Abstract
Software documentation is essential for software maintenance, onboarding, and collaboration, but it is often incomplete, outdated, or inconsistent with the current state of a project. Large language models offer new opportunities to assist documentation tasks by generating summaries, explanations, examples, and structured text from software project context.
This position paper proposes a lightweight workflow for AI-assisted software documentation using large language models and Markdown-based repositories. The workflow treats model output as a draft that must be reviewed by developers before it becomes part of the project. It supports common tasks such as README generation, API explanation, changelog drafting, and conversion of issue discussions into documentation updates.
The paper also discusses risks, including hallucinated content, outdated context, incorrect examples, privacy concerns, and excessive trust in generated text. The main contribution is a simple human-centered workflow for integrating large language models into software documentation practices while preserving developer control and review.
Citation
Keywords
- large language models
- foundation models
- generative AI
- software documentation
- Markdown
- software engineering
- human review
Conference Submission
The manuscript was also submitted to FLLM 2026, The 4th International Conference on Foundation and Large Language Models, under the FLLM Main Track.