Chris Pedregal - Building Granola - [Invest Like the Best, EP.412]

Episode Summary

Executive Summary: Patrick O’Shaughnessy interviews Granola founder Chris Pedregal about AI as a new “tool for thought,” why meeting memory is a first killer use case, and how Granola uses transcription + editable notes to augment human judgment. They cover AI-native workflows, privacy norms, model competition, product philosophy, and why small teams can build outsized software in the LLM era.

Main Topics: Tools for thought across history (Priority: 5/5): Pedregal frames writing, notation, and visualization as predecessors to AI-powered cognition. Granola’s product vision (Priority: 5/5): Granola transcribes meetings, preserves context, and helps users synthesize notes and follow-ups. Human augmentation vs replacement (Priority: 5/5): He argues AI should remove busywork while leaving judgment, creativity, and decision-making to people. Privacy, recording, and social norms (Priority: 4/5): Granola avoids storing audio to reduce invasiveness and sees workplace norms diverging from social settings. Small teams and AI app-building (Priority: 4/5): Foundation models let tiny teams ship powerful products by focusing on UX, edge cases, and speed. Competition, moats, and model providers (Priority: 4/5): He expects model providers to improve underneath apps while defensible products win through use-case focus. Personal motivation and product philosophy (Priority: 3/5): Pedregal emphasizes emotional clarity, user control, and building things that realize human potential.

Key Arguments: AI should extend human capability, not replace judgment; Granola is built to make users better. Meeting notes work best when AI captures rote details and people only add critical thoughts. A tool’s value rises when it brings the right context at the exact moment it’s needed. Workplace recording will likely become normal, but privacy and invasiveness must be balanced. High-frequency, high-stakes workflows need specialized apps; general assistants can’t always match the UX. Granola’s moat comes from context, habit, and product design—not just model intelligence. Small teams can move fast because LLMs provide a powerful technical layer, but edge cases still matter. AI-native users already work differently: they write fewer notes and query the transcript instead. The biggest unanswered problem is the interface for collaborating fluidly with AI, beyond chat. Personalization is still surprisingly weak in frontier models and remains a major gap.

Data Points: expense reviews automated: 85% - Ramp ad read claims AI automates most expense reviews expense review accuracy: 99% - Ramp ad read claims its AI review automation is highly accurate company savings: 5% - Ramp ad read claims it saves companies five percent current product penetration: 5% - Pedregal says Granola is only about 5% along its vision follow-up completion target: 80, 90, 95% - He says Granola should get users most of the way through follow-up work meeting note behavior: a couple notes per meeting - Heavy users write only a few internal-thought notes per meeting launch timing: May - Granola launched in May, roughly eight or nine months prior to the interview Mac OS support at launch: Mac OS 13.4 - Early Granola only worked on that version and newer Mac user coverage at launch: 15% - That OS version represented about 15% of Mac users then late-stage product time: a year - They built Granola for a year before launching

Pivotal Quotes: "humans are toolmakers" — Chris Pedregal: Explaining the long history of tools that extend cognition "we see the role of AI as being a tool to make you better" — Chris Pedregal: Describing Granola’s core product philosophy "the competitor that we have chosen to worry about at Granola is the one that hasn't launched yet" — Chris Pedregal: Talking about competitive dynamics in fast-moving AI markets

Implications: The open question is which AI workflows will become daily habits; listeners should watch for products that combine strong context, control, and privacy.