A featured contribution from Leadership Perspectives: a curated forum reserved for leaders nominated by our subscribers and vetted by the Construction Tech Review Advisory Board.
Evan Yang
But there's a B-side. Spending continues to climb, while business value often fails to keep pace. Gartner predicts 60 per cent of AI projects will be abandoned because they fail to meet production requirements, most often due to a lack of high-quality data. Between a model's general intelligence and the intelligence an enterprise actually needs lies a gap of business context. Closing it is, at bottom, a data problem. The next generation of enterprise AI will not be limited by model intelligence alone. It will be limited by whether enterprises can turn their data into real-time, trusted, and actionable context.
This brings me to the question I've spent the past year considering, and the real subject of this piece. In the age of AI, what should a database become? What must change, and what must be preserved?
Two Assumptions Just Broke
For decades, databases rested on two assumptions: the user was a person writing SQL or reading dashboards, and the data worth managing was structured in tidy rows and columns. AI has broken both assumptions.
The user is no longer just human. Increasingly, it is an agent that operates continuously, calls tools, reads and writes data, maintains state, and acts at machine speed. Gartner expects 33 per cent of enterprise applications to embed agentic AI by 2028, up from less than 1 per cent in 2024.
At the same time, enterprise data is no longer primarily structured. Roughly 80 per cent consists of documents, messages, images, audio, video, and logs. For the first time, AI makes this information computable rather than simply archival.
Neither shift is simply another feature request. Together, they fundamentally redefine what enterprises should expect from a database.
What AI Agents Demand
Building for AI agents over the past year has revealed three essential requirements.
The first is scale, but not in the traditional sense. Instead of making one database larger, organizations must support millions of lightweight, mostly dormant databases that activate instantly while remaining isolated and cost-efficient.
The second is context. AI agents continuously search across structured and unstructured information to build the context required for accurate reasoning. Hybrid retrieval combining structured queries, full-text search, and vector search becomes essential. Ultimately, an agent's effectiveness depends less on the model itself than on the quality of context it can assemble.
The third is evolution. Agents improve through constant experimentation. They need isolated environments where they can safely test new workflows, modify data, evaluate results, and discard unsuccessful approaches without affecting production systems. Much like software developers use code branches, AI agents require branching environments for continuous learning.
As AI evolves, enterprise data itself takes on new responsibilities.
Unstructured information, once treated as a storage problem, becomes a computable business asset. Documents, emails, recordings, and videos must be governed and queried alongside structured records using unified hybrid retrieval.
Data flow must also evolve. Traditional architectures separate operational databases, data warehouses, and AI training systems through slow pipelines. AI requires a continuous feedback loop where operational data immediately becomes learning data, allowing systems to improve continuously.
-
The future of enterprise AI will not be defined by who has the most data or the largest model, but by who can connect data, context, and meaning well enough to make AI trustworthy, explainable, and actionable.
Finally, interaction with data is changing. SQL remains valuable, but AI increasingly interacts through natural language and semantic understanding. Databases therefore need semantic layers that understand business entities, relationships, metrics, and policies, allowing AI to interpret information in meaningful business terms rather than simply retrieving records.
What Must Never Change
Although AI changes how databases are used, it makes their foundational principles even more important.
Consistency must now extend across structured records, text, vectors, and files. Scalability shifts from supporting one massive database to efficiently managing millions of smaller ones. Reliability becomes critical because AI agents operate continuously without human supervision. Real-time performance also becomes non-negotiable, as every inference depends on retrieving accurate context instantly.
AI demands architectural innovation, but it cannot compromise the engineering foundations that businesses have relied upon for decades.
Defining the AI Database
These changes lead to a new definition of the AI database.
It must unify structured and unstructured data on a single real-time foundation. It must support multimodal information, including text, vectors, JSON, and traditional relational data, within one engine. It must be agent-friendly, providing native capabilities for memory, isolation, branching, rollback, and natural-language interaction. Finally, it must remain open, allowing organizations to retain control over their infrastructure, models, and data.
The industry increasingly describes this architecture as a lakebase, where the capabilities of the data lake and the transactional database converge into one system.
OceanBase's Perspective
Our answer is the OceanBase AI Database.
Rather than simply combining a lake with a database, we believe a lakebase must extend from a mature transactional core capable of financial-grade consistency, reliability, and real-time performance. That foundation enables multimodal storage, hybrid search, native support for AI agents, open architectures based on S3-compatible storage and Iceberg tables, and a unified catalog without sacrificing transactional guarantees.
For the past fifteen years, OceanBase has been proven in demanding production environments where data accuracy, availability, and rapid recovery are non-negotiable. Today, we are applying those same principles to some of Alibaba's and Ant's most demanding AI workloads.
Just as the Double 11 shopping festival once pushed distributed databases forward, AI is now driving the next evolution of database architecture. Our goal is a simple one, to build OceanBase for the AI era.
