# Large Language Models as Software Components: A Taxonomy for LLM-Integrated Applications

**Authors:** Irene Weber  
**arXiv:** [2406.10300](https://arxiv.org/abs/2406.10300)  
**Date:** 2024-06-13  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2406.10300v1/

## Key points

- Large Language Models (LLMs) are increasingly popular as autonomous agents and tools for software engineering.
- Integration of LLMs into applications allows for more efficient performance of challenging or time-consuming tasks.
- There is a lack of established terminology, concepts, and methods specific to LLM-integrated application engineering.
- Irene Weber conducted a study that presents a taxonomy for LLM-integrated applications to address this gap.
- The taxonomy serves as a framework for analyzing and describing LLM-integrated systems, showcasing effective utilization of LLMs within applications.
- The study identifies multiple dimensions along which an LLM component can be characterized, including skills leveraged and format of output generated.
- The taxonomy effectively describes and categorizes LLM-integrated applications, contributing valuable insights to theory building in this field.
- Researchers and practitioners are exploring innovative ways to leverage LLM capabilities within applications despite existing challenges.
- Integrating large language models has the potential to revolutionize traditional software development practices by enabling more efficient and sophisticated systems.

## Summary

Large Language Models (LLMs) have become increasingly popular in recent years as they are utilized as autonomous agents and tools for software engineering. However, a new area of exploration has emerged with the integration of LLMs into applications. This integration allows for tasks that would typically be challenging or time-consuming to code to be performed more efficiently. Despite the growing interest in this field, there is a lack of established terminology, concepts, and methods specific to LLM-integrated application engineering. To address this gap, Irene Weber conducted a study that presents a taxonomy for LLM-integrated applications. This taxonomy serves as a framework for analyzing and describing these systems while showcasing various ways in which LLMs can be effectively utilized within applications. By examining a sample of recent LLM-integrated applications and applying the taxonomy to additional cases, the study identifies multiple dimensions along which an LLM component can be characterized. These dimensions include the skills leveraged from the LLM and the format of output generated. Together, these components define how an application integrates with LLMs. The effectiveness of this taxonomy is demonstrated through its ability to describe and categorize LLM-integrated applications. It offers valuable insights that can contribute to theory building within this emerging field of study. Despite existing challenges, researchers and practitioners continue to explore innovative ways to leverage LLM capabilities within applications. Ultimately, integrating large language models has the potential to revolutionize traditional software development practices by enabling more efficient and sophisticated systems to be built.