On the First Day of 26ai, Viscosity Gave To Me…

Bringing Agentic AI Right to My APEX Apps 

Oracle 26ai moves AI from "bolted on" to "built-in." I'm most excited about MCP Server support and the APEX AI Application Generator, which finally lets us build secure, agentic applications directly within the database, eliminating brittle glue code and making AI a first-class citizen in the Oracle stack. 
 

Last year, I wrote about Oracle 23ai and how it set the table for doing real AI work next to  my data. This year, 26ai shows up and basically pulls the table straight into the kitchen. AI is now baked into the database and the developer workflow. 

If you only remember two things from 26ai, make them these: 

1. Model Context Protocol (MCP) Server support, and 
2. APEX AI Application Generator (the natural-language builder that uses your data and context to scaffold real apps). 

Both land right in my sweet spot as someone who’s previously wired Oracle APEX to an LLM (I used Claude) and has been waiting for a native, supported path to make that experience first-class inside the Oracle stack. 

Why MCP Inside the Database Matters 

MCP is a standard that lets AI agents reach out for tools and data in a controlled way. With MCP Server support in 26ai, agents can talk to the database through well-defined interfaces, iterate, ask for more context, and come back with better answers grounded in our private data. No more brittle glue code or bespoke adapters for every experiment; we get a clean, standards-based lane between LLMs, agents, and the Oracle AI Database. 

For me, this means the next time I connect APEX to an LLM, I’ll reach for MCP instead of custom shims. I can keep sensitive logic and data governance in-db, let the agent reason step-by-step, and still meet the privacy and audit bar my customers expect. (Bonus: the database now treats agents as first-class citizens via  Select AI Agent, so I can define and govern agents inside Autonomous AI Database and even wire in MCP tools alongside in-db and REST tools.)  

The APEX Moment I’ve Been Waiting For 

I’ve leaned on “vibe” coding tools daily, such as Replit, Lovable, Antigravity, and Windsurf, because natural language is an amazing accelerator when you're exploring ideas.  APEX AI Application Generator  brings that same energy to Oracle APEX. Speak your intent and get an enterprise-grade starting point that already respects the data model and the guardrails. For teams that need to ship apps quickly (and safely), this is the missing native piece.  

Two related features make this smarter than “code genie” hype: 

• Data Annotations: you can describe purpose/semantics/constraints, so AI tools generate better pages, validations, and queries from the start. 
• Unified Data Model: the same data can be used relationally (SQL), as JSON docs, or as a graph, without fragile ETL hops. This is perfect fuel for an AI builder.  

Retrieval That Matches How Apps Actually Search 

26ai’s  Unified Hybrid Vector Search  blends vectors with relational, JSON, text, graph, and spatial predicates in a single query. That means my APEX page that filters by account status and date can also bring in semantically similar docs, images, or notes. No split systems or weird fan-outs. It’s the practical path to RAG that respects how business apps query today.  

Agents, but Make Them Operational 

Beyond MCP and Select AI Agent, 26ai ships a  Private AI Services Container  for running your own models (embeddings, open-weight LLMs, NER) privately, and it can integrate with NVIDIA’s NIM services for embeddings and RAG pipelines. In short, these are real deployment options, not just demos.  

For shops standardizing on Exadata, vector offload  to intelligent storage and the new  Exascale  architecture sweeten both performance and elasticity. This is handy when your proof-of-concept suddenly becomes a thing everyone wants. 

Security and Governance You Don’t Have to Duct-tape 

I care a lot about least-privilege and audit trails. 26ai brings  Built-in Data Privacy Protection (row/column/cell-level controls, dynamic masking) and  SQL Firewall  to keep unauthorized SQL/injection out, so both humans and agents only see what they should. Add  quantum-resistant ML-KEM  for data-in-flight, and you’ve got a modern posture that travels with your data.  

A Mini Plan for My First 26ai + APEX Experiment 

1. Stand up  an Autonomous AI Database and enable  Select AI Agent  with an MCP Server. 
2. Annotate a small, real dataset (purpose, semantics, constraints). 
3. Prototype  with the  APEX AI Application Generator  to scaffold the core pages. 
4. Add RAG  via Unified Hybrid Vector Search for the “smart search” experience users actually want. 
5. Lock it down  with data privacy policies + SQL Firewall, then run a quick load test. 

I’ll write a follow-up with code and results once I’ve put this through its paces. 

In Summary 

If 23ai laid the groundwork, 26ai consolidates it into a cohesive, AI-native database. You get agentic workflows, MCP alignment, pragmatic RAG, and AI-aware developer tooling, all without scattering your stack. That combo is what gives me confidence that I can ship useful AI into APEX apps that my customers already trust.  

Here’s to a new era of agentic APEX apps, and to more building, learning, and experimenting in 2026. 

 
Happy Holidays! 🎄 

References:  Oracle’s 26ai announcement and technical blog provide a deeper breakdown of MCP Server support, Select AI Agent, Unified Hybrid Vector Search, APEX AI Application Generator, and more.  

• Oracle Press Release: Oracle AI Database 26ai Powers the AI for Data Revolution 
• Oracle Blogs: Introducing Oracle AI Database 26ai: Next-Gen AI-Native Database for All Your Data 

Ready to Take Your Oracle Skills to the Next Level?

Join OraPub, Viscosity’s training hub for Oracle professionals, packed with expert-led courses and exclusive paid member benefits.

Check out Viscosity’s event page for upcoming virtual and on-site training opportunities.