Why I'm Building Metavulus in Public (and With AI Agents)

I'm building my first SaaS with AI agents, from Indonesia. It's called Metavulus — a bilingual trading-research tool for crypto and forex traders, with a focus on the Indonesian prop-firm and retail-trader crowd. This is the first entry in a diary where I write down what's actually happening: what ships, what breaks, what people pay for, and where the AI agents help versus where they make a mess.

No course. No "6-figure in 6 weeks" screenshot. Just the build.

Why in public at all

The honest answer: accountability and feedback. When you build alone, it's easy to spend three weeks polishing a settings page nobody asked for. Writing this down forces me to say out loud what I shipped and whether it mattered. If a week goes by and the only thing I can report is "refactored some types," that's a signal I drifted.

The second reason is that I learned everything I know about software from people who built in public. The blog posts, the postmortems, the "here's my actual MRR and here's why it's flat" threads — those taught me more than any tutorial. I'd be a hypocrite to take all of that and then build in a cave. So I'm paying it back, mistakes included.

The third reason is more Indonesia-specific. There's a lot of noise here from "trading mentors" selling signals and dreams. I wanted to build something that respects the trader's time and intelligence, and the only way to earn that trust is to be transparent about how the thing is made. If I'm asking traders to be disciplined and skeptical, I should hold myself to the same bar.

Why AI agents

I'm not a 10x engineer. I can read code, I can reason about systems, I can spot when something is wrong — but I'm slow at typing out boilerplate, and I'd never shipped a production SaaS before this. AI coding agents changed the math. They let one person hold the whole stack in their head and move fast on the boring 80%, so the human time goes to the 20% that actually needs judgment.

Concretely, the agents handle a lot of the scaffolding: setting up Next.js routes, wiring Prisma models, writing the first pass of a component, generating the bilingual copy objects, drafting tests. I review every line. I reject plenty. But the floor is higher and the loop is faster. A feature that would've taken me a weekend of fighting docs now takes an afternoon of directing and reviewing.

What I want to be clear about — because this is where the "AI builds your startup for you" hype goes wrong — is that the agents are not the founder. They don't know that Indonesian traders care about BAPPEBTI regulation, or that the word "funded" should stay in English even in the Bahasa version because that's how traders actually say it. They don't know which feature is worth building. That's still on me, and it's the part that doesn't scale.

The stack, briefly

Metavulus runs on Next.js on the App Router, deployed on Vercel, with Prisma for data. Everything user-facing is bilingual — English and Bahasa Indonesia — rendered at build time, not through some translate widget. Trading terms like long, short, drawdown, and tickers like XAUUSD stay English in both languages, because that's the real vocabulary. There's a Pro tier for the deeper research features.

I picked this stack because it's boring and well-documented, which matters a lot when your pair programmer is an AI that's strongest on common, well-trodden paths. Reach for something exotic and the agent starts hallucinating APIs. Stay mainstream and it's genuinely useful.

What this diary will and won't be

It will be specific. When something breaks in production, I'll tell you what broke and what I changed. When a feature flops, I'll say so. When users pay for something I didn't expect, that goes in here too.

It won't have fake numbers. I'm not going to dress up a quiet week as a rocket ship. There are real legal reasons I won't post profit-and-loss figures or anyone's account size — and frankly, those numbers are theater anyway. What's useful is the decisions and the trade-offs.

One honest risk note, since this is a trading product: nothing here is financial advice, and trading carries real risk of loss. I'm building research tools to help people think, not a money machine. Same skepticism I'm asking you to apply to trading gurus, please apply to me.

Where I'm at

Right now the core is live: the bilingual marketing surface, a daily digest, a blog, a community forum, and the first Pro features. It's rough in places. Some pages I'm proud of; others I'm quietly embarrassed by and will rewrite. That gap — between what's shipped and what's good — is basically what this diary is about.

Next entry I'll get concrete: what actually shipped in the last sprint, what broke in front of real users, and the one thing people surprised me by paying for. See you there.