Today, many software developers rely on prompt-driven development, where a developer prompts the AI to “build me a point-of-sale system” or “build a feature” and hope for the best. Any professional software engineer knows that approach is very unreliable. In many ways, prompt-driven development is equivalent to rolling the dice with every new feature you add. Spec-driven development reduces uncertainty about how a project is built out. And this article introduces how to do spec-driven development and powerful tool that makes this complex process fast and easy.
Spec-Driven Development
Spec-driven development (SDD) is an approach to AI-driven software development where project definitions, project design, rules, constraints, and acceptance criteria are documented in spec files. Think of a “spec” as a documented design specification for a project. Any design detail the coder may need to tell the AI would be stored in a spec file. Suppose, for example, website users who have different privileges will have access to different parts of the website. It’s important to document those rules in spec files so that the AI knows that before making substantial changes.
The AI is never going to know the intentions or vision of the project designer without having that information explicitly written down. Those intentions and vision could include complex business rules, database architecture, security, data, performance, and user experience.
When the AI does not know the business rules or database architecture and it begins making substantial changes, it will break things. Spec-driven development reduces that possibility by explicitly documenting the intentions, vision, business rules, and architecture of the designer before AI begins making substantial changes to code.
Real-World Examples
To get started with spec-driven development, create a SPECS folder in your project. Within this specs folder, you may have sub-folders that define different parts of your project, such as Database, User Interface, and so on. This is where all of the design documentation and rules for your application will live.
1) When starting a new session with AI, it helps to give the AI a general idea of your project. That’s where a project definition spec file comes in handy.
ProjectDefinition.md: Create a markdown file that describes the purpose of your project, what technologies you intend to use, and what purpose it will serve.
2) Anytime AI needs to write code that interacts with the database, it needs to know the structure of that database. If that information is in a spec file, simply refer to that spec file when prompting the AI to write that code.
DatabaseDefinitions.md: Contains DDL scripts, ER diagrams, and so on.
3) If an application has many complex rules, those rules need to be reexplained to AI every time it interacts with specific parts of the application. For example, if an application has three different subscriber types and each has different rules in how they are handled, those rules need to be explicitly documented in a spec file. Otherwise, when the AI works on your application, it will not be aware of those rules.
SubscriberTypeRules.md: Describes the rules as they apply to subscriber types.
4) Many business applications have complex rules for how the user interacts with the application. The rules for each page and each section need to be documented in a spec file. Otherwise, the AI will not be aware of those rules when it makes changes to your application.
RegistrationPageRules.md: Describes how the user is supposed to interact with the registration page.
If you’re looking for an easy and powerful spec-driven development tool, try Acceledesk’s MD Creator AI Agent. The MD Creator markdown editor can be downloaded for free. The AI Agent makes it super easy to generate precise documentation using Mermaid diagrams to do UMLs, ERDs, Dataflows, and other diagrams. It also generates LaTeX expressions powered by MathJax to express mathematics. The AI Agent significantly reduces the time it takes to do Spec-Driven Development from hours to minutes.