I have a few half-baked ideas for software products that could help me or my colleagues in our work. But I’ve found the challenge of actually sitting down and developing a prototype to be insurmountable with my time and programming ability. For a while I have been use Claude.ai and Claude Code for small personal projects and research, not for trying to build working prototypes.
So, I figured, given that the future is poorly distributed, I would like to see if I could redistribute it a bit in my favor. I came up with this plan:
- Write an initial product description that is close to a stream of consciousness, I want to get ideas out and try to make connections later.
- Open up a terminal in vs code and ask Claude Code to read the text file.
- Use that as the start of a conversation to develop two other text files:
- PRODUCT_VISION.md: Overview of the product including all the features I would like to see in it for it to really be impactful with some details regarding the tech stack to use.
- PROTOTYPE_SPEC.md: Detailed spec of what to build now, it has a reduced feature list that is buildable, and convinces the user that they want the full-blown product.
- Once the two documents above are complete (after significant iteration and discussion with Claude), start building with two goals in mind:
- Get granular enough to learn how the whole thing works (the convenient thing with Claude is that they are a patient tutor even when you are a demanding task master).
- See how laissez-faire I can be on the details (just how good is Claude when left alone?)
- I realize the above two are a bit antithetical to each other, but its an important balance to strike in the new world we live in.
- After the prototype is built, refine it as needed then show it to some people to see what they think.
I have one product in the works, so interested to see how it plays out. On the next one I will probably add a file to the two listed above, “PERSONAS.md” along of the lines of Alan Cooper’s thinking.