A PM at Amplitude connected Claude Code to his analytics stack. What happened next took 90 seconds instead of 3 hours.

An agent parsed a chart URL, pulled the underlying data, cross-referenced customer feedback, checked feature flag changes, and returned a structured root cause analysis. The entire thing took 90 seconds. A data analyst doing the same work manually would need three hours.

That wasn’t a demo. That was a live workflow from Frank Lee, a principal PM at Amplitude who builds MCP and agent products for a living. I recorded the whole thing on screen, and what Frank showed in that session fundamentally changed how I think about what PM work should look like in 2026.

Vibe coding changed engineering. Vibe PMing with Claude Code just changed PM.

What the vibe of PMing actually looks like on screen

Frank’s setup connects Claude Code directly to his analytics platform, his ticket system, and his customer feedback tools through MCP servers. The AI doesn’t just answer questions about the data. It pulls the data, processes it, cross-references multiple sources, and delivers structured output that a PM can act on immediately.

The deep chart analysis workflow was the one that made me pause the recording. Frank pointed the agent at a single chart URL showing a metric decline. The agent pulled the underlying data, identified the inflection point, cross-referenced customer feedback from the same time period, checked whether any feature flags had changed, and came back with a structured analysis explaining the most likely cause.

That investigation would normally involve opening three tools, writing a SQL query, waiting for the data team, reading through Zendesk tickets, and then spending an hour connecting the dots yourself. Frank got the same answer in 90 seconds from a single prompt.

Then he took that analysis, combined it with screenshots of the current product experience, and had Claude Opus brainstorm the solution. Two rounds of feedback later, a full PRD drafted in his team’s exact template. Twenty minutes from insight to spec.

The workflow that killed me personally

Automated Monday business reviews.

At Epic Games, I spent three to five hours every Sunday night pulling metrics manually for the weekly business review. Opening five dashboards. Screenshotting the important charts. Writing up the week-over-week changes. Identifying the one urgent thing that needed executive attention. Every Sunday. For years.

Frank’s setup eliminates this entirely. You point a dashboard agent at the five dashboards you care about. Every Monday morning, a clean report hits your inbox with the top insights, week-over-week changes, and the one urgent thing that needs immediate attention. The agent writes the narrative. The agent flags the anomalies. The agent compares this week to last week and tells you what changed and why.

The Sunday night ritual that every PM dreads just became a Monday morning email that writes itself.

Five complete workflows, start to finish

Frank walked through five workflows during the episode, and each one represents a different piece of the PM operating system that most people are still doing by hand.

1. Deep chart analysis. Point the agent at a chart URL showing a metric change. Get a structured root cause investigation in 90 seconds that cross-references data, feedback, and feature changes automatically.
2. Automated dashboard reporting. Connect your key dashboards to a scheduled agent. Get a weekly report with insights, trends, and flagged anomalies delivered to your inbox before you open your laptop on Monday.
3. Customer feedback synthesis. The agent processes feedback from Zendesk, Gong call transcripts, NPS surveys, and app store reviews in a single prompt. It clusters by theme, assigns severity, and surfaces patterns that only become visible when you analyze all sources together instead of checking them one at a time.
4. Insight to spec. Take the analysis output from any of the above workflows and feed it directly into a PRD. The agent drafts the spec in your team’s template using the data it already processed. No copy-pasting between tools. No starting from a blank document.
5. Spec to shipped. Once the PRD is reviewed, the agent routes it to Linear as a ticket or prototypes the solution in Cursor. The workflow goes from analysis to spec to engineering handoff without a single context switch.

What makes this different from using ChatGPT as a research assistant

The PMs who use AI as a research assistant are asking questions and getting answers. They copy the useful parts into their docs. Their workflow is fundamentally the same as 2023, just with an AI tab open next to their Google Doc.

Frank’s setup is different because the AI is connected to the actual data sources. It doesn’t answer from general knowledge. It pulls real numbers from real dashboards, reads real customer tickets, and cross-references real feature flag changes. The output isn’t a generic analysis. It’s a specific analysis based on what actually happened in Frank’s product last week.

That connection layer is what MCP servers provide. Without them, you’re asking an AI to help you think. With them, you’re asking an AI to do the work and bring you the results.

The compounding advantage

The PMs who wire up this kind of system in the next six months are going to operate at a level that’s genuinely hard to compete with using manual workflows.

Not because the tools are secret. Anyone can set up Claude Code and MCP connections. But because the system gets smarter over time. Every CLAUDE.md rule you add improves future output. Every skill you build saves time on the next hundred uses. Every data connection you add gives the agent more context to produce better analysis.

After three months, a PM running this system has an operating advantage that a PM starting from scratch can’t close in a week. After six months, the gap is measured in shipped features and strategic decisions that the manual PM simply didn’t have time to make.

You describe the problem. The agent pulls the data, synthesizes the feedback, drafts the spec, and files the ticket. That’s vibe PMing. And it’s already here.