The context problem: It’s what’s really breaking AI agents at scale

The article contends that AI models and tokens are improving quickly, yet many companies still fail to see the return they expected from AI spending. The gap, according to the writer, is context: agents need the right information at the right moment, quickly, to move beyond demos into dependable automation. Simply linking a company's data to a model through MCP servers is not enough, because stuffing everything into a context window is costly in tokens and slow. The piece presents Elastic's approach as one answer. Company data would be pre-organized into an index holding metadata about where information lives and how it connects, which the author says can speed up agent responses and cut token use sharply in some cases. The index can also be kept free of stale information, which the company says helps teams anticipate how the model will reason. The writer notes that pre-processing data for AI raises privacy and agent autonomy questions, which are explored in an accompanying interview. Overall, the author comes away encouraged that industry responses to these new problems appear well considered.