AI embedded in databases
Databases that understand context, generate metadata and participate in governance. From traditional modelling to co-modelling with embedded intelligence.
Artificial intelligence has gained prominence for its generative capacity, but a quieter and deeper change is beginning to transform system architecture: the direct integration of AI into databases. This evolution not only improves efficiency; it redefines how organisations interpret, manage and activate their information.
It is no longer just about querying data: we work with databases that understand context, collaborate in their own governance and enrich analytical processes. This article summarises three key innovations and proposes a viable strategy for local, secure and operational data engineering environments.
Three examples that mark the direction
1. IBM Watsonx.data: Lakehouse with embedded generative AI
IBM has strengthened Watsonx.data by integrating hybrid architectures (SQL, noSQL and unstructured data) with foundational models and natural language. Companies like Lockheed Martin have reduced information retrieval errors by up to 40%, surpassing traditional RAG systems.
Source: IBM Newsroom.
2. Google Agentspace: AI agents on top of structured databases
At Google Cloud Next 2025, agents integrated directly into AlloyDB and BigQuery were presented. These agents generate pipelines, document schemas and analyse data from within the system itself, without exporting information.
Source: Futurum Group / Google Cloud Next.
3. AnDB: a database with native AI
An emerging engine designed to simultaneously support OLTP workloads, vector analytics and semantic queries with extended SQL. Its main innovation is knowledge-based query planning, allowing the engine to decide how and why to execute a particular strategy.
Source: arXiv / May 2025.
What do they have in common?
These technologies do not limit themselves to enriching queries with AI. They make databases actively participate in the data lifecycle:
- They generate metadata automatically.
- They validate data quality and consistency.
- They suggest relationships between entities.
- They create new views adapted to usage contexts.
The database stops being a passive repository and becomes an intelligent actor in the system.
A proposal from data engineering: towards local cognitive infrastructures
We propose a reproducible strategy to bring these capabilities into secure and controlled environments:
Local agents in containers
Tools like Docker AI (Gordon) and MCP toolkit allow deploying services that incorporate embedded models to automate technical tasks. They are configurable, scalable and do not expose data outside the environment.
Lightweight models as data copilots
Through private LLMs like Phi-3, LLaMA or Mistral running locally, it is possible to validate schemas, detect anomalies or generate technical documentation automatically. All without sacrificing privacy or depending on external services.
Intelligent databases with embedded governance
By integrating engines like AnDB, or through agents coupled to PostgreSQL or ClickHouse via MCP, databases can suggest optimisations, apply quality rules and cooperate in data-level decision-making.
From storing to co-modelling
Intelligence is no longer a peripheral layer. It is at the core of the system:
- Databases that understand meaning.
- Models that deploy knowledge in real time.
- Agents that correct or suggest improvements.
Data engineering enters a new stage: from modelling to co-modelling with embedded intelligence.
And most powerful of all: this transformation is not futuristic, it is happening now.
We are at the beginning of a new era in data engineering. An era in which databases not only store, but think, dialogue and participate in processes. Embedded AI is not a complement: it is a paradigm shift that opens the door to truly cognitive infrastructures.
This shift invites us to rethink everything:
- What if data could explain itself?
- What if the system not only responded, but also proposed?
- What if storage stopped being a destination, and became an intelligent partner?
Horizons are no longer measured only in capacity or speed, but in capacity for understanding. Betting on these technologies is more than modernising systems: it is building ecosystems that learn, improve and collaborate.
The question is no longer whether to integrate AI into our data, but how to lead that leap with sense, security and vision.
And for those of us working in data engineering, the moment to open that conversation is now.
