Good idea, I went a step further and created a GPT plugin that would take in a PRD and convert into a prompt spec. Here's the link, try out and let me know
This will be a norm. The caveat which I see currently is LLMs have this habit to suggest your next prompt and trying to be cheeky :) That's where human intent is so important to stick to while you are building product through prompts. I am learning it and keeping LLMs on the check to stick to my intent.
Thanks for sharing this article—I really enjoyed it. It’s made me rethink the way we document and learn as we build. Compared to the traditional approach, I find prompt sets brilliant for encouraging bottoms-up experimentation. But how do we ensure they don’t fragment product thinking? A PRD (hopefully well written!) forced you to connect the dots — between features, edge cases, and the broader narrative. How do we protect that connectivity in this case?
Great read, and interesting take. I agree, the encoding of judgement-inducing questions that comes from such a prompt could even uncover different paths to a solution that a PRD may not uncover.
Good idea, I went a step further and created a GPT plugin that would take in a PRD and convert into a prompt spec. Here's the link, try out and let me know
https://chatgpt.com/g/g-68b1c4e846f88191885be9e950efcca1-prompt-spec-architect
I used the below PRD to test it out. This is a gap in Microsoft teams that could be a great value add that I thought of
Write an example spec based on the above principle for the below PRD
Product Requirements Document (PRD): AI-Powered Meeting Insights with LLM Integration in Microsoft Teams
Overview
Integrate a Large Language Model (LLM) into Microsoft Teams to analyze meeting transcripts, understand intent, and provide:
Alternate dimensions and ideas for discussion topics
Strategic plans and suggestions
Missed points identification with value addition
Relevance scoring for discussed items
Key Features
Meeting Transcript Analysis: Analyze meeting transcripts using an LLM to identify key topics, intent, and context.
Idea Generation: Generate alternate dimensions and ideas for discussion topics.
Strategic Planning: Provide strategic plans and suggestions.
Missed Points Identification: Identify potential missed points and their value addition.
Relevance Scoring: Score discussed items based on relevance to the context.
Functional Requirements
LLM Integration: Integrate a suitable LLM with Microsoft Teams.
Transcript Ingestion: Ingest meeting transcripts.
Contextual Understanding: Analyze meeting context.
Output
Meeting Summary: Summary of key points.
Alternate Ideas: Alternate dimensions and ideas.
Strategic Plans: Strategic plans and suggestions.
Missed Points: Potential missed points with value addition.
Relevance Scorecard: Scorecard for discussed items with relevance scores.
Relevance Scoring
Scoring Algorithm: Develop an algorithm to score relevance (e.g., 1-5).
Factors: Consider factors like context, intent, and impact.
Value Addition for Missed Points
Potential Impact: Estimate potential impact of missed points.
Recommendations: Provide recommendations for incorporating missed points.
Success Metrics
User Adoption Rate: Measure adoption.
Insight Quality: Evaluate relevance and usefulness.
This will be a norm. The caveat which I see currently is LLMs have this habit to suggest your next prompt and trying to be cheeky :) That's where human intent is so important to stick to while you are building product through prompts. I am learning it and keeping LLMs on the check to stick to my intent.
Thanks for sharing this article—I really enjoyed it. It’s made me rethink the way we document and learn as we build. Compared to the traditional approach, I find prompt sets brilliant for encouraging bottoms-up experimentation. But how do we ensure they don’t fragment product thinking? A PRD (hopefully well written!) forced you to connect the dots — between features, edge cases, and the broader narrative. How do we protect that connectivity in this case?
Great read, and interesting take. I agree, the encoding of judgement-inducing questions that comes from such a prompt could even uncover different paths to a solution that a PRD may not uncover.
Someone at OpenAI did a talk about this recently as well. Calling prompts the new requirements. Great read.