As AI moves into enterprise workflows, databases must deliver trusted context, secure access and scalable performance. Sanjeev explains how Oracle is adapting its data platform for generative and agentic AI.
Enterprise databases are evolving beyond their traditional role as systems of record. As organisations move from transactional applications to generative AI and agentic systems, their data platforms must support real-time access to structured and unstructured information, preserve security and governance, and provide the business context needed for reliable AI-driven decisions. This shift is also changing the responsibilities of developers, who must now account for model selection, accuracy, latency and token costs alongside application performance.
In this conversation with CIO&Leader, Sanjeev Chauhan, Vice President – Cloud Engineering, Oracle discusses how Oracle is addressing these demands through AI Vector Search, Select AI, Model Context Protocol (MCP) support and flexible deployment options. He explains why trusted enterprise data will be central to AI adoption and how databases can help developers build context-aware applications without sacrificing control over critical information.
CIO&Leader: How has the role of databases changed as applications have evolved from traditional transactional systems to cloud-native, generative AI and agentic applications?
Sanjeev Chauhan: Databases were traditionally systems of record, storing business information and serving applications. Their role has changed significantly with the growth of analytics, self-service data access and AI. Organisations can now analyse enormous volumes of information to identify patterns and understand customer relationships and buying behaviour that were previously difficult to uncover.
The rise of cloud-native development and AI-assisted coding has also expanded the developer base to include citizen and business developers. This makes database-level security, access control and scalability increasingly important. Applications can be built faster, but developers must ensure that users access only the information they are authorised to see.
As AI agents increasingly take action on enterprise data, databases must provide scale, availability and security alongside persistent storage. They are evolving from passive repositories into critical platforms supporting intelligence and action.
CIO&Leader: What has fundamentally changed for developers building AI applications that work with enterprise data?
Sanjeev Chauhan: Traditionally, developers worked primarily with structured data, building applications to retrieve or update information. Today, applications must combine structured transactional records with unstructured information such as KYC documents, photographs and other customer data. They must do this in real time while preserving data lineage, access controls and security.
Developers must also make decisions about how data is chunked, embedded, retrieved and filtered, and which models are appropriate for particular use cases. Some organisations want AI models running within their own environments, while others are comfortable using public-cloud APIs. Developers must account for these deployment preferences and the requirements around data sovereignty.
Developers must therefore evaluate whether the cost of running an AI capability is justified by the value it delivers.
Accuracy, latency and token costs have become equally important. Unlike traditional application code, AI workloads can incur significant costs depending on the model and the volume of tokens consumed. Developers must therefore evaluate whether the cost of running an AI capability is justified by the value it delivers. Their responsibility now extends beyond writing code to managing the entire AI application lifecycle.
CIO&Leader: How does Oracle AI Vector Search help developers combine semantic search with business and operational data to create context-aware AI experiences?
Sanjeev Chauhan: Enterprises have large numbers of applications and tables, but understanding the relationships between them in business terms is increasingly important. Concepts such as knowledge graphs, ontologies and semantic layers help establish those relationships.
Vector search helps retrieve information based on semantic meaning rather than relying solely on exact keyword matches. It can also help systems interpret the intent behind a user’s query. When combined with business and operational data, this enables applications to retrieve information that is relevant to the user’s context rather than simply matching words.
That business context is important for generative AI and agentic applications, where understanding what a user intends to achieve can influence the relevance of the response and the information retrieved.
CIO&Leader: What role does Oracle Select AI play in helping developers interact with enterprise data through generative AI?
Sanjeev Chauhan: Traditionally, developers needed to understand SQL and, in many cases, database-specific capabilities to retrieve information and work with database internals. These capabilities can deliver significant performance benefits, but learning and mastering them requires time.
With AI-assisted development, we wanted to make database capabilities more accessible. Oracle Select AI enables users to interact with the database using natural language. Instead of manually defining SQL statements, users can express what they want to know in ordinary language, which is translated into SQL to retrieve the relevant results.
This can help developers and other users access database information without needing to master every detail of SQL. It also makes it easier to explore database performance and investigate bottlenecks through a more accessible interface.
CIO&Leader: As enterprises move from chatbots and copilots to agents that can reason and act, what new demands does this place on the underlying data platform?
Sanjeev Chauhan: The underlying data platform becomes even more important. Agents require trusted, current and governed context, not simply a static copy of enterprise data. They also need to interact with the data platform without extensive integration work.
Agents require trusted, current and governed context, not simply a static copy of enterprise data.
We have introduced Model Context Protocol (MCP) support for Oracle databases to help agents communicate directly with the database. The objective is to reduce the amount of additional integration required to explain database structures and enable these systems to access information.
The database must provide both the context and the scale that AI tools need. As agents move beyond answering questions towards taking actions, the quality, currency and governance of the underlying data become critical.
CIO&Leader: How does Oracle support developers working across different programming languages, AI frameworks and deployment environments?
Sanjeev Chauhan: Developers work with different use cases, skill sets and technology preferences. Oracle supports a broad range of programming languages and frameworks, including SQL, PL/SQL, Python, Java, Node.js and REST interfaces. We also support frameworks such as LangChain and LlamaIndex, and developers can use models within the database or connect to OCI Generative AI.
Deployment flexibility is equally important. Some organisations need databases behind their own firewalls to meet security, sovereignty or operational requirements. Others need the scalability of public-cloud infrastructure for workloads with substantial transaction volumes.
Oracle supports cloud, multicloud and hybrid deployments. Customers can run Oracle databases on Oracle Cloud Infrastructure and on other major cloud platforms available in India, including Amazon Web Services, Google Cloud and Microsoft Azure. Oracle Cloud@Customer also enables organisations to access cloud capabilities while keeping the infrastructure within their own environments.
These choices allow enterprises to align their data architecture with their workloads, security requirements and deployment preferences.
CIO&Leader: Looking ahead, how will enterprise application development and deployment change as AI becomes native to applications and the data stack?
Sanjeev Chauhan: AI will increasingly become an embedded application capability, supporting workflows, search, decision-making and automation. Applications will combine deterministic business logic with models, retrieval, semantic memory and agents that can recommend or execute actions under policy controls. Human oversight will remain necessary for certain use cases.
The differentiator will be the trusted business context applications can securely derive from high-quality enterprise data.
The differentiator will be the trusted business context applications can securely derive from high-quality enterprise data. Organisations that connect AI to relevant, reliable and governed information will be better positioned to produce useful outcomes.
AI will become an integral part of enterprise applications, but its effectiveness will depend on the quality of the data and the context in which it operates.