Building a Solid Foundation for AI Development Through Planning

AI development is not about picking up a tool and immediately writing code. True efficiency comes from thorough upfront planning. Before diving into development, every step of requirement breakdown and architectural design directly impacts the quality and maintainability of the final product.

Planning with AI Even Without Development Experience

For those unfamiliar with code, knowing which command to use to instruct the AI for development is already a hurdle. Even experienced developers often make a common mistake when using AI: telling it to do whatever comes to mind, resulting in chaotic code architecture and redundant functionality.

Several solutions are already available on the market. For example, AWS developed the Kiro editor, and there are open-source projects like OpenSpec that use Commands and Skills to break down and analyze the planning process. Kiro follows a very rigorous workflow — from planning to requirements analysis, writing user stories, and breaking down development checklists — each step requiring manual approval before proceeding to the next.

A More Flexible Planning Approach

However, if you are just tweaking small features or need to revise plans after initial planning, a strict linear process can be counterproductive. That is why we designed a “Planning Agent” that starts from the user’s stated requirements and automatically breaks large features into sub-features, which are further broken down into execution items such as: database structure design, admin interface, frontend interface, and frontend-backend data integration.

The biggest advantage of the Planning Agent is that while breaking down tasks, it automatically searches for existing development Commands, so users don’t need to study each command’s purpose — they get automatically matched to the corresponding to-do items.

How It Works in Practice

Suppose the requirement is “Send a message to the admin’s Telegram after a customer completes checkout.” The workflow would be:

1. Prepare the WooCommerce Hook documentation and Telegram technical docs
2. Use Skill Seeker to organize official documentation into structured Skills
3. Enter /plan to invoke the Planning Agent and tag relevant Skills
4. The Agent’s output is stored in the spec folder, organized by feature folders with sub-features as Markdown files

Each sub-feature file includes user stories, development checklists, code references, and corresponding Commands. The Planning Agent scans the existing codebase before each run to ensure it doesn’t plan duplicate or conflicting features.

From Planning to Execution

After the documentation review is complete, the “Execution Agent” starts development based on the documentation. The Execution Agent also scans the current folder structure before each development task to ensure project architecture consistency. Through this “plan, review, execute” iterative cycle, even people unfamiliar with code can effectively use AI to complete high-quality development tasks.

Good planning is not a constraint — it is the foundation for maximizing AI’s potential.