AI in Documentation Principles

Jun 11, 2025

Summary

  • Alex Garnett, Senior Curriculum Developer at Temporal Technologies, presented on effective and principled uses of AI in documentation, focusing on Retrieval Augmented Generation and the Kappa AI platform.
  • Key discussion points included practical AI integration in documentation workflows, the importance of citations and feedback loops, accessibility, privacy considerations, moderation, and ethical challenges in LLMs.
  • The session included attendee Q&A on workflow adaptation, AI ingestion of community content, the feedback loop process, accessibility, regulated industry constraints, ethical/licensing issues, technical implementation, and moderation strategies.
  • The importance of maintaining high standards, context, and authoritativeness in documentation when leveraging AI tools was emphasized.

Action Items

  • None assigned or volunteered in the transcript.

AI in Documentation: Approaches and Principles

  • AI is increasingly promoted for use in documentation, but common large language models remain prone to errors (“hallucinations”), requiring oversight and domain knowledge.
  • Effective uses of AI include automating repetitive or reactive documentation tasks, enabling writers to focus on their preferred or value-adding activities.
  • The most interesting AI applications go beyond writing to user interaction, such as embedding chatbots in documentation sites that draw answers from existing, trusted corpora.
  • Retrieval Augmented Generation (RAG) allows leveraging a company’s own documentation and community-generated content to inform AI-generated responses while relying on robust citation mechanisms.
  • Integration of support, documentation, and community content (forums, Slack, GitHub, LMS) enhances the knowledge base available to users via AI tools.

Practical Implementation: Kappa AI Platform at Temporal

  • Temporal uses Kappa AI to provide a chatbot interface on documentation sites, enabling users to interactively query across multiple content sources.
  • Kappa’s analytics dashboard provides metrics on usage, certainty of answers, and source distribution (docs, Slack, courses, GitHub, etc.), facilitating ongoing content improvement.
  • The AI flags uncertain answers, encouraging users to further research and highlighting documentation gaps for writers.
  • Temporal did not need significant additional preparation for Kappa ingestion; existing markdown-based Docusaurus docs were compatible by default.
  • The feedback loop from AI analytics is operationalized by reviewing flagged “uncertain” queries and generating Jira tickets to address documentation deficiencies, especially to clarify compatibility and negative cases explicitly.
  • Community-generated content ingestion (e.g., public Slack, Discourse) is pruned regularly to maintain quality and prevent misinformation from contaminating the chatbot’s knowledge base.

Accessibility, Ethics, and Industry Considerations

  • Increasing access routes to documentation via AI is framed as an accessibility gain, provided content remains authoritative and well-moderated.
  • Challenges in regulated industries include the risk of AI-generated misinformation with potentially severe consequences; in these contexts, AI chatbot implementation may be inadvisable.
  • Ethical and licensing concerns are significant: there is skepticism about major actors’ respect for licensing, and the best recourse is to support good vendors and practices that elevate standards, such as robust citation and transparency.
  • Active moderation of community channels ensures quality control, but automating pruning with additional AI may introduce further complexity and unintended consequences.

Decisions

  • No significant decisions or approvals were made or recorded during the session.

Open Questions / Follow-Ups

  • None explicitly called out for follow-up; several attendee questions were answered live, including technical, ethical, and implementation aspects of documentation AI integration.