This is the way…..
- Have three day planning meeting with 15 people, for a large product with 2000+ people involved in total
- Use Jira or other tracking software to refine Market Problems and Features
- Use auto-transcribe for most meetings, but have a couple of non-recorded meetings to discuss sensitive topics
- Share the transcripts with the 15 people
- Figure out that the MCP server you’re using doesn’t work with watchers in Jira
- Use Claude code to extend the MCP (it tells you it’s simple)
- Install the modified MCP server locally
- Use the new extended MCP server to….
- For every Market Problem your team is actively working on, have your agent Analyze the call transcripts and pull out: 1. all applicable points in each meeting 2. Action Items 3. Important notes to stress
- Make it a Claude Skill so that it’s a single command for all future comments
- Do the same for Market Problems that you’re refining
- Submit a PR to Atalassian with these changes for their MCP Server, to make it more useful in an Agentic workflow
This took hours and hours and hours before, and no matter how unbiased you are as a Product Manager, Marketer, Engineer, Director, or Vice President, lots of details and nuance got lost. Most of the time, nobody ever did the summaries. The people in the meetings had a fuzzy understanding of what the team was doing, and the rest of the wider team was never communicated the details the level they needed. Now we can do it right! We can communicate better in large organizations tackling super difficult business problems.
AI is still non-deterministic, you must read each comment and have it ask you for approval (and changes), but this is WAY better information than we were operating with before Agentic AI.