Every startup I have ever run was bootstrapped. Soda Music. Musikator. Konserama. All of them. No VC money, no seed round, no fancy office with a ping pong table. Just a problem, a budget that made me nervous, and the stubborn belief that the right tool beats the expensive tool every time.
That instinct didn’t change when I started Malau Digital, my new venture building WordPress websites for small businesses in the San Gabriel Valley. What changed is the tooling available to me. Because in 2026, “building something yourself” includes AI, and how you use it says a lot about how you think.
Before I had a single client, I built a CRM. Not because I was overengineering things. Because I knew exactly what I needed and nothing off the shelf was worth paying for.
The problem was specific. I needed to find local businesses with weak or missing websites, research each one properly, write a specific pitch, track who I contacted, and follow up without losing the thread. A real CRM could do all of that. I just didn’t need all of that. I knew what I needed, and I wasn’t going to pay for the rest of it.
Google Workspace I already pay for. A spreadsheet, Gmail, and Apps Script. That’s the infrastructure. The question was what to build on top of it, and who helps me build it.
This is where I have to be honest about something people get wrong with AI tools. Claude and Codex are not the same thing. Using both interchangeably is like hiring a copywriter and a developer and asking them to swap jobs every Tuesday. I tried it. The work suffered.
The point is to know which one does what.
Claude is my writing and strategy partner. When I needed to think through the business model, the positioning, the pitch angle, what kinds of businesses to go after first and why — that was Claude. When I sit down to draft an outreach email, that’s Claude. The thinking partner. The one I talk to when the question is “what should this say and why.”
Codex is my systems partner. Once the research process was clear, I brought Codex in to turn it into something that actually runs. It built the Apps Script layer: the pipeline sheet, automation menus, color coding, the sync between Gmail and the spreadsheet, timestamps, follow-up date logic. Things that need to work the same way every time I open the sheet. That’s Codex territory.
The result is a CRM that does one thing well. It tracks prospects from first research through email sent through follow-up. Each row has the business name, city, category, website status, what the trust gap is, what the pitch angle should be, and where I am in the conversation. Not copied from a template. Built from what I actually needed.
The handoff is clean. Codex qualifies and organizes the prospect. Claude drafts the outreach once the row has enough context. I review it, tweak it, and send it myself. The system logs it. The row moves from Researched to Drafted to Contacted, with dates and notes attached.
I didn’t invent anything here. Google Sheets has existed for years. Apps Script is not new. AI coding assistants are everywhere. None of these tools are magic on their own. The system works because I was clear about the problem before I touched a single tool. The tools filled in the shape I already had.
Four days of actual work. Figuring out the right tools, making mistakes, learning from them, fixing what broke. That’s the real cost. Two AI subscriptions I already had, one Google Workspace account, and four days of not knowing what I was doing until I did.
I built the CRM before I had a single client. That’s when you build infrastructure. When there’s no pressure and you can still think straight.