Jerry Liu, co-founder and CEO of LlamaIndex, says the scaffolding layer developers built for LLM applications is collapsing. That includes indexing layers, query engines, retrieval pipelines, and orchestrated agent loops. Liu frames this collapse as inevitable progress, not failure.
The shift reflects how LLM capabilities have matured. Developers no longer need lightweight frameworks to compose deterministic workflows. Larger models handle complexity that once required intricate engineering layers.
Liu argues context becomes the real competitive advantage as scaffolding disappears. Companies that build better systems for managing context in LLM applications will win. LlamaIndex, which raised funding as an indexing and orchestration platform, is repositioning itself around this insight.
The market opportunity expands here. As basic scaffolding commoditizes or gets absorbed into model capabilities, startups must move upstream. They need to solve context management, knowledge organization, and semantic retrieval at scale. The winners won't be framework companies. They'll be platforms that help enterprises maximize what LLMs can do with their proprietary data.
Liu's thesis predicts winners and losers across the AI infrastructure stack. Shallow abstraction layers disappear. Deep context systems survive.