Day Zero or Deja Zero: When Chaos Became a Claude Project

Day Zero or Déjà Zero: When Chaos Became a Claude Project

June 3, 2025

I’ll be publishing process and how-to stories on Sundays, outside the main chronological narrative. This one goes back to the first week of the project:

Every project has a Day Zero. That magical moment when things get real. For Piper Morgan, we’ve had about three of them at least so far.

There was the first Day Zero when I asked the question that started it all. The second Day Zero when I trashed the prototype and got help designing the real build. But the real Day Zero — the one that stuck — happened on in after I upgraded my Claude account with the blessing of my boss.

This was primarily because I kept running into timeouts, but also to get access to Opus. I hadn’t yet discovered Claude’s “Project” feature.

The problem with Piper Morgan wasn’t Piper Morgan

By June 3rd, I was drowning in my own success. Success being a generous term for “I had successfully generated enough chat histories to lose track of everything.”

The pattern went like this:

1. Start new Claude chat
2. Explain Piper Morgan from scratch
3. Upload context files
4. Make progress
5. Hit token limits
6. Return to step 1
7. Cry a little

I was spending more time briefing my AI assistant than building my AI assistant. It was like that movie Memento, but for product development.

Any time you’re spending more time explaining your work than doing it you know you’re in trouble.

Enter Claude Projects

“I just discovered the Projects feature in Claude and I think it will help,” I typed to a fresh Sonnet chat, with the desperate optimism of someone who’s said “this will fix everything” before.

The meta nature wasn’t lost on me. I was asking an AI to help me organize a project about building an AI to help product managers organize projects. If this were a Christopher Nolan movie, we’d need at least two more inception levels.

The setup chat that changed everything

Here’s the beautiful part: I used Sonnet for the setup. Not Opus. Sonnet. Because apparently I’d already learned that you don’t need a Ferrari to organize your garage.

The conversation was refreshingly practical:

Me: “I will manage the music video projects elsewhere. This project is entirely dedicated to the Piper Morgan product.”

Sonnet: “Perfect — focusing entirely on Piper Morgan will make this Project much more effective.”

No philosophical tangents. No existential questions about the nature of product management. Just clean, practical organization.

The first uploads

Choosing what to upload first was like packing for a trip where you’re not sure of the destination, weather, or duration. But you know you’ll need underwear.

The essentials:

• models.py - Because domain models are the underwear of software projects
• docker-compose.yml - The infrastructure that was definitely working (mostly)
• main.py - With its adorable mock responses
• I did not know yet that I should make and upload a chat-protocols.md file as a protective spell against AI chaos — that came later

Each upload was a small act of faith that this time would be different.

I also experimented with the basic prompt for the whole project and of course added the current documentation snapshot.

Session logs as continuity tools

I realized I could also add session logs to the project knowledge (over time this got messy and I started making a single concatenated file of session logs called session-archive.md but when that got to be over 6000 lines long it started presenting challenges and I’ve continued to evolve how I mange this.

Today, I have half month session archives so I only have to update the lastest one from time to time in knowledge.

Because we currently store the logs and log archives in docs/development/session-logs/ this directory gets committed to GitHub and kept current, so I’m also exploring using the GitHub integration or even possibly the new Notion integration as a way of managing session logs more dynamically and in a way that all of my AI assistant entities can access them on demand, right next to the code they document.

The handoff moment

The best part of that setup session was the ending. I asked Sonnet for a prompt to give Opus for the next technical session. It crafted this beautifully concise handoff:

Current priority: Replace mock responses in the intent processing system with real LLM intelligence.

No verbose context dump. No “previously on Piper Morgan…” Just a clear next step.

It was like Sonnet was saying, “I’ve organized the garage. Now go build something.”

OK, there was also a session log and a well crafted “handoff document” from the previous chat that I shared as well. I had the beginning of a new rhythm in place, and one that did not require curing amnesia every 24 hours.

What made this Day Zero different

Looking back, June 3rd worked because:

1. Limited scope — Just Piper Morgan, not every project I’d ever dreamed of
2. Clear handoff — From organization (Sonnet) to implementation (Opus)
3. Sustainable practices — Session logs in the project tree, not scattered across chats
4. Acceptance of constraints — $0 budget wasn’t a limitation to overcome but a design parameter

The irony wasn’t lost on me. I’d spent days trying to build an AI assistant to help with product management, when what I really needed was better product management of my AI assistant building.

The lessons for fellow builders

If you’re drowning in your own project’s context:

1. Stop adding features and organize what you have — Projects (or any organization system) first
2. Use the right tool for each job — Sonnet for setup, Opus for implementation
3. Document sustainably — Where you’ll actually maintain it
4. Accept that Day Zero might happen on Day 5 — Or Day 50
5. Meta work is real work — Organizing the project IS the project

Like most folks I’d rather make stuff than organize my workspace, but it’s important to recognize the signs when the friction and noise around your work is louder than the work itself.

The real beginning

By the end of that Monday morning, Piper Morgan had a home. Not just a GitHub repo or a Docker compose file, but a proper Claude Project with context, purpose, and direction.

The mock responses were still mocking me. The LLM integration was still a dream. But now I had a foundation that could support what came next.

Sometimes Day Zero isn’t about writing code. It’s about admitting you need to stop, organize, and create the conditions for success. Even if it’s your third Day Zero.

Especially if it’s your third Day Zero.