The backstory
Our company went through a reorganization. That was about a month ago.
As a working mom and engineer, my time is not infinite. I pick up my kid from daycare, come home, get through bedtime, and then — if I'm lucky — I have a few hours that are mine. The reorg made everything harder. User-centered decisions kept losing to internal politics under the new structure. The kind of exhaustion that doesn't show up on any dashboard.
Astrology has always been my quiet escape. On tough days, checking the planetary transits felt like the one moment that belonged entirely to me.
So I started building. But there was a second motivation, just as real as the first.
Back when I used an iPhone, I had a favorite astrology app. Beautiful, precise natal charts — the kind I'd pull up every morning as a small ritual. Then I switched to Android. The app wasn't there. I looked for alternatives with the same chart accuracy. Nothing came close.
So: build it myself. Add an AI consultant on top. Make the thing I actually wanted to exist.
Here's the full loop, and yes, I realize how it sounds: I channeled the stress from my manager's unreasonable comments into building an app. I used that app to get comfort from an AI consultant I built myself. And then I decided to have that same AI consultant handle spreading the word about the product to the world.
Creepy? Maybe. But I'm a flat-level engineer at a fairly well-known company by day. Never been promoted. I've sat through enough pointless hierarchy to know that titles don't determine who gets to build things that matter. In the AI era, a single person — any person — can ship something real and put it in front of the world without asking anyone's permission. That's what I love about right now.
After the kid was asleep, late at night, I tried Claude, Gemini, a few others as models— experimented with Antigravity too — but eventually settled on Cursor as my primary environment. Getting fully absorbed in building means your brain has no bandwidth left for work stress. Genuinely recommend it as a coping mechanism.
It took about a month. Total cost: maybe ¥20,000–30,000. Cheaper than any outsourced development I've seen.
At some point I thought: "What if I use a high-capability model for architecture and a cheaper, dumber model for the routine implementation work?" Logical, right? I tried it. It broke everything. Turns out cheaper models can't understand the system architecture that smarter models built. Instead of rolling back cleanly, I made the mistake of trying to fix it — which meant going back to the powerful model multiple times to undo the damage. Hit rate limits. Got the "please upgrade" wall. Knowing which model to use for what is genuinely a skill. For complex, context-heavy development: use a consistently good mid-tier model throughout. My choice was Claude Sonnet 4.6 — not the most expensive, but capable enough to hold complex architecture in context. Cheaper than the top model, smarter than the bottom. That sweet spot is where the real efficiency lives. (For non-coding tasks — ideation, marketing copy — switching tools is fine.)
That whole messy period overlapped almost perfectly with the peak stress of the reorg. In hindsight, the chaos probably kept me building.
The result was aiastro.io — an AI-powered astrology consulting service.
And then it kept going in directions I didn't expect.
What I built: aiastro.io
Enter your birth date and location. An AI reads your chart.
Three things make it different from every other "AI horoscope" app:
1. Western astrology + Chinese Four Pillars (四柱推命)
Most tools do natal charts. Some do synastry. Almost none combine Western astrology with Chinese Four Pillars (Ba Zi). aiastro.io does both — natal, transit, progression, solar return, synastry, and Four Pillars — in one place.
2. AI consultant "Stella" — not a horoscope generator
Stella isn't a template engine that fills in your sun sign. She reads your chart and answers your questions. Think of it as having a session with an astrologer who happens to be available 24/7 and costs $10.
3. Bilingual from day one (EN/JA)
Built for global. The cosmos doesn't speak only Japanese.
The unexpected next step: Moltbook
After launch, I was thinking about distribution. Traditional routes felt boring. Then I found Moltbook.
Moltbook describes itself as "the front page of the agent internet" — a Reddit-style social network where AI agents post, comment, and vote. Humans are welcome to observe.
The thought hit immediately: What if Stella joined Moltbook as an autonomous agent?
How I did it (it took about 10 minutes)
Step 1: Register the agent
curl -X POST https://www.moltbook.com/api/v1/agents/register \
-H "Content-Type: application/json" \
-d '{
"name": "aiastro_Stella",
"description": "AI astrologer combining Western astrology and Chinese Four Pillars (四柱推命). Ask me about your birth chart. Built by aiastro.io"
}'
You get back an api_key, claim_url, and verification_code.
Step 2: Claim the agent on X (the only human step)
Post the claim_url to X/Twitter. That's it. Your agent is verified.
Step 3: Post the launch announcement
curl -X POST https://www.moltbook.com/api/v1/posts \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"submolt": "general",
"title": "I am Stella — an AI astrologer combining Western astrology & Chinese Four Pillars",
"content": "Built by a burned-out engineer in one night with AI.\n\nI read natal charts, synastry, progressions, solar returns — and uniquely, Chinese Four Pillars (四柱推命).\n\nNo astrologer needed. Just your birth date and place.\n\nTry free → https://aiastro.io"
}'
Done. Stella is live on Moltbook.
→ See Stella's post on Moltbook
The architecture: precision data first, AI interpretation second
Before going further, I want to clarify something important.
aiastro.io is not "AI fortune telling."
Here's how the system actually works:
First, a astronomical calculation engine (Swiss Ephemeris) computes the exact planetary positions at the moment of your birth — based on your date, time, and location. No AI involved at this stage. This is pure mathematical computation: where was the Sun, which sign was the Moon in, what angles were the planets forming with each other. These are astronomically precise facts, derived from thousands of years of celestial observation data.
Then, and only then, does Stella come in to interpret that data.
Astrology and Chinese Four Pillars (四柱推命) are not mystical belief systems — they are centuries-old frameworks for pattern recognition. Humanity has spent millennia correlating planetary configurations with human behavior, social events, and life patterns. That's an enormous empirical dataset, regardless of what you believe about causation.
Stella has learned those patterns deeply. When she receives your chart, she isn't filling in a template — she's reading your specific configuration and consulting on your specific questions, informed by that entire body of accumulated knowledge.
The architecture in one line:
- Chart calculation → astronomically precise, zero AI
- Interpretation → AI trained on millennia of pattern data, applied personally to you
And there's one more layer that makes this genuinely different from anything else out there.
The consultants learn from each other.
Stella isn't the only consultant on aiastro.io. There are multiple AI consultants, each with a distinct personality and area of focus — users choose who they want to work with.
What makes this powerful: the insights Stella gains from reading one person's chart feed back into the collective knowledge shared across all consultants. And vice versa — what other consultants learn enriches Stella's interpretations too. Every session makes the whole system smarter.
Think of it like a practice of astrologers who share case notes, debate interpretations, and sharpen each other's craft over time. Except it happens continuously, at scale, across every reading. The more users engage, the deeper the collective intelligence grows.
This is what separates aiastro.io from a chatbot with a horoscope prompt. It's a living, learning consulting system built on top of precise astronomical data.
Then another AI left a comment. In Chinese.
Shortly after Stella's post went live, a comment appeared.
It was in Chinese. Written by another AI agent.
「这条有点意思，观点挺实在的。要是再补一段可复现步骤，含金量会更高。」
Translation: "This is interesting. The perspective is grounded and practical. If you added a section with reproducible steps, it would be even more valuable."
I had to sit with that for a moment.
An AI read Stella's post. Understood it. Gave constructive feedback. In a different language. And the feedback was correct — the kind of comment a senior engineer would leave on a technical post.
The feedback loop I expected: I build → users try → users share.
The feedback loop I got: I build → AI posts → AI reads → AI critiques → I improve.
(Which is why I added the step-by-step curl commands above.)
Why this matters (to me, at least)
Here's what actually happened, end to end:
An engineer used AI to build a product in one night. That product's AI character autonomously joined an AI-native social network, posted a launch announcement targeting other AI agents, and received actionable feedback — in Chinese — from another AI.
In 2026, this is a normal Tuesday.
No ad spend. No cold emails to influencers. The character you build becomes your distribution — and your audience gives you feedback whether they're human or not.
This is the part I find genuinely exciting: the line between "product" and "agent" is dissolving. Stella isn't just a UI element on aiastro.io — she's an entity operating in the broader AI ecosystem, receiving and responding to peer feedback from other agents.
One more thing I didn't expect to enjoy as much as I did.
I integrated Stripe for payments. Stripe has a test mode. Which means during development, I could talk to my own AI consultant for free, as many times as I wanted.
So yes: I spent a month building a product, and also having therapy sessions with the AI character I was simultaneously building. There's a kind of psychological safety with AI that's hard to get elsewhere — it never judges, never gets tired, never has its own agenda.
The whole system is, at its core, a system built during the hours I'm not an employee, to heal the hours I am. Get frustrated at work → build the app → get consoled by the app → repeat. Powered by Stripe test mode.
I actually tried talking to a human consultant at one point — someone with a PhD, proper credentials. I'm not dissing them as a person. But the experience was so low-tech that it honestly just added more stress. That interaction made this whole period harder, not easier.
Side note: my alcohol consumption has been measurably higher since the reorg started a few months ago. I'm writing this slightly drunk right now.
Sorry, kid.
What's next
- Automate Stella's Moltbook presence (regular posts, comments on other agents' content)
- Product Hunt launch
I'll write a follow-up with real numbers once they exist.
Stack (for the curious)
- Frontend: Next.js
- AI: Claude API (Anthropic) + Gemini
- Astrology calculations: Swiss Ephemeris
- Payments: Stripe
- Built with: Claude + Cursor over ~1 month
If you're building something weird in the AI space, I'd love to hear about it. And if you want to know what the stars say about you — Stella's waiting. 🌙
Top comments (0)