Elastic exec: enterprise AI needs "context engineering," not bigger prompts

Elastic’s head of field engineering Ravindra Ramnani has spent the last year helping enterprises move AI from demos to production. What he’s found isn’t a shortage of model size or compute power, but a deeper problem: most enterprise data isn’t ready for AI, and most AI isn’t ready for enterprise data. The solution, he argues, isn’t bigger prompts but a new discipline he calls “context engineering.”

The framing is more than semantic. For the last two years, the enterprise AI conversation has been dominated by “tokenmaxxing”—the race to stuff ever-larger prompts with as much context as possible. Vendors have sold this as the path to more accurate, more useful AI. But Ramnani’s experience suggests the opposite. Enterprises that try to brute-force their way through AI adoption with massive context windows quickly hit diminishing returns. The real bottleneck isn’t how much data the model can hold in memory, but how well that data reflects the messy, unstructured reality of enterprise systems.

Context engineering, as Ramnani describes it, is the work of shaping data so it’s actually useful to AI. That means converting analytical data into formats models can understand, mapping relationships between disparate systems, and encoding human expertise in ways that don’t require manual prompt engineering. It’s less about the model and more about the plumbing around it. If the last wave of enterprise AI was about picking the right foundation model, the next one will be about building the infrastructure to feed it.

This shift aligns with broader trends in the market. Some startups, for example, appear to be targeting similar challenges—whether by automating data conversion, improving governance, or tracking AI costs. The underlying issue seems consistent: enterprises struggle when AI systems lack the right context or when data isn’t structured for practical use. How companies address this—whether through infrastructure, tooling, or governance—may define the next phase of adoption.

What makes Ramnani’s argument notable is that it comes from a company that sells search infrastructure, not AI tooling. Elastic’s pitch to enterprises has long been about making data searchable and usable. Now, with AI adoption stalling in production, Elastic is positioning itself as the connective tissue between legacy systems and AI. That’s a smart pivot, but it’s also a crowded space. Many enterprise AI startups are now trying to solve some version of the same problem, whether they frame it as data engineering, governance, or cost management.

The open question is whether enterprises will buy into this as a distinct category. So far, most AI adoption has been driven by bottom-up experimentation—teams trying to solve immediate problems with off-the-shelf tools. Context engineering, by contrast, requires upfront investment in data infrastructure. That’s a harder sell in a market where many enterprises are still in pilot purgatory. But if Ramnani is right, it’s also the only way out. The next phase of enterprise AI won’t be about who has the biggest model, but who can make their data actually work with it.

Sources: yourstory.com