How I AIHow this PM uses AI for PRDs, JIRA tickets, and replying to coworkers | Dennis Yang (Chime)
CHAPTERS
- 0:00 – 3:37
Cursor as a writing-first AI hub for PM work
Claire introduces Dennis Yang’s unusual use of Cursor: not for coding, but as a central workspace for PM writing and coordination. Dennis frames Cursor as the UI that connects models, files, and day-to-day PM systems so he can ship faster communication and better documentation.
- •Cursor used primarily for writing and PM workflows, not code
- •Value proposition: models + filesystem + rules + tool connectivity
- •Goal: reduce time on status/translation while improving clarity across the org
- 3:37 – 4:58
Why Cursor wins for PM workflows: models, files, rules, and MCP interoperability
Dennis explains why Cursor is his preferred AI interface: easy model switching, persistent artifacts in a filesystem, customizable Cursor Rules, and deep interoperability via MCP. They also note desktop performance and the “three-pane” mental model that fits PM work.
- •Access to multiple frontier models in one place
- •Filesystem-based artifacts enable drafting and iteration
- •Cursor Rules standardize how the AI works for you
- •MCP connects to Jira/Notion/Confluence/Figma/GitHub
- •Desktop app speed improves the day-to-day experience
- 4:58 – 6:38
Daily setup for non-coders: panes, MCP toggles, and reliability realities
They walk through Cursor’s layout and a practical trick: docking the settings/tools pane so you can quickly confirm MCPs are “green.” The conversation also surfaces the real-world friction—MCPs can be flaky—so quick enable/disable and scoping active tools becomes a core habit.
- •Three panes: files (left), artifact (center), chat (right)
- •Bottom pane used for Cursor settings/tools instead of terminal
- •Checking MCP health (green/red) is part of the workflow
- •Toggling MCPs on/off is a common troubleshooting step
- •Limiting enabled tools reduces ambiguity and wrong-tool calls
- 6:38 – 7:41
MCP stack tour: Atlassian, Notion, Figma, GitHub—and custom MCPs
Dennis lists the MCP servers he relies on, including Jira/Atlassian and Notion, plus Figma and GitHub. He also shares that he built custom MCPs (News API and Semrush) using Cursor, highlighting ‘AI building tools for AI’ as a practical pattern.
- •Two Atlassian/Jira MCP options (including local/server variations)
- •Notion MCP for publishing and retrieval
- •Figma MCP for design context (when Figma is running)
- •GitHub MCP for repo interactions
- •Custom-built MCPs (News API, Semrush) authored with Cursor
- 7:41 – 9:54
Markdown-first workflow: making docs readable with live preview
Dennis explains why Markdown is the lingua franca for LLM-assisted writing and why rendering matters for human review. He shows the Markdown Preview Enhanced extension and the settings that auto-open a preview pane, making an IDE feel like a polished doc editor.
- •LLMs work especially well with Markdown structure
- •Markdown preview reduces cognitive load vs raw syntax
- •Markdown Preview Enhanced extension recommended
- •Cursor settings: auto-preview and quick preview access
- 9:54 – 10:24
PRDs in Cursor + Git: source control for product artifacts
They move into Dennis’s end-to-end flow: drafting PRDs in Cursor, keeping them in Git, and treating docs like versioned artifacts. A key idea emerges: keeping product artifacts adjacent to code can keep engineers and coding agents continuously grounded in context.
- •PRDs authored in Cursor and stored as repo artifacts
- •Git provides change tracking/version history for docs
- •Docs next to code improves discoverability for engineers/AI
- •Idea: Cursor Rules could nudge teams to update PRDs as learning evolves
- •Discussion of ‘source of truth’ shifting toward repos + markdown
- 10:24 – 22:00
Publishing PRDs to Confluence/Notion via MCP—and the importance of interoperability
Dennis demonstrates pushing a Markdown PRD into Confluence (and also supports Notion) so the broader company can comment. They emphasize interoperability as the winning property of future tools—PM content shouldn’t be locked inside any single SaaS.
- •Confluence/Notion used because teams have different preferences
- •MCP enables one-command publishing from Cursor
- •Tool descriptions help users build a mental model of what AI can do
- •Interop is a key selection criterion; avoid content lock-in
- •Round-trip pattern: publish and then write the URL back into the PRD
- 22:00 – 25:52
AI-assisted comment triage and replying in Confluence (human-in-the-loop)
Dennis shows a workflow where Cursor reads Confluence comments, categorizes them by priority, drafts suggested replies, and posts responses as Dennis after approval. They discuss inserting the human at the highest-leverage moment—judgment and prioritization—while offloading the mechanical writing.
- •AI reads PRD comments and organizes them (high/medium/clarifications)
- •Drafts replies that Dennis reviews/approves
- •Authenticated MCP posts replies as the user
- •Human-in-the-loop placement: judgment over typing
- •Ruthless prioritization supported by AI triage
- 25:52 – 30:18
From PRD to Jira: generating epics and well-structured story tickets
Next, Dennis uses Jira MCP to translate the PRD into an epic and associated stories. He notes that the AI-generated tickets are often more thorough than what he’d write under fatigue—complete with acceptance criteria and structured detail—while Cursor Rules prevent common mistakes like orphaned tickets.
- •Command: read PRD and create epic in a target Jira project
- •Auto-generates story tickets from PRD sections
- •Tickets include clear descriptions, Gherkin/acceptance criteria
- •Cursor Rules enforce conventions (e.g., link stories to parent epic)
- •Reduces PM ‘translation toil’ and improves engineer handoff
- 30:18 – 35:14
Automated status reports from Jira using JQL + reusable Cursor Rules
Dennis demonstrates generating a status report by querying Jira activity since a date or for a specific epic. Over time, a refined Cursor Rule becomes a repeatable weekly status workflow, and a cultural side effect appears: engineers add better updates in Jira because they know it’s being used.
- •AI writes JQL and synthesizes progress into a narrative status report
- •Iterate interactively, then codify into a Cursor Rule template
- •Works even with minimal ticket metadata (title + state changes)
- •Improves communication upward and across teams while saving time
- •Encourages better async documentation habits in Jira
- 35:14 – 40:36
Morning briefing in ChatGPT: memory as a prototype—and its failure modes
Dennis shares a lightweight daily habit: asking ChatGPT for a morning briefing “based on what you know about me,” leveraging memory and a long-running project thread. They observe that quality can drift as context grows or models change, making this a useful personal way to study AI reliability and UX.
- •Simple daily prompt: “morning briefing” in a ChatGPT project
- •ChatGPT memory can make briefings surprisingly personalized
- •Over time it may ‘lose the plot’ (relevance drift)
- •Using AI daily builds intuition about model changes and weaknesses
- •Habit doubles as personal productivity + product learning loop
- 40:36 – 46:42
The ‘super MVP’ method: prototype agentic products inside Cursor (zero code)
Dennis explains how he prototypes real AI products without writing code by using Cursor’s tool-calling and model switching. He writes a PRD, asks Cursor to produce a TDD, then creates “super MVP” instructions that call MCP tools (e.g., News API) to generate a working briefing—fast enough to iterate before engineering implementation.
- •Start with PRD → have Cursor draft a TDD (technical design doc)
- •Create ‘super MVP agent instructions’ as executable guidance
- •Use MCP tools (News search) to gather data, summarize, compile report
- •Add practical guardrails (e.g., run terminal date to avoid time confusion)
- •Swap models (Claude/GPT/Gemini) to compare outputs with same instructions
- 46:42 – 50:07
Lightning round: transparency, getting started with Cursor, and prompting style
They close with rapid-fire questions: colleagues know Dennis uses AI to reply, his recommended Cursor ‘blank directory’ quickstart for PMs, and his calm, polite prompting approach. The episode ends with where to find Dennis and final show wrap-up.
- •Colleagues are aware he’s ‘fully AI-enabled’ in workflows
- •Quickstart: create a blank folder for product artifacts and begin chatting
- •Prompting style: polite, reset threads when things go off rails
- •Practical mindset: restart and simplify rather than fight the tool
- •Where to find Dennis (social + upcoming conference mention)
