PRODUCT NAME: Rovo AI - Atlassian WHAT IT DOES: Helps write Wiki docs and Jira tickets WHAT CAN'T BE FULLY SPEC'D: With multiple formats existing for both writing documents including but not limited to PRDs, Technical docs, business docs as well as Epics and user stories. This is what an assistant will need to do to match the expectations of the writer: 1. Understand the context 2. Understand the writing style 3. Format of the content 4. Depth of understanding the type of product 5. End user type 6. Predict output There are multiple factors involved in writing/re-writing a document for varied purposes and user types, this will need to include varied outputs. Think of an article being written in a newspaper and an editor reviewing and re-writing bits to improve it. WHY AI (NOT SOMETHING SIMPLER): Everything about this problem can revolve around prediction of what user wants rather than it being a simple output. It will have to be reviewed multiple times and given feedback on to produce multiple outputs which is a problem of relativity and context instead of algebra. A formula based product would be lacking here as in case of early chatbot assistants which could just be used to gather context rather than understanding it. HOW IT HANDLES BEING WRONG: With reference to AI, It might be better to use the terms Acceptable and Unacceptable. This product has a way of tracking both kinds of responses with ‘Replace’ prompt being an ideal way to track any output produced that is acceptable to be directly used into the documents created. ‘Thumbs down’ or abandonment of the output or providing AI multiple iterations of prompts being used as way to track unacceptable responses. ONE THING THE TRADITIONAL PM WOULD MISS: A traditional PM might just focus on the output to be in a set format for different types of documents or tickets, track the keywords in the output. This would greatly limit the capability of the product.