In the earlier days of my career, I was a developer. However, a decade spent in UX design and product management means that while I understand the landscape, my knowledge of the “flavour of the week” frameworks is naturally dated. In the past, whenever I had a spark of an idea for a hobby project, I would get caught in the trap of analysis paralysis. I would spend weeks researching whether to use Next.js or Remix, which ORM was currently in favour, or which cloud provider offered the best free tier. Usually, I would exhaust my creative energy before I even ran npx create-react-app.
This is where the STACK.md file in my context folder becomes a complete game-changer. Once I have the PRD.md and JOURNEYS.md defined, I do not go to Google; I go to the AI. I ask it to propose a tech stack that is best suited for the specific logic and user flows we have just mapped out.
The AI as a technical consultant
The beauty of using an AI agent like Cursor to define your stack is that it suggests tools it is actually “comfortable” building with. Because LLMs are trained on vast amounts of documentation and open-source code, they tend to suggest stacks with strong communities and clear patterns. When the AI suggests a stack, it is essentially telling you: “I know how to build this well.”
I treat the creation of STACK.md as a high-level technical negotiation. I provide the requirements, and the AI provides a proposal. It lists the frontend framework, the database, the authentication provider, and the hosting environment. For a product leader, this removes the friction of technical debt before the first line is even written. We are not just picking tools; we are picking a partner for the build.
Pushing back on the enterprise bloat
Even though the AI is brilliant, it occasionally suffers from “enterprise bias.” Because I work in high-stakes live broadcast environments during the day, I am used to expensive, robust solutions. However, for my hobby projects and MVPs, my resources are minimal.
I often find myself poking holes in the AI’s first draft. If it suggests a complex AWS setup or a suite of paid third-party services, I push back. I tell it: “This is a hobby project with a budget of zero. Give me the high-performance, low-cost alternative.” This back-and-forth is where the STACK.md file truly crystallises. We swap out expensive cloud functions for simpler alternatives or choose a database with a generous free tier. My role is to act as the strategic filter, ensuring we stay lean while the AI ensures we stay functional.
A foundation for the agent
The final version of STACK.md acts as a set of rules for the AI. When it comes time to actually generate the code, the AI refers back to this file to ensure it is not mixing paradigms or importing unnecessary libraries. It keeps the codebase clean and aligned with our agreed-upon architecture.
By letting the AI handle the heavy lifting of technical research, I have reclaimed the ability to ship. The STACK.md file ensures that I am building on a foundation that is both technically sound and resource-light. It is the bridge between a good idea and a working prototype. Up next in this series, we will look at DESIGN.md and how I maintain high UX standards without ever opening a dedicated design tool.
