5.6 Creating an App Using AI and Spec-Driven Development (SDD)

Create an application using AI and spec-driven development.

When using AI for generative development, and spec-driven development (SDD) in particular, you can use a blueprint to define a high-level application description that APEX uses to generate a full application. An APEX blueprint is an importable application design artifact written in Markdown that describes exactly what APEX should build: pages, navigation, regions, reports, forms, charts, filters, actions, and page behavior. The blueprint is inspectable, meaning that you can review it before creating the application, and can be regenerated when the application specification changes.

A functional specification defines what an application should do. An Oracle database schema defines the data it works with. By providing both these inputs to an AI coding agent, the agent can generate an APEX blueprint. Blueprint scaffolding can be faster and more token-efficient than generating APEXlang directly because it encodes common APEX page patterns into reusable structures.

In order to create an application blueprint, consider three phases:

1. Inputs, where you provide the context, requirements, and guidance:
   • a functional specification, containing the requirements and business context
   • schema metadata, describing the database objects and structures
   • a system prompt that guides AI with best practices, page patterns, and generation rules
2. Blueprint Generation, where AI analyzes the input files and produces a Markdown blueprint,
3. Import, where you import the Markdown file into APEX.

From the Markdown file, APEX creates the application.

Inputs

Spec-driven development starts with the functional specification, where you define what the application is, and what it does. For example, a supply chain management application could describe inventory and warehouse management flows for goods receipt, storage, stock movements, fulfillment, and returns.

The second component is the schema metadata that describes the database schema. You can export this information directly from APEX using the Describe Tables utility. For more information, see Describing Tables in Oracle APEX SQL Workshop Guide.

The system prompt that instructs the LLM to produce a properly structured output is available on Github: github.com/oracle/apex/tree/26.1/blueprints.

Blueprint Generation

Combining all three inputs together enables your AI agent to produce a blueprint. The blueprint defines the application's flows, entities, and relationships in a format that APEX expects. Since the blueprint is human-readable, you can review it prior to import to understand how the functional specifications will be translated into the application.

Import

You can now take the AI-generated blueprint and import it into the APEX import wizard. Every section of the blueprint maps to an application component, and produces a baseline application that you can further enhance in the App Builder. For more information, see Importing an Application Blueprint.