Dell And Elastic Target The Data Bottleneck Holding Back Enterprise Agentic AI

As enterprise customers race to move agentic AI from demos into production, many are encountering a common obstacle: enterprise data that remains scattered across databases, object stores, applications, documents, tickets and logs. For solution providers, the challenge is not only integrating and governing that data, but making it searchable in a way that allows AI agents to retrieve and rank the right context quickly and accurately at scale.

That puts new demands on enterprise search. Agents need to work across structured and unstructured information while finding relevant context, preserving metadata and permissions, and combining meaning-based retrieval with the precision required to find exact information. Elastic describes hybrid search as combining semantic and keyword retrieval, particularly for AI applications where both meaning-based relevance and exact-match precision matter.

CRNtv’s Matt Harris spoke with Sri Desikan, vice president of product management at Elastic, and Vrashank Jain, director of product at Dell Technologies, about how Elastic and the Dell AI Data Platform aim to address those data and search challenges, giving partners a validated foundation for agentic AI while leaving room for higher-value services, consulting and implementation.

Matt Harris: How is the shift to agentic AI changing what enterprise customers actually need from search.

Sri Desikan: Customers are moving from demo to production in terms of their agentic AI strategy. Though everyone wants to do it, many of them are not ready. The block is in the data. Enterprise data is scattered across various object stores, databases, applications, and most of it is never intended to be used. Organizing it, making it [ready] for production, is important.

Earlier, it was managing structured data. Now with the age of agent AI, it’s more about combining the structured data with unstructured data, such as documents, contracts and images. How do you best make use of it? Taking agentic workflows into production is a big challenge enterprise customers face.

Vrashank Jain: Building on Sri's point, the hard part isn't the agent, it's the context that you feed into it. Context is actually a pretty hard problem to solve. The reason is because the data lives all over the place. It's in structured records and file repos. It’s in an unstructured document, in tickets and logs.

It's increasingly multimodal as well, which adds another level of complexity. If an agent needs to act on your behalf, it [needs] a governed set of data, permission and freshness so you can believe what the agents tells you. Otherwise, it's going to hallucinate, and you'll have no idea because it sounds so realistic. That’s a data problem before it’s an AI problem.

That's one of the reasons we're partnering with Elastic closely at Dell. Rather than building this side ourselves, we want to rely on best of breed here, which is a combination of text search, vector search and hybrid search combined into one.

Matt Harris: What role does data preparation play in the context layer that agents rely on?

Vrashank Jain: [That] plays a much bigger role than ever before. The quality of search on enterprise data is dependent on the data that's feeding into it. We added the data orchestration engine recently to discover, prepare, enrich and govern [data] because we think this is worth investing in. We then hand it over to the Elastic side for search. The metadata we preserve becomes filterable for the agents and models, and the permissions get carried through. If we set a permission all the way at the beginning, we know that it will get respected when an agent comes along and starts asking that question on your behalf.

Sri Desikan: Enterprises are realizing that models are pre-trained on a different set of data. When it comes to applying their enterprise data, they have to do metadata prep. How do you understand where this data lives? That’s the metadata. What’s the shape of that data? That’s the schema. What questions can you ask of that data? That’s human-curated context. You also want to decide what systems contain what kinds of data [and] sample data using LLMS and store them as knowledge indicators.

Data prep becomes crucial as the models get more powerful. The need for the centralized context layer is super important now. How do you search for all this? Humans go search on Google and get ten blue links and you know what links are important. You have that in-built context. Agents don't have that context. They're going to search and get a result, and you have to rank those results.

The spectrum of retrieval, ranking and relevance becomes important. Search becomes the foundation for all of agentic AI. Without search, you can't operate the agents at scale and in production on your private data and feed that to your models. That’s the overall context. We're bringing data to AI and making it more usable for the enterprise.

Matt Harris: What does the Elastic and Dell Technologies partnership give partners that they couldn't go and build on their own?

Vrashank Jain: It’s a great opportunity for partners. Dell brings the end-to-end data foundation. We start with our Dell storage engines, with PowerScale and ObjectScale in the file and object market. The data orchestration engine is part of the purpose-built set of data engines that provide discovery, search and preparation,.

We also have purpose-built data engines, deployment and life cycle integration, and the data search engine powered by Elastic integrated deeply inside that data platform. It's not a bolt on, and that's a deliberate choice for us. Elastic has spent more than a decade improving relevance and retrieval at enterprise scale. We would rather integrate that level of expertise than reinvent a version that we could probably never match without a whole lot of investment and time.

For partners, it means one validated foundation for them instead of a month-long integration or having to piece together various software and hardware components. VARS and system integrators have been assembling this for themselves for a while, but these can become a support burden later on.

Partners value a validated foundation they can build on top of for their services and consulting offerings and for their implementation and value-add. This becomes a perfect underpinning for them to then layer on their value.

Matt Harris: Can you walk us through how this looks in practice when a partner or customer puts this into production?

Sri Desikan: If customers have to build AI data platforms, they have to assemble all the pieces that Dell brings together by themselves. They have to bring in 10 to 15 different technologies. In the age of AI, all of these independent stack items are changing fast. We are adding more and more features every week. If partners use a package offering like this with the Dell AI Data Platform, it comes with best-of-breed products that are fully supported.

We work with Dell to bring the latest offerings we have directly into the AI Data Platform, so the customers don't have to stitch it together themselves. That’s the biggest benefit for resellers and SI’s because they can focus on higher value-added services on top of it versus thinking about how to bring all this plumbing together.

Matt Harris: What would you say separates production grade search from a good demo and what changes with agents?

Sri Desikan: The bottleneck in enterprise AI isn’t the model; it’s the data underneath. So how do you unlock the data? There are two ways to do this. One, you have to operate on unstructured data which means you need the traditional keyword search, but you also need semantic search at scale. Think about thousands of documents and being able to make sense out of it. You want to have agentic search on it, which means you want hundreds of agents operating at millisecond latency in terms of retrieval, retrieving the right context at the right time.

The second part is organizing the data. Elastic provides the context layer. Context Engine allows you to bring data and metadata from various data sources in an enterprise and organize them so that the agents are grounded on your private data as that information is sent to the model.

A combination of hybrid search, semantic search, keyboard search and context layer — all as part of one platform — is what differentiates Elastic search. And given that search is the foundation for AI, agents need to have access to data fast and relevant data. Elastic is built as a real-time, low-latency, high-throughput engine, so we are meeting the market at the right time.

Vrashank Jain: One of the demo shortcuts people often take is that they move all the relevant data into essential repo, consolidate a nice clean copy and then show something really impressive. What you discover is that you've actually built a great demo built on a new silo of data that is now stale. It's lost permissions because you never bothered with it at the beginning, because running towards the answer.

You realize that the search is just one stage in a governed data path. It’s not the destination you're migrating into. It’s just one of the pieces in a long pipeline. Making that work means that the retrieval has to be aware of the permission that were [there] at the beginning. It has to be fast enough that agents can query repeatedly and it needs to fan out across the data that's all over the environment and not in one single place or a single format. [Not] all the PDFs or all documents even look alike. If you go back in time, you'll find documents that were electronically written with multiple formats, and maybe even some handwriting all the way back. That’s the hard part that a lot of us are guilty of skipping when we're trying to create a good demo — and [we] have to rethink when it’s in production.

Learn more about how Elastic and Dell help organizations scale agentic AI production.