Harnessing Conversational AI: The Interrogatory LLM Approach

Introduction

Large language models (LLMs) are increasingly used for complex tasks that require extensive context, such as designing a new feature or implementing a guideline. Typically, a human must manually write this context—often spanning multiple pages of documentation. However, an alternative strategy is emerging: instead of having humans compose the context, we can let the LLM itself interview the human to extract the necessary information. This technique, known as an interrogatory LLM, flips the conventional workflow and offers several advantages for both context creation and review.

Harnessing Conversational AI: The Interrogatory LLM Approach — Source: martinfowler.com

The Interview Method for Context Generation

In this approach, the LLM is prompted to act as an interviewer, asking the human a series of questions to gather all the required details. The model can request descriptions of user-facing features, implementation guidelines, external systems to consult, and any other relevant data. Once the interview is complete, the LLM compiles a comprehensive context report that can be fed into another session—perhaps using a different model—to execute the next step of the task.

Harper Reed first popularized this technique on his blog, emphasizing a key constraint: the LLM should ask only one question at a time. Practitioners note that models frequently need reminders to adhere to this rule, but it helps maintain a focused and manageable conversation, preventing the human from becoming overwhelmed.

One Question at a Time: Ensuring Clarity

Requiring the LLM to pose single questions may seem trivial, but it significantly improves the quality of the interaction. It mimics a structured interview, where each answer builds on the previous one. This method also reduces cognitive load on the human, allowing for thoughtful responses rather than skimming multiple inquiries. The resulting context tends to be more thorough and accurate because the LLM can adapt follow‑up questions based on the human’s replies.

Document Review Through Dialogue

The interrogatory LLM is not limited to generating new context. It can also be employed to review existing documents, such as software specifications. In this scenario, the LLM is given a document that captures domain knowledge, then interviews a human expert to verify its accuracy. This dialogic approach replaces the traditional method of having the expert read and critique the document—a task many find tedious and difficult, especially if the text is poorly written.

By conversing with the LLM, the expert can more naturally identify errors, omissions, or ambiguities. The LLM guides the discussion, probing each section and prompting the human to reflect on correctness. The result is often a more thorough review than a simple read‑through would achieve, and it accommodates people who struggle with long, dense documents.

Broader Applications for Knowledge Capture

While the interrogatory LLM is valuable for creating and assessing context in technical workflows, its utility extends far beyond. Many individuals find writing to be a challenging and even painful process. For such people, an LLM interview offers a more accessible way to transfer knowledge from their minds into a structured, shareable format.

Instead of forcing them to draft emails, reports, or documentation, the LLM can conduct a guided conversation that captures the essential information. The output may carry an unmistakable “AI‑writing” tone—something that seasoned writers often dislike—but it is far better than having the information lost or rushed into an unusable form. This method democratizes knowledge sharing, enabling those for whom writing is a barrier to contribute effectively.

Conclusion

The interrogatory LLM represents a paradigm shift in how we interact with AI for complex tasks. By turning the tables and having the model ask questions, we can generate rich context, review documents more effectively, and capture expertise from people who find writing difficult. Whether applied to software development, business analysis, or general knowledge management, this technique leverages the conversational strengths of LLMs to bridge the gap between human expertise and machine processing.

As the technology matures, we can expect to see more tools and frameworks that support this interactive approach, making it a standard practice for knowledge work. The key takeaway is clear: sometimes the best way to get information from a human is to let the AI do the talking.