“The frontier models are getting better and better and better. But they're like a giant beast, and you need to feed it the context. If you don't feed it the context, all the money that you're spending on tokens is going to waste.”
The New Stack Podcast · Hands-On Engineering Podcasts · November 2026
Context
Alex Williams asked how far the average enterprise has got in moving AI agents past pilots and demos. Sri Desikan says the main obstacle is getting company data organized and ready for AI, not the models themselves.
Transcript
…AI company, integrates its expertise in search technology with artificial intelligence to help everyone transform data into answers, actions, and outcomes. Elasticsearch AI platform, the foundation for its search, observability, and security solutions, is used by more than 50% of the Fortune 500.
…have questions. And hopefully we can figure out what's breaking in this enterprise push for Agent capabilities and how we can fix it. So first of all, starting with AI pilots, demos, and prod, how far along is the average enterprise pursuing AI in terms of moving past the pilot or demo stage today?
Yeah, I would say they are in a better stage than they were last year, right? Things are moving in the right direction. But I think the biggest challenge today is organizing the data and making it ready for AI. And that's still a pretty big challenge because on one side, you have the AI moving very fast, meaning the frontier models are getting better and better and better. But they're like a giant beast, and you need to feed it the context, right? If you don't feed it the context, all the money that you're spending on tokens is going waste. And I think this is where the enterprises are trying to figure out, well, I need to get my data in order first, right? I have it sitting in 15 different repositories. I have documents. I have PDFs. I have images. I have relational data. And then I have Elastic. How do I make sense of all this? Feed the right context to the LLMs and get better ROI.
Just to extend the analogy a little bit, if you have these thundering Goliaths, these Frontier Labs, and you don't feed them context, then they eat you. And in this case, that would be eating your budget.
Pretty much. Yeah, and taking
away all of your money. Token maxing. So it seems that the key stumbling block here is not the models at all. And they've reached a level of intelligence that is sufficient to run the Agentic workloads that we want to throw at them. So we have the cars, but not the gas, perhaps?
More from The New Stack Podcast
