Couchbase is strengthening its enterprise AI capabilities with an expanded AI Data Plane designed to help organizations move artificial intelligence (AI) agents from pilot projects to production while addressing data fragmentation, security, performance, and operational costs.
In an exclusive interview with AI Times, Michael Cronin, Vice President for Asia-Pacific, Japan, and India at Couchbase, identified fragmented data infrastructure as one of the biggest obstacles to enterprise AI adoption. He explained that organizations often maintain separate systems for operational data, vector search, analytics, and AI memory, each with its own security requirements, synchronization processes, and potential points of failure.
According to Cronin, the challenge is no longer simply choosing the right AI model but building an infrastructure that allows agents to access reliable, relevant, and governed data at scale.
Couchbase’s expanded AI Data Plane addresses this challenge by bringing operational data and AI context together in a single governed platform that operates across public cloud, private data centers, and edge environments.
The expansion introduces four generally available capabilities: native Amazon Bedrock integration with semantic caching, Managed Agent Memory on Amazon Web Services (AWS), a managed Model Context Protocol (MCP) server on AWS, and vector search within Couchbase Operational Insights.
Reducing AI costs through semantic caching and persistent memory
One of the key enhancements is the native integration with Amazon Bedrock, which provides access to foundation models and vectorization through private AWS networking without requiring separate GPU infrastructure.
Cronin explained that the integration enables semantic caching, which identifies similarities in the meaning and intent of incoming prompts rather than relying on identical keywords. This allows applications to retrieve previously generated answers for similar queries instead of repeatedly invoking large language models (LLMs).
According to Couchbase, customer deployments have demonstrated a 40% to 60% reduction in LLM calls through this approach, with sub-millisecond query responses for cached results.
Another major capability is Managed Agent Memory, which enables AI agents to retain user profiles, conversation histories, and relevant information across sessions. Rather than starting each interaction without previous context, agents can use stored information to provide more consistent and relevant responses.
Cronin said the database-native memory capability provides transactional persistence for agent sessions while keeping information within the enterprise’s governed database environment. It also supports role-based access control and field-level encryption to protect stored information.
Tested against the independent LoCoMo benchmark, Managed Agent Memory demonstrated a 33% to 45% reduction in token usage compared with leading alternatives. Vector-based context retrieval can also produce prompts that are 70% smaller, helping reduce the amount of information processed by AI models.
The capability supports popular AI development frameworks, including LangGraph, CrewAI, and LlamaIndex, allowing developers to change orchestration frameworks without rebuilding the underlying memory infrastructure.
Secure access to enterprise data and real-time analytics
Couchbase’s managed MCP server on AWS provides an open-standard interface that allows AI agents to interact with enterprise data and tools. As a fully managed service, it eliminates the need for developers to build, host, scale, and maintain their own MCP infrastructure.
Cronin described the service as a governed policy gate that enables security teams to determine which information agents can access and what actions they are authorized to perform. It also reduces the development work associated with custom APIs, connection management, and authentication while maintaining enterprise auditing and access controls.
The company has also integrated vector search into Couchbase Operational Insights, allowing organizations to analyze structured and unstructured information together through a single SQL query.
Cronin explained that traditional analytics typically handles numerical information such as sales figures, inventory, and margins, while vector search identifies similarities, intent, and patterns in information such as customer feedback and support conversations.
Combining the two allows businesses to generate insights using current operational data without separate extract, transform, and load (ETL) pipelines or secondary systems for synchronization.
For example, retailers can combine customer preferences with live inventory, pricing, and promotional information to generate relevant product recommendations and available checkout bundles. Couchbase’s write-back capability can then send analytical results directly into operational data used by business applications and AI agents.
Supporting AI agents at enterprise scale
As AI agents increasingly perform multiple tasks simultaneously, infrastructure must support high volumes of concurrent requests without compromising response times or reliability.
Cronin said Couchbase’s memory-first architecture and shared-nothing topology are designed to handle these workloads, allowing parallel queries to resolve at sub-millisecond latency rather than waiting in queues.
He cited Mavenir Networks as an example of an enterprise deployment operating at significant scale. The telecommunications company has used Couchbase for subscriber profiles and live session state serving hundreds of millions of mobile users since 2017. It is now expanding the platform to support AI-powered voice and subscriber agents that require reliable customer context and high availability.
Beyond telecommunications, Cronin identified financial services, global travel, and retail as industries seeing growing demand for production AI agents.
He cited Emirates, which uses Couchbase Lite for offline-first mobile applications supporting pilots and cabin crew. The technology allows critical applications to remain functional even when network connectivity is unavailable.
Couchbase Lite runs database capabilities and native vector search directly on devices. Its synchronization technology allows locally stored information to synchronize across systems when connectivity is restored.
For organizations operating across multiple locations, this approach supports applications that need to continue functioning despite network interruptions, including environments where reliable connectivity cannot always be guaranteed.
Philippine enterprises face opportunities and infrastructure challenges
Cronin sees significant opportunities for Philippine organizations to adopt AI agents, particularly in banking, business process outsourcing (BPO), retail, and telecommunications.
He noted that local enterprises are beginning to move beyond basic chatbots toward autonomous AI agents capable of handling more complex tasks and executing transactions.
However, he cautioned that fragmented databases, connectivity limitations, technical capabilities, and data governance must be addressed before organizations can successfully scale these applications.
Cronin pointed to the country’s growing digital investments, including cloud spending that accounts for 29% of enterprise information and communications technology budgets and the ₱16.1-billion Philippine Digital Infrastructure Project (PDIP), which aims to improve nationwide connectivity.
In an archipelago where broadband availability and network reliability remain important considerations, he said organizations need infrastructure that allows applications and AI agents to continue operating even when connectivity is interrupted.
Couchbase’s unified platform, supported by Couchbase Lite and bidirectional cloud-to-edge synchronization, is designed to address this requirement by enabling local data access and allowing transactions to synchronize once connections are restored.
Data protection and regulatory compliance are equally important. Cronin emphasized that enterprises deploying autonomous agents must ensure that access to customer and business information is properly controlled.
He highlighted role-based access control, field-level encryption, and centralized governance as important safeguards, particularly for organizations handling sensitive information under the Philippine Data Privacy Act and applicable National Privacy Commission requirements.
Moving from AI experimentation to production
Couchbase said more than half of its customers report total cost of ownership reductions exceeding 50% after consolidating fragmented data services onto its platform. Early adopters have also reported shortening AI production deployment timelines from quarters to weeks.
Cronin advised enterprises to begin by identifying fragmented data systems that increase integration complexity and operational expenses. Organizations should also prioritize AI workflows with repetitive prompts, where semantic caching can immediately reduce unnecessary model calls and associated costs.
He said consolidating operational data, persistent agent memory, vector search, and analytics allows development teams to focus more on building useful AI applications rather than managing disconnected infrastructure.
All four expanded AI Data Plane capabilities are generally available. Existing Couchbase Capella customers using AWS can provision the services through their existing console without deploying separate infrastructure.
More information is available at www.couchbase.com.
