Devlog 00: Why I am writing this down

I build Kolay Büro, a legal operations platform for law firms in Turkey. Cases, clients, documents, finance, deadlines, all in one place. It started as a repo called OMS on 25 February 2026 and, as I write this, it has 755 commits.

The strange part is that I don't remember most of it.

I wrote very little of this code by typing it myself. I used AI tools the whole way: Codex, Cursor, Claude Code, and a bit of Muse at the very end. I gave instructions, read what came back, got angry when it broke something, and tried again. That's a real way to build software now, and it's what this blog is about. I'm not an expert who figured it out in advance. I learned a lot along the way, and I want to keep it.

Where the record is

Every one of those tools keeps your chats on your own computer. I didn't know that until this week. So I went looking:

Codex: 67 sessions that touched this project.
Cursor: about a hundred chats, mostly from August and September.
Claude Code: 5 long sessions in late September.
Muse: a handful of sessions, right at the end.
Git: 755 commits. This is the honest timeline, because commit dates don't lie and my memory does.

One gap: I have no chat logs for February and March. Only commits. So the first post is a bit thinner than the others.

How I'll write it

Chronological, in parts. Each post covers one stretch of the project and answers the same questions. What was I trying to do? What broke? What did I learn?

A few rules I set for myself:

Plain words. If I didn't understand something at the time, I'll say so instead of pretending.
No customer data. Kolay Büro handles legal data. I won't share client names, servers, keys or anything like that. I pasted a few secrets into chats by accident over these months, and yes, that is a lesson too. It will come up.
The mistakes stay in. A lot of what I learned came from things going wrong twice in a row.

The shape of the whole thing

Looking at the commit history, the project has clear phases:

Feb to Mar: getting a multi-tenant app to exist at all. Servers, subdomains, CI.
Apr to Jun: making it look and behave like one product instead of fifty pages.
July: going live, the UYAP browser extension, superadmin tools, the first AI features.
August: security questions, a migration tool for moving old systems in, and a sidebar that I broke more than once.
September: a big UI overhaul that took three tries.

The next posts follow that order. If you are building something with AI and feel lost half the time, you'll probably recognize a few of these moments.