Agentic APEX: Embedding AI Into the Full Development and Support Lifecycle

Oracle APEX has a problem that no other enterprise low-code platform shares: you cannot script the page builder. The Page Designer is a browser-based drag-and-drop tool with no CLI, no API, no programmatic access. Every button, every region, every dynamic action requires a human clicking through the interface. This is fine for building apps. It is a serious constraint when you want AI to help.

We built a multi-agent system for a consultancy in Central Europe. The goal was not to replace the developer - it was to handle everything around the Page Designer work so the human spends their time exclusively on the parts that require a browser.

How the agents are structured

• architect: the lead agent. Reads the ticket, triages the work, breaks it into tasks, reviews everything before handover.
• database developer: writes the database layer following strict schema conventions and deploys directly to the client environment.
• spec writer: generates step-by-step Page Designer instructions that a human developer follows in the browser.
• quality verifier: after the human builds the page, exports the result, diffs against the spec, and runs automated browser checks.

The architect decides the routing. Many support tickets turn out to be database-layer fixes that need no Page Designer work at all.

Connecting to client infrastructure

The agent system handles multiple clients through automated VPN activation - it autonomously establishes encrypted connections to reach the correct client database. Tickets, notes, and attachments are read directly from the ticketing platform through a custom MCP (Model Context Protocol) server. Handover documentation uploads back to the same ticket automatically.

Understanding the existing application

Before any agent writes code, the system reads application exports - the full application structure including pages, regions, items, processes, and dynamic actions. The agents are loaded with specialised skills encoding the consultancy’s actual standards for database development, APEX page conventions, security practices, and naming standards.

The full workflow

Ticket arrives. The architect reads it via MCP, classifies the work, creates subtasks. The db-developer builds the database layer and deploys. The page-spec-writer generates numbered instructions. The human opens Page Designer and follows the steps. The qa-verifier exports, diffs, and runs browser checks. Handover documentation is generated and uploaded.

Developers interact through simple slash commands in the terminal. The system does not act on tickets without a developer initiating the workflow.

What this means in practice

Developers now spend their time on two things: building pages in Page Designer and reviewing agent output. The database work, spec writing, QA verification, and documentation - all handled by agents. All connections encrypted, no client data leaves the infrastructure, every action logged to the ticket for audit.

This is not a generic AI assistant bolted onto a helpdesk. It is a purpose-built system designed around a specific constraint: the Page Designer cannot be automated, so automate everything else.