aiagents.org

Build a software project from an idea

Coding agentKINDA1 method

KINDA — AI can help, with a person in the loop.

1 method on the page: 1 agent.

Use cases

Methods

Build a software project from an idea

agent

Aiderrepomix

Steps

  1. Talk through the idea with a chat model, a single question each turn, until the details are covered. Then ask it to compile everything into a developer-ready spec and store it as spec.md.
  2. Give the spec to a reasoning model and ask for a step-by-step blueprint, cut into small chunks, then a series of test-driven prompts for a coding model. Each prompt extends the last and closes by connecting the pieces; save the result as prompt_plan.md.
  3. Ask for a todo.md checklist that the coding tool can tick off as it works, which keeps state between sessions.
  4. Set up the repository and tooling yourself, then paste the prompts into your coding tool one at a time, in order, and run the tests after each.
  5. When a step fails, pack the codebase with repomix and give it to the model to debug, or answer questions in Aider; once it works, move to the next prompt.
  6. Repeat steps 4 and 5 for each prompt in the plan.
  7. For an existing codebase, the author uses a lighter path: pack the code with repomix, have a model write a review, a list of missing tests or GitHub issues from it, and work through those items one at a time.
  8. The author estimates steps 1 to 3 take about 15 minutes. He says the workflow works now and may change quickly.
Run this method with your agent
Download .mdCopied 0
---
name: "build-a-software-project-from-an-idea-method"
description: "Use when you need to Build a software project from an idea. Build a software project from an idea"
---

# Build a software project from an idea

Job: [Build a software project from an idea](https://aiagents.org/canai/build-a-software-project-from-an-idea)

## Need

- ai agent: [ChatGPT](https://aiagents.org/agents/chatgpt), [Claude](https://aiagents.org/agents/claude), [Aider](https://aiagents.org/agents/aider), [repomix](https://aiagents.org/agents/repomix)

## Steps

1. Talk through the idea with a chat model, a single question each turn, until the details are covered. Then ask it to compile everything into a developer-ready spec and store it as spec.md.
2. Give the spec to a reasoning model and ask for a step-by-step blueprint, cut into small chunks, then a series of test-driven prompts for a coding model. Each prompt extends the last and closes by connecting the pieces; save the result as prompt_plan.md.
3. Ask for a todo.md checklist that the coding tool can tick off as it works, which keeps state between sessions.
4. Set up the repository and tooling yourself, then paste the prompts into your coding tool one at a time, in order, and run the tests after each.
5. When a step fails, pack the codebase with repomix and give it to the model to debug, or answer questions in Aider; once it works, move to the next prompt.
6. Repeat steps 4 and 5 for each prompt in the plan.
7. For an existing codebase, the author uses a lighter path: pack the code with repomix, have a model write a review, a list of missing tests or GitHub issues from it, and work through those items one at a time.
8. The author estimates steps 1 to 3 take about 15 minutes. He says the workflow works now and may change quickly.

## Credit

Method author Harper Reed
Source: https://harper.blog/2025/02/16/my-llm-codegen-workflow-atm/

Method author Harper ReedSource