I’ve noticed a trend of product professionals quickly leveraging LLMs to generate documents. It’s very tempting: the push of a button that generates a perfectly structured Product Requirements Document (PRD), a complete backlog of user stories, or entries on a roadmap. In fact, I see prototypes for generating the entire lifecycle of product documents from a single prompt. This leads to an ever-increasing stream of artifacts: documents, files, slides, illustrations, and visualizations.
I’m not denying document automation has its place; LLMs make all this a reality. Document generation is an easy, short-term win, which is why it’s the typical starting point for product people using LLMs. I'm not suggesting we stop doing it, but I’m not convinced that it is truly adding much value. Are we simply creating documents faster than the existing bottlenecks in the process? This often results in more and more unread documents piling up.
To be honest, I’ve worked on projects where we wrote up all these detailed product documents to kickstart an effort, but we rarely glanced at them again afterward. In reality, the biggest bottleneck in product development isn't the writing of these documents—it’s usually everything that follows. People may look past the fancy LLM-generated PRD and lament that, despite the AI, the delays persist in getting stakeholders on board, securing necessary approvals, and so on.
If one of the goals of AI is to remove friction, I’m not sure this is it.
It’ll be interesting to watch how this automated document generation plays out in the real, messy world of product management, particularly concerning how the documents are maintained, communicated, and acted upon. Will the LLM own the documents and manage their updates?
If a product person generates a whole lifecycle of documents based on prompts/requirements, and changes occur, how are the documents updated and communicated to the rest of the team who depend on them to inform their work? Perhaps they can auto-update and then auto-communicate the updates? I’m thinking particularly of downstream team members who need specific product functional requirements, such as developers, QA, and documentation writers.
So many questions!
Data as a Starting Point?
For me, one area where LLMs have made a significant difference in the product space is with product and user data. I feel this offers a better starting point because the medium- to long-term value-add is stronger; value can be demonstrated quickly by creating a UI and aggregating data. It is an area you can have more control over, own the process, and operate free from the constraints of stakeholder time or influence.