As Q1 2026 closes, let’s take stock of where Oracle 23ai stands and what signals point to the next steps.
Adoption status:
Oracle 23ai has been in production on OCI for nearly two years. On-premises, Enterprise adoption is accelerating — the Oracle community surveys from late 2025 showed roughly 30% of large Oracle shops either already on 23ai or planning a migration within 18 months.
The SQL features with highest reported adoption are, in order: IF [NOT] EXISTS, GROUP BY ALL, the BOOLEAN data type, JSON enhancements, and schema-level privileges. These align with what I’d have predicted — they’re the features with immediate, tangible developer productivity benefits and low migration risk.
AI Vector Search maturity:
The biggest shift in early 2026 is the growing number of production Vector Search deployments. Embedding Oracle AI Vector Search into RAG architectures — using Oracle as the single data store for both relational and vector data — has become a viable, production-tested pattern.
DBMS_VECTOR_CHAIN and the automatic embedding pipeline (generate embeddings without leaving the database) are the features driving this. Oracle processing embeddings natively reduces the architecture to: data in Oracle → embedding in Oracle → vector search in Oracle. No Python preprocessing pipeline required.
What to watch:
Oracle has signaled continued investment in: ONNX model import (run ML models directly inside Oracle), expanded MLE language support (Python stored procedures are being discussed), and deeper APEX integration with AI features for low-code AI application development.
On this blog:
We’ll keep this series going. Weekly Oracle posts, specific and practical. If you have topics you’d like covered — leave a comment or reach out. The Oracle 23ai feature set is rich enough to sustain another full year of deep dives.
Thanks for reading. See you next week.
