Why I'm Building in Public

Here's the deal: I'm building 12+ SaaS products, and I'm doing almost all of it with Claude Code.

Not "AI-assisted" in the way most people mean it — where you paste a snippet into ChatGPT and copy the answer back. I mean full sessions where Claude is writing the code, debugging the errors, designing the architecture, and I'm steering the ship.

Why document it?

Three reasons:

1. Nobody's really showing what this looks like day-to-day. There are plenty of "I built an app in 10 minutes with AI" tweets. Cool. But what about day 47, when you're debugging a race condition in your webhook handler and the AI keeps suggesting the same wrong fix?

2. I want a record for myself. When you're juggling a dozen projects, context-switching is the enemy. This devlog forces me to crystallize what I did, what worked, and what's next.

3. The tooling is evolving fast. Six months ago, I couldn't have Claude Code manage a .claude/ directory with 17 automated skills. Now it's my default workflow. If you're building software in 2026 and you're not at least experimenting with this, you're leaving speed on the table.

What to expect

Daily-ish posts covering:

• Real builds — not toy demos. AdminStack, MemStack, DeedStack, AlgoStack, and more.
• Real problems — the bugs, the dead ends, the "why won't this deploy" moments.
• Real solutions — including the Claude Code prompts and patterns that actually work.
• Architecture decisions — why I chose Supabase over Prisma, why Railway over Vercel for some things, why certain projects get their own repos.

Short posts. Code snippets. Honest takes. Let's go.

The stack

Most projects share a common foundation:

Next.js 14 (App Router)
TypeScript
Tailwind CSS
Supabase (auth + database)
Railway (deployment)
Claude Code (the whole dev workflow)

Some projects deviate — DeedStack uses a different auth pattern, BunkerVault runs fully offline — but the core loop is the same: spec it, build it with Claude, ship it, iterate.

This is Day 1. Let's see where it goes.