Oracle has launched the APEX AI Application Generator in Oracle Cloud Infrastructure on Autonomous AI Database Serverless as part of Oracle APEX 26.1.
The release introduces APEXlang, a new Application Definition Language that lets AI agents generate full data-centric applications as structured definitions rather than conventional source code. Oracle APEX then compiles and runs those definitions within its managed runtime.
Oracle is presenting this as a different approach to AI-assisted software development at a time when many tools focus on producing large volumes of code from prompts. In this model, the output is an abstract application definition that can be reviewed, version-controlled, and moved through existing development processes.
Michael Hichwa, Senior Vice President of Software Development, described the shift as a change in what AI should produce during application development.
"As we all know, AI is changing how applications are built. Developers, business analysts, and line-of-business teams are clearly using AI to accelerate application development, with or without IT involvement. For enterprises, however, speed is only part of the equation. Business-critical applications must also be secure, governed, auditable, maintainable, and integrated with established data, development, and deployment processes. This raises an important question: How do we empower business users and developers to build AI-generated applications while maintaining security, data access controls, and operational governance? At Oracle, we believe the answer isn't simply to generate more code. It's to rethink what AI should generate in the first place," said Michael Hichwa, Senior Vice President of Software Development at Oracle.
How it works
According to Oracle, APEXlang defines the full structure of an Oracle APEX application, including pages, navigation, forms, reports, dashboards, and business logic, in a declarative format. The format is human-readable and can be handled like other software artefacts used in corporate development environments.
That means teams can inspect changes, compare versions, review updates, and manage releases without relying solely on generated implementation code. Oracle argues this should fit more easily with established processes for source control, testing, deployment, and rollback.
The company drew a parallel with SQL, where developers describe the desired outcome and the database determines how to execute it. Here, AI agents create an application definition, while Oracle APEX handles execution inside the managed runtime.
Oracle says this structure is intended to keep application generation within defined boundaries. Because APEXlang uses a documented grammar and a constrained format, AI agents work with a narrower set of instructions than in general code generation.
Development options
Developers can work with APEXlang through several routes, including the low-code App Builder, native AI agents, and the Oracle SQL Developer extension for Visual Studio Code, where teams can review differences, merges, and AI-made changes in text form.
The language is also publicly documented with examples and grammar rules, with the aim of allowing external AI models to work with it directly. Oracle highlighted prompt-based changes such as adding a chart, translating an application, renaming buttons, or introducing an AI agent to answer questions about customer orders.
"A useful analogy is SQL. Developers use SQL to describe what they want to do with data, while Oracle AI Database determines how to execute those operations. APEXlang applies the same philosophy to applications. Instead of generating implementation code, AI agents generate an application definition. Oracle APEX handles its execution within a managed runtime, allowing the platform to continue enforcing consistency and operational best practices," said Hichwa.
Broader update
Alongside the generator, Oracle APEX 26.1 adds other AI-related functions. These include tools for building AI agents conversationally inside applications and a new AI Interactive Reports feature that lets users alter reports through natural language requests.
Users can request changes such as pivoting sales by region, filtering to a geography, or sorting by customer count, with the system translating those instructions into report actions. The aim is to reduce the need to manually configure report controls.
Oracle linked the new features to its long-standing APEX platform, which it says has been used to build millions of data-centric business applications over more than two decades. By keeping generated applications inside the APEX runtime, older applications can continue to receive platform updates and security patches without being rewritten.
"The question is no longer whether AI can help developers build applications; it clearly can. The more important question is how AI can help build enterprise-class applications. Oracle APEX AI Application Generator enables organizations to exploit the full power of AI agents to generate enterprise-class, data-centric applications. Best of all, enterprises can leverage the AI coding agents they are already using (ex: Codex), and can manage the generated artifacts using existing organizational source control standards. The same artifacts can move through established enterprise development lifecycles-from source control and review to test, production deployment, rollback, and recovery. This means organizations can innovate faster without giving up the data governance, security, reliability, and operational consistency they depend on," said Hichwa.