Step 1: take an established no-code application stack, with over 20 million applications currently running, more than two decades of lineage and evolution, and a track record of transcending platforms shifts. Step 2: provide the building blocks necessary to let AI generate metadata-driven domain-specific language (DSL) code that defines the application, rather than its implementation. Step 3: While you're there, add the ability to inject the applications themselves with agentic AI capabilities, for reporting, and conversational analytics. In a nutshell, that's what Oracle’s APEX AI Application Generator announcements, made earlier this month, were all about.

APEX’s BACKGROUND, CONTEXT AND HISTORY

In 1999, Mike Hichwa and Joel Kallman first began work on a project simply called Flows.

The idea behind the project was to help customers build enterprise web applications without needing to learn the patchwork of Web server application platforms and the ever-changing stack of client-side JavaScript libraries and frameworks necessary to create a more interactive experience. Flows was a no-code application development platform before the term was even coined.

The platform went through several name changes but was eventually renamed Application Express, or APEX for short and, over time, grew in adoption and sophistication. APEX has always been based on expressing the application's structure, design, and behavior. And the APEX platform runs on Oracle AI Database, inheriting its enterprise-grade security and governance capabilities without making developers responsible for building it in.

As a result, platform disruptions didn't need to disrupt the applications themselves. As application development platforms changed, across client/server, client-side Web interactivity, mobile, and the cloud, the APEX execution engine evolved with them, allowing customers' applications to adopt and implement new technologies without requiring significant code changes.

APEX’s EVOLUTION

In 2015, Oracle added the Page Designer IDE to APEX, allowing development of an application's pages, components, and logic in a single interface, through drag-and-drop authoring. This added a visual abstraction over the implementation abstraction. Also added were Universal Theme and Responsive Design, allowing development of applications that could run on both desktop and mobile form factors. Then came the fully managed APEX App Dev Service on Oracle Cloud Infrastructure (OCI), followed by support for Progressive Web Applications, declarative components such as Cards, Smart Filters, and enhanced Map regions, REST Data Sources for external data connectivity, and Oracle REST Data Services (ORDS), which facilitates authoring and hosting of data APIs.

Now, as we enter the age of agentic application development, APEX is put to the test once more. Can APEX, still under the oversight of Hichwa, now Oracle's SVP of Database Tools, make the transition? On August 11, Oracle made a set of announcements around APEX 26.1 that point to yes. And, spoiler alert, APEX's focus on application structure, rather than implementation, turns out to deliver advantageous, qualitative differences, compared with other agentic and "vibe coding" approaches.

APEXlang

First off, Oracle has introduced a domain-specific language (DSL) for defining APEX applications, called APEXlang, with a formally defined syntax and grammar. APEXlang is human-readable, declarative, and premised on delegating the actual implementation of an application to the APEX runtime. APEXlang-generated applications leverage capabilities of the Oracle AI Database, including its platform-level data privacy and the perf and scalability it’s known for. APEXlang files are formatted in clear text and are fully DevOps-manageable, including through source code control and continuous improvement/continuous deployment (CI/CD). And just as APEXlang code generates APEX applications, existing applications can be imported back to APEXlang, edited, and redeployed.

APEXlang was designed to be readable not only by humans, but by AI agents. With that in mind, Oracle is providing Public Skills to support AI-agent based APEXlang generation. So not only can the visual designer generate APEXlang code, but so can conversational, agentic interaction. That means developers, and even non-technical users, can use natural language to generate APEXlang code and thus APEX applications.

ENTERPRISE PRODUCTION-READY, BY DESIGN

Just as before the agentic era, agentic APEX development generates the application definition, not the implementation. The APEX runtime will implement the application itself, and along the way, supply security and single sign-on identity provider integration, along with protections against parameter tampering, cross-site scripting (XSS), and SQL injection. This rids the APEX application development cycle itself of building this enterprise plumbing, because it’s platform-inherited and part of APEX applications automatically, implicitly, and assuredly. So too are auditing, governance, risk management, and compliance capabilities.

Other agentic application generation scenarios don’t cover this. Instead, they generate the actual application source code, which often will lack the enterprise safeguards described above. The dirty secret of most vibe coding is that unless it is done by experienced architects, using highly governed methodologies, it’s a breeding ground for technical debt.

Yes, vibe coding can create functional applications very quickly, but that really only gets customers to prototype-level code quality. Code that is consistent, well-architected, secure, and built to last – rather than just built to run – is not easily produced through ad hoc vibe coding. Generating APEXlang avoids this problem, because implementation is left to the platform, not the agent.

TOOLING OPTIONS, END-USER BENEFITS

The tooling options for agentic APEX app building are broad. The Low-Code APEX Application Builder supports APEXlang generation through conventional drag-and-drop development, and conversational generative development, too, care of a built-in agent. For experienced agentic developers who want to bring their own LLMs, Oracle’s Native AI Agent hosts agentic development via Codex and other agents/models, with a built-in Web browser to preview the execution of the generated APEXlang code. Developers who want to use a standard IDE can also use Visual Studio Code (VS Code), by installing Oracle's SQL Developer for VS Code extension – which supports SQL, PL/SQL, and APEXlang development – and Oracle's Codex extension (which supports LLMs beyond Codex, as well).

Finally, Oracle is adding support within APEX applications for APEX AI agents, providing end-users a chat interface over their data, not as a standalone capability, but embedded within the generated application. These agents can be grounded in specific domains, allowing them to work well within the context of specific parts of an application.

Additionally, AI Interactive Reports provide a conversational interface for modifying existing reports.

THE PIECES ALIGN

We’ve now seen that APEX lets developers build application designs and specifications agentically, avoiding the pitfalls of generating source code. And now we’ve seen that APEX lets agents generate applications that themselves contain agents, bound to specific tasks. The recursion inherent in the idea may make it a bit confusing at first. But it becomes more intuitive the longer it’s considered. After all, if developers can have the productivity gains of natural language and agents, shouldn’t end-users as well? And if the answer to that question is yes, then shouldn’t the developers’ agents be able to generate end-user agents within the applications?|

Oracle’s APEXlang, the agentic capabilities of its Low Code AI Application Builder, the Native AI Agent, and the agentic APEX extensions for VS Code, together, bring a mature application development platform into the agentic age without sidelining the security, governance, compliance and manageability that enterprises demand. For customers that are “Oracle shops” this ought to be a no-brainer, especially as APEX requires no additional licensing beyond the Oracle AI Database – across all its deployment targets, including OCI, Azure, AWS, and Google Cloud, on-premises, and in hybrid environments. For folks wanting more vendor-neutral platforms, it’s a different equation, but the approach merits careful study. That goes for competitors, too.

Oracle is a client of Blue Badge Insights.