Aakash Gupta

The Claude Code Analytics Workflow Top AI PMs Don’t Want You to Know

Frank Lee breaks down the complete AI PM workflow using Claude Code and MCP. Deep chart analysis, automated dashboards, customer feedback synthesis, PRD generation, and pushing directly to Linear or code all live demos. Full Writeup: https://www.news.aakashg.com/p/frank-lee-podcast Transcript: https://www.aakashg.com/mastering-analytics-and-claude-code-the-complete-aipm-workflow-with-frank-lee/ --- Timestamps: 0:00 - Intro 1:37 - Guest Introduction 1:47 - Most Powerful AIPM Workflow 3:45 - Setting Up Claude Code + MCP 6:28 - Context Management in Claude Code 9:41 - Ads 11:08 - Top 5 Use Cases for PMs 19:29 - Automating Dashboard Reporting 30:48 - Ads 33:01 - Customer Feedback Analysis 33:54 - Converting Insights into PRDs 39:24 - Pushing to Linear or Code 40:35 - Biggest Mistakes with MCP 45:19 - What Amplitude Is Shipping 50:28 - Outro --- 🏆 Thanks to our sponsors: 1. Amplitude: The market-leader in product analytics - https://amplitude.com/session-replay?utm_campaign=session-replay-launch-2025&utm_source=linkedin&utm_medium=organic-social&utm_content=productgrowthpodcast 2. Pendo: The #1 software experience management platform - http://www.pendo.io/aakash 3. Testkube: Leading test orchestration platform - http://testkube.io/ 4. Product Faculty: Get $550 off the AI PM Certification with code AAKASH550C7 - https://maven.com/product-faculty/ai-product-management-certification?promoCode=AAKASH550C7 5. Bolt: Ship AI-powered products 10x faster - https://bolt.new/solutions/product-manager?utm_source=Promoted&utm_medium=email&utm_campaign=aakash-product-growth --- Key Takeaways: 1. Claude Code + MCP is the most powerful AIPM workflow today - Connect your analytics tool via MCP, load your product context into a repo, and let the agent do analysis that used to take hours in minutes. 2. Deep chart analysis now takes 90 seconds instead of 3 hours - Drop a chart URL into Claude Code, trigger the analyse chart skill, and the agent navigates your data taxonomy, finds anomalies, and hypothesises why metrics changed. 3. Automate your entire weekly business review - Point Claude Code at your dashboards Monday morning. Get 3-5 top insights and the one urgent issue to tackle — no manual dashboard scanning ever again. 4. Customer feedback synthesis across all channels in one pass - Zendesk, Gong, Salesforce, Slack, app stores all unified. Claude Code navigates the MCP, clusters themes, and surfaces what customers love and hate that week. 5. PRDs write themselves from insights - Take the analysis output, point it at your PRD template in Cursor or Claude Code, and get a first draft spec in under 2 minutes. Iterate with command L or command K. 6. Skills are the most important Claude Code feature - A skill is just a named prompt with heuristics and tool instructions. It loads only when relevant, preventing context bloat and giving the agent a repeatable workflow. 7. The biggest MCP mistake is connecting too many servers - Every tool description burns context. Load only what's relevant to the workflow. Remove or hide tools that aren't being used for a given task. 8. MCP is not for complex orchestration - it's for data access - Set the right expectation. MCP connects AI to external systems easily. It's the first step, not the whole pipeline. 9. Granola has no MCP, so build a script instead - Frank used Claude Code to write a local script that dumps Granola meeting notes into his product repo. Now he can pull all meeting context with a single at-command. 10. The future of PMing is vibe PMing - Chart analysis, dashboard reporting, feedback synthesis, spec writing, and prototyping — all agent-driven. PMs who adopt this workflow now will have a massive advantage in 2-3 years. --- 👨‍💻 Where to find Frank Lee: Twitter: https://www.twitter.com/franklee LinkedIn: https://www.linkedin.com/in/franklee/ 👨‍💻 Where to find Aakash: Twitter: https://www.x.com/aakashg0 LinkedIn: https://www.linkedin.com/in/aagupta/ Newsletter: https://www.news.aakashg.com #claudecode #aipm --- 🧠 About Product Growth: The world's largest podcast focused solely on product + growth, with over 200K+ listeners. 🔔 Subscribe and turn on notifications to get more videos like this.

AGAakash GuptahostFLFrank Leeguest

Feb 25, 2026·53m·Watch on YouTube ↗

📖 CHAPTERS

1. 1

   Why Claude Code + MCP is a “10x” analytics workflow for PMs

   Aakash and Frank open by framing the core idea: run your entire product analytics-to-spec workflow inside Claude Code/Cursor by connecting your product context and tools through MCP. They preview the end-to-end arc—chart analysis, reporting automation, feedback synthesis, PRDs, and even shipping prototypes/tickets.

2. 2

   Meet Frank Lee (Amplitude AIPM lead) and what he’s optimizing for

   Frank introduces his role at Amplitude working on agents and MCP products. The conversation sets up a practical, PM-focused tutorial: how to “surface product context” into an AI agent so it can reason over your metrics and product artifacts.

3. 3

   Claude Code, Cursor, Skills, MCP—plain-English breakdown

   Frank explains what Claude Code is (terminal-based coding agent) and what MCP is (Model Context Protocol for connecting models to external tools/data/actions). He also introduces Skills as reusable prompt/tool bundles that standardize repeated workflows.

4. 4

   Initial setup: build a “product repo” as your context source of truth

   Frank demos his setup in Cursor: a dedicated repo with folders per product line and markdown context (roadmaps, specs, vocabulary). This repo becomes the shared substrate that Claude Code/Cursor can reference to brainstorm, analyze, and draft documents consistently.

5. 5

   Creating an Auto-Insights Skill: teaching the agent how to analyze charts

   Frank walks through a real Skill definition: metadata (name/description), a structured prompt, heuristics (spikes, seasonality, anomalies), and tool selection via MCP. The Skill replicates hours of manual exploratory slicing by having the agent fetch related charts and form hypotheses.

6. 6

   Context management realities: tool descriptions, compaction, and session resets

   Aakash and Frank dig into how much context is “too much,” and what actually consumes the context window (often MCP tool schemas/instructions). They cover practical tactics: starting new sessions/tabs, Claude Code’s /compact, and writing progress to a markdown file before compaction fails.

7. 7

   Ingesting meeting notes without a native MCP: the Granola script workaround

   Frank shows how he uses Claude Code to create scripts that pull meeting notes (e.g., from Granola) into his repo by date-based folders. Once notes are local files, the agent can reference entire folders for summaries, action items, and downstream PRD inputs.

8. 8

   Parallel agent orchestration: where Conductor and Git worktrees matter (and when they don’t)

   Frank explains Conductor’s value for running many coding agents in parallel without branch conflicts, using Git worktrees. He notes that most PM tasks don’t require heavy parallel orchestration, but it becomes crucial for multi-agent coding work.

9. 9

   The “Top 5” PM analytics use cases (the full loop)

   Frank lays out the five repeatable workflows: anomaly investigation, automated weekly reporting, qualitative feedback synthesis, converting insights to PRDs/actions, and finally routing work to Linear or prototyping in code. This becomes the blueprint for the rest of the demos.

10. 10

    Demo 1: Deep chart analysis—investigating a metric spike via Amplitude MCP

    Frank demonstrates dropping an Amplitude chart link into Claude Code and triggering an ‘analyze chart’ Skill. The agent parses the URL, queries underlying data, explores properties and related charts, and produces a structured explanation with hypotheses (e.g., feature flags, data quality, power users).

11. 11

    Demo 2: Dashboard reporting automation—TL;DRing a dashboard and scheduling agents

    They switch to dashboard-level synthesis: an ‘analyze dashboard’ Skill fetches charts in batches to manage context limits and summarizes key movements. Frank then shows Amplitude’s scheduled dashboard agents that push insights to Slack/email, reducing the need for recurring WBR slide prep.

12. 12

    Demo 3: Customer feedback analysis—unifying Zendesk/Intercom/Salesforce/G2 into one view

    Frank demonstrates using an ‘analyze feedback’ Skill over Amplitude’s AI Feedback product to pull recent themes, issues, bugs, and delights. The key value is centralized ingestion plus flexible, queryable analysis (e.g., focus only on praise, competitors, or urgent issues).

13. 13

    Demo 4–5: From insights to PRDs, then to Linear or prototypes (Cursor/Claude Code)

    They generate draft PRDs/specs directly into markdown files, then iterate to improve quality (reduce verbosity, prioritize acceptance criteria). Finally, they discuss routing: push tickets to Linear via MCP, message engineers, or even prototype changes in code using Cursor/Claude Code agents.

14. 14

    Common MCP mistakes and the real critiques (context bloat, auth pain, configuration)

    Frank outlines two major user errors: expecting MCP to do everything, and connecting too many irrelevant MCP servers/tools. He then addresses broader criticisms—hype, context waste, auth friction, and configuration—arguing the ecosystem is rapidly improving (managed connectors, dynamic tool calling, Skills).

15. 15

    What Amplitude is shipping next: embedded global agent, sub-agents, and Slack-first access

    Frank previews Amplitude’s broader agents launch: a global embedded agent across the platform, plus specialized sub-agents (dashboard, session replay, feedback, website optimization). MCP underpins both internal agent capabilities and external workflows in Claude/Cursor/ChatGPT, with strong emphasis on “agents where you work,” especially Slack.

16. 16

    Closing: “Vibe PMing” as the future operating model

    Aakash wraps by positioning this as the new default for PM effectiveness: integrate analytics + qual data + spec writing + execution through agents. He encourages viewers to request access (Claude Code, Cursor, GitHub, MCP-enabled tools) and points to follow-up resources and the newsletter walkthrough.