Why ‘vibe PMing’ is the future of product management

This article summarizes an episode of Aakash Gupta’s video series featuring Frank Lee, principal product manager at Amplitude.

Image credit: Arsal Ysfin

Frank Lee, principal product manager at Amplitude, argues that software development has moved past manual data analysis. Automating these routine tasks changes the job completely, leading to an era of “vibe PMing.” In this new world, success comes from a manager’s taste and vision, rather than their ability to just churn through paperwork.

However, Lee warns that insight is useless without speed. For automation to matter, it must compress the gap between identifying a problem and fixing it. By integrating AI tools directly into his workflow, he ensures teams can move instantly from diagnosis to execution.

Rethinking the product system

To make this possible, Lee rethinks how information is organized. Most product leaders struggle to make AI find information across the scattered software their companies use. Lee avoids this by organizing roadmaps, contexts, and notes into local files within his code editor, allowing AI tools to access everything simultaneously.

Lee says, “On the left-hand side [of my code editor], I have a bunch of contexts that I’ve already actually aggregated… within Claude Code or within Cursor, I can easily refer to some of those pieces of context, brainstorm about them, and draft a new spec.”

Bridging integration gaps
Beyond organizing context, Lee also addresses workflow friction. He observes that product managers often waste time manually bridging information gaps between tools. To eliminate that inefficiency, he writes custom scripts that pull data from specialized applications into his central workflow.

“Granola [a meeting notes tool] actually does not have a dedicated MCP [command palette] right now. So I tried to hack Claude Code to build some type of automation,” Lee explains. “I basically could run a command to pull in my recent Granola notes using the script we wrote.”

Automating data insights

Similarly, investigating unexpected metrics by hand can consume entire weekends. To avoid this, Lee outlines a five-step process that automates the heavy lifting:

• Chart analysis: Give an agent a link to a chart. It sorts through the data, finds what’s unusual, and guesses why numbers changed.
• Automated reporting: Give an agent access to dashboards. It summarizes the main findings and problems, so you don’t have to check them by hand.
• Summarizing feedback: The agent gathers feedback from all sources (Zendesk, Gong, Slack). It then groups similar comments to find the main problems.
• Creating product documents: Take the results of the analysis and give it to a document template. The agent writes a first draft of the plan in minutes.
• Prototyping and sending tasks: Simple ideas can be turned into early models in the agent’s code editor. Harder tasks are sent to teams through Linear.

“What I would have had to do manually to investigate… the agent did it in a minute and a half,” Lee explains. “I’ve automated basically all of my weekly business reviews.”

Defining the automated workflow

Taken together, this system removes bottlenecks created by human-dependent review processes. Rather than relying on managers to read every chart and dashboard, Lee assigns defined roles to AI:

• Analysis reads charts to find unusual patterns and explains why numbers changed.
• Reporting summarizes weekly progress automatically, so managers do not have to check every dashboard.
• Building turns written ideas into early code examples or sends tasks directly to the developers.

The cost of system complexity
But giving an AI agent access to too many data sources also often creates confusion rather than clarity. Lee warns that overloading the system with irrelevant connections slows down response times and degrades the quality of answers.