\n
![](\"https:\/\/images.ctfassets.net\/jdtwqhzvc2n1\/3PNVXTyfSXJhvGjd00Ia1C\/d051128f97407ff20b6b4db84c907811\/Upscaled_already.png?w=300&q=30\")
<\/p>\n
\n<\/p>\n
The scaffolding layer that developers once needed to ship LLM applications \u2014 indexing layers, query engines, retrieval pipelines, carefully orchestrated agent loops \u2014 is collapsing. And according to Jerry Liu, co-founder and CEO of LlamaIndex, that's not a problem. It's the point.<\/p>\n
\u201cAs a result, there's less of a need for frameworks to actually help users compose these deterministic workflows in a light and shallow manner,\u201d Jerry Liu, co-founder and CEO of LlamaIndex, explains in a new VentureBeat Beyond the Pilot podcast.\u00a0<\/p>\n
Liu\u2019s LlamaIndex is one of the foremost retrieval-augmented generation (RAG) frameworks connecting private, custom, and domain-specific data to LLMs. But even he acknowledges that these types of frameworks are becoming less relevant.\u00a0<\/p>\n
With every new release, models demonstrate incremental capabilities to reason over \u201cmassive amounts\u201d of unstructured data, and they\u2019re getting better at it than humans, he notes. They can be trusted to reason extensively, self-correct, and perform multi-step planning; Modern Context Protocol (MCP) and Claude Agent Skills plug-ins allow models to discover and use tools without requiring integrations for every one independently.\u00a0<\/p>\n
Agent patterns have consolidated toward what Liu calls a "managed agent diagram" \u2014 a harness layer combined with tools, MCP connectors, and skills plug-ins, rather than custom-built orchestration for every workflow.<\/p>\n
Further, coding agents excel at writing code, meaning devs don\u2019t need to rely on extensive libraries. In fact, about 95% of LlamaIndex code is generated by AI. \u201cEngineers are not actually writing real code,\u201d Liu said. \u201cThey're all typing in natural language.\u201d This means the layers between programmers and non-programmers is collapsing, because \u201cthe new programming language is essentially English.\u201d\u00a0<\/p>\n
Instead of manual coding or struggling to understand API and document integration, devs can just point Claude Code at it. \u201cThis type of stuff was either extremely inefficient or just would break the agent three years ago,\u201d said Liu. \u201cIt's just way easier for people to build even relatively advanced retrieval with extremely simple primitives.\u201d<\/p>\n
So what\u2019s the core differentiator when the stack collapses?\u00a0<\/p>\n
Context, Liu says. Agents need to be able to decipher file formats to extract the right information. Providing higher accuracy and cheaper parsing becomes key, and LlamaIndex is well-positioned here, he contends, because of its developments with agentic document processing via optical character recognition (OCR).\u00a0<\/p>\n
\u201cWe've really identified that there's a core set of data that has been locked up in all these file format containers,\u201d he said. Ultimately, \u201cwhether you use OpenAI Codex or Claude Code doesn't really matter. The thing that they all need is context.\u201d<\/p>\n
There\u2019s growing concern about builders like Anthropic locking in session data; in light of this, Liu emphasizes the importance of modularity and agnosticism. Builders shouldn\u2019t bet on any one frontier model, or overbuild in a way that overcomplicates components of the stack.\u00a0<\/p>\n
Retrieval has evolved into \u201cagent-plus-sandbox,\u201d as he describes it, and enterprises must ensure that their code bases are tech debt free and adaptable to changing patterns. They also have to acknowledge that some parts of the stack will eventually need to be thrown away as a matter of course.\u00a0<\/p>\n
\u201cBecause with every new model release, there's always a different model that is kind of the winner,\u201d Liu said. \u201cYou want to make sure you actually have some flexibility to take advantage of it.\u201d<\/p>\n
Listen to the podcast to hear more about:\u00a0<\/p>\n
LlamaIndex\u2019s beginnings as a \u2018toy project\u2019 with initially only about 40% accuracy;\u00a0<\/p>\n<\/li>\n
How SaaS companies can tap into complicated workflows that must be standardized and repeatable for average knowledge workers;<\/p>\n<\/li>\n
Why vertical AI companies are taking off and why \u2018build versus buy\u2019 is still a very valid question in the agent age.\u00a0<\/p>\n<\/li>\n<\/ul>\n
You can also listen and subscribe to <\/b>Beyond the Pilot<\/b> on <\/b>Spotify<\/b>, <\/b>Apple<\/b> or wherever you get your podcasts.<\/b><\/p>\n