Sri Desikan of Elastic explained that frontier models continue to improve yet still require proper context. Without it, companies spend heavily on tokens that produce little value because agents cannot locate the right information at the right time.
Simply connecting a company’s data corpus to an MCP server is not enough. Shoving all data into a model’s context window is computationally expensive and slow, so Elastic recommends pre-conforming corporate data into an AI index that knows where data lives and how pieces relate to one another.
The index acts as a container storing metadata about everything in the enterprise. Agents can then fetch correct, useful information faster and with lower token usage. Pre-computing the index also lets companies exclude stale information and gain visibility into how the LLM will reason.
Desikan noted that using AI to prepare data for other AI agents raises privacy and autonomy questions that remain open. The approach is presented as one industry response to the context gap rather than a complete solution.