The Rise of Extensible Web Software Driven by LLMs

Limitations of Static Web Software

Current web software focuses on features for the largest user base, leaving a long tail of individual user needs unaddressed. Adding too many features to a product can also make it overly complex and unusable for the majority of users.

LLMs Enable Personalized Software

Large Language Models (LLMs) have proven effective in building highly personalized applications, referred to as "Software for One." These applications are tailored to individual workflows, bypassing the complexities of enterprise software development. Y Combinator's Pete Koomen identifies this as an opportunity for "Small Software."

LLM-Native Software and Extensibility

An example of "LLM-native software" is Pi, which features a core functionality but allows extensive customization through user requests. This model enables users to create and share their own modifications, leveraging the ability of LLMs to generate code from natural language. Most existing software cannot integrate this capability.

Extensible Software on the Web

While pluggable software exists in local professional tools like IDEs and CAD extensions, the web has not fully adopted this model. The author proposes an opportunity for "Extensible Software" on the web, where LLMs reduce the cost of authoring extensions and modern sandbox primitives ensure secure deployment. This approach allows developers to build a stable core application while enabling users to safely extend it with LLM-generated components.