Brett Malinowski ran a software agency that billed $6,000 per month per client and took six weeks to deliver each project. With Claude Code and a handful of connected tools, that same output now takes one person under an hour. Here is the exact workflow, and the business model behind it.
The Creator-Software Agency Model
The core insight is simple: educational creators already have the three things that kill most software startups before they launch. They understand the problem deeply because they live it, they have an audience ready to buy, and they have a peer network that can accelerate distribution faster than paid ads ever could.
The Atlas AI case makes this concrete. The founder was an e-commerce YouTuber with over a million subscribers. His main growth channel was not even his own audience. It was peer creators in his niche who gave him favorable sponsorship rates because he was one of them. That product is now doing over $500,000 per month.
The pitch to a creator is straightforward: find an educational creator in a niche you understand, identify the software their audience actually needs, and offer to build it. Three deal structures work here. A monthly build fee of $2,000 to $6,000 depending on scope and proof of work. Pure equity if you believe strongly in the idea. Or a hybrid, which is the most defensible option: a smaller monthly fee plus ownership stake. You cover costs while you build, and you have real upside if the product works.
Automating the PRD Phase
A product requirement document is the foundation of any software build. It defines what the app is, who it is for, what features ship in version one, and what the technical constraints are. Traditionally, a project manager handles this after a discovery call. That step is now fully automatable.
The workflow: record the client call with an AI note-taker like Fellow, connect Fellow to Zapier, pipe the transcript into Claude with a system prompt that instructs it to act as a senior product manager and output a build-ready PRD, then save the result to a labeled Google Drive folder. The whole Zap takes minutes to configure and runs automatically after every future client call.
One high-leverage detail: embed your preferred tech stack directly into the PRD system prompt. When Claude Code receives that PRD later, it already knows which backend, database, authentication, payments, and hosting stack to use. This is what makes the workflow repeatable across every future project without re-specifying preferences each time.
Building the App Without a Design Team
The first version Claude Code produces from a PRD will be functional but generic. The fix is two assets: a brand kit and a component library. A brand kit from Canva establishes the color palette and typography. A free component library from Figma provides the prebuilt UI elements, buttons, cards, modals, and navigation bars that give the product a consistent, polished feel.
Feed both into Claude Code alongside the PRD and iterate. The difference between an app that looks AI-built and one that feels like a real product is almost entirely in these two inputs. If you are charging $6,000 a month, the output cannot feel like a prototype.
Infrastructure: What to Build vs. What to Plug In
This is where most AI-assisted builds fail. Payments and real-time chat are not features, they are infrastructure problems. A chat system that works for ten users breaks at ten thousand. Marketplace payouts require compliance, webhooks, and edge-case handling that vibe-coded solutions cannot handle reliably.
The correct approach is to tell Claude Code to integrate pre-built components rather than build from scratch. For payments and chat, plug in an existing SDK. For authentication, use OAuth through the same provider. Claude walks you through the API key setup, webhook configuration, and environment variables step by step. You follow the instructions, not write the code.
For the database, Supabase handles schema setup for users, listings, purchases, and categories with minimal configuration. For hosting, Vercel connects to a custom domain in minutes. Once these accounts exist on your first project, Claude can replicate the same stack on every future build automatically.
The Bottom Line
What used to require a project manager, a designer, and a developer working for six weeks now fits inside a single automated workflow. The business model works because creators bring distribution and you bring execution, and AI has made execution cheap enough that one person can run what used to be an agency. Watch the full video to see every prompt and tool configuration in action.