By Teresa Torres · producttalk.org · @ttorres on X · LinkedIn
Teresa Torres interviews Matthias and Charlotte Kleverud, co-founders of Momental, about building a "GitHub for product management" that ingests documents, transcripts and recordings to map goals, decisions and ownership into a structured context layer. The system uses AI agents to detect conflicting strategic bets across teams and surfaces them for human resolution, much like merge conflicts in code. The conversation traces the company's path from early GPT-3 experiments in 2022, through a multi-agent team that revealed agents face the same alignment problems as humans, to a document-processing agent that maintains a living knowledge graph. It matters because misalignment between teams is one of the costliest and least visible failure modes in product organizations.
01Key takeaways
- Treat strategic misalignment like a merge conflict: surface divergent bets early so humans can resolve them before they derail a quarter.
- Attach metadata such as speaker, date and context to every piece of organizational knowledge to reduce AI hallucinations.
- Model how signals turn into learnings, decisions and principles so decision reasoning stays traceable.
- Expect chunking and basic RAG to break down at organizational scale and invest in structured context models instead.
- Build feedback loops that let agents improve their own prompts from regular user input.
- Favor UI-first and proactive agent designs over chat-only interfaces for product workflows.
02Key sections
- The GitHub-for-PM concept
- The founders frame the product as a way to find merge conflicts in strategy rather than in code. Conflicting bets, such as one team optimizing retention while another optimizes conversion, get flagged early.
- The product chain and context model
- Momental models how signals become learnings, decisions and principles, and organizes organizational knowledge into interconnected trees for goals, decisions and people/time.
- Origin story and pivots
- The company started in 2022 with early GPT models, pivoted through a multi-agent team, and discovered that agents need aligned context just as humans do.
- How the document agent works
- An OODA-loop-driven agent extracts and connects context across documents, with metadata about speaker and timing used to reduce hallucinations, and RAG-style chunking is avoided at scale.
- Human-in-the-loop and product design
- Conflicts are auto-resolved or escalated with merge options, feedback is collected weekly to rewrite agent prompts, and the design moves from chat-first to UI-first to proactive agents.
03From the post
“Listen to this episode on: Spotify | Apple Podcasts What if an AI could spot the moment two product teams start pulling in opposite directions -- before it derails a quarter? In this episode of Just Now Possible, Teresa Torres talks with Matthias and Charlotte Kleverud, co-founders of Momental, about their”
04Frameworks mentioned
Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.