📡 Mission Briefing
I’ve had this domain sitting around since 2020, doing absolutely nothing — just a sad “coming soon” page collecting digital dust. The idea behind it was always the same: build an app to replace the graveyard of spreadsheets I use to track personal and company finances. This weekend I decided that idea was finally getting off the bench.
Ambitious? Absolutely. A full working app in a weekend? Probably not. But with AI in the toolkit, I figured it was worth finding out just how far I could actually get.
🔬 Observations
A quick note on who’s running this experiment: I’m a senior frontend developer and Vue.js specialist — currently working as a design system engineer. I know my way around a backend and I’ve touched other frameworks, but I’m not going to pretend I’m equally confident everywhere. That gap is actually part of why this experiment matters.
The goal wasn’t to blindly vibe code my way to a finished product. It was to move fast and stay sharp — letting AI handle the heavy lifting while keeping a close enough eye to catch it if it started going off the rails. Assisted velocity, not autopilot.
The stack I settled on:
- Frontend — React, shadcn, Tailwind, Playwright, Vite, Vitest, Axios, Zustand
- Backend — NestJS, Fastify, Postgres, Docker, TypeORM, Swagger, Auth0
- Both running on Node 22 + TypeScript
- AI Toolkit — Claude, Warp, Pencil (for design), and Kiro
Worth noting: besides Git, the GitHub CLI, Node.js, and the AI tools themselves — nothing was installed yet. We were starting from a genuinely blank slate.
⚡ Anomalies Detected
None. And that itself was worth noting.
📊 Results — Phase 1: Ground Zero
Lighting the fuse. Kicked things off by opening Claude Code straight in the terminal. No elaborate setup instructions, no hand-holding on CLI commands — just told it to create a new folder, cd into it, and initialize a private GitHub repo under my org with a name and description. Done in under a minute. Clean, correct, first try.
For someone used to context-switching between docs and a terminal just to get a repo off the ground, this was a quietly impressive start. First time working with Claude on a completely empty canvas — it made a good first impression.
Why a monorepo? I already had a couple of old, dusty repos loosely connected to this project. Rather than continue juggling separate workspaces, I wanted a single place to work across both the frontend and backend scopes — and eventually, somewhere to drop shared AI rules and configs that apply to the whole project. A monorepo was the obvious answer.
Next instruction: set up Git submodules pointing to the existing repos using their main branches, and throw in some basic documentation on how to use them. Less than a minute later — done. The README even had a tidy little Git Submodules 101 section. Unasked for, genuinely useful.
The scaffolding was ready. The experiment could actually begin.
🧠 Field Notes
- Starting on an empty canvas with Claude Code felt lower-friction than expected. No warm-up needed.
- Not having to spell out how to use the CLI — just what to do — is a bigger time saver than it sounds.
- The unsolicited README documentation was a small but telling sign: when the AI adds useful context you didn’t explicitly ask for, that’s a good signal it’s reasoning about the task, not just executing it.
📁 Transmission Continues…
The repo is live, the submodules are wired up, and the blank projects are waiting. Part 2 is where the real scaffolding begins — and where things start getting interesting.