Bliki: Interrogatory LLM
Summary
An "Interrogatory LLM" is a technique where a large language model is prompted to ask a human questions to gather necessary information or validate existing content. This approach, described on May 14, 2026, offers two primary applications. First, an LLM can interview a human to generate detailed context, such as multi-page markdown descriptions for complex feature designs, which can then be used by another LLM for subsequent tasks. Harper Reed's blog notably inspired this, emphasizing the LLM asking one question at a time. Second, an interrogatory LLM can review documents like software specifications by interviewing a human expert to assess accuracy, providing an alternative to traditional document review, which many find challenging. The technique also extends to helping individuals who struggle with writing to externalize their knowledge into consumable forms.
Key takeaway
For AI Architects or Technical Leads designing complex systems, consider deploying an Interrogatory LLM to streamline context gathering for new features or to validate existing documentation. This method can significantly reduce the human effort in writing detailed specifications or reviewing lengthy documents, especially when experts find direct writing or reading challenging. Implement a "one question at a time" protocol for optimal results, ensuring efficient and accurate information extraction from human experts.
Key insights
An Interrogatory LLM interviews humans to gather context or validate documents, streamlining information extraction and review.
Principles
- LLMs can proactively gather context.
- Interviewing aids human information extraction.
- One question at a time improves LLM interviews.
Method
Prompt an LLM to ask questions to a human. Feed initial information, direct to other sources, then generate a context report or validate a document.
In practice
- Generate feature design context.
- Validate software specifications.
- Help non-writers externalize knowledge.
Topics
- Interrogatory LLM
- Context Generation
- Document Validation
- Knowledge Elicitation
- Human-AI Interaction
- LLM Workflows
Best for: AI Engineer, AI Architect, Director of AI/ML
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by Martin Fowler.