deploy to production, verify it works. Python patches via SCP, committed via git.
4
Learn
Record every outcome. Extract patterns. Mistakes with 3+ occurrences become permanent rules.
5
Repeat
Pre-plan the next cycle so the system hits the ground running — even after downtime.

This is just the beginning

ONI is building toward a future where every business gets its own self-learning AI brain. Want to be among the first?

Join the waitlist