Granola: Building a "Life-Changing" AI Workspace

AI is poised to fundamentally alter how humans work and think. Granola, an AI-powered workspace, aims to be the tool that supports this transformation, emerging as a darling in Silicon Valley for its ability to become an indispensable daily tool for many in the tech industry. This article delves into the philosophy, design choices, and growth strategies behind Granola, offering insights into building a beloved product in the age of AI.

The "Second Brain" Vision

The core idea behind Granola is to act as a "second brain," an AI-powered workspace that provides contextually aware support. This vision stems from the understanding that as AI capabilities advance, humans will increasingly outsource lower-level tasks, freeing up cognitive capacity for higher-level thinking.

This concept, however, raises questions about outsourcing memory and its impact on human cognition. The analogy of Google Maps is often used: while it has atrophied our innate navigation skills, it has dramatically reduced the time spent lost and increased efficiency. The key, therefore, lies in thoughtful consideration of what to outsource and how.

Granola's philosophy aligns with the concept of augmentation, inspired by pioneers like Douglas Engelbart, who envisioned computers as tools to enhance human intelligence and collective problem-solving. This contrasts with dystopian futures where technology leads to human atrophy, as depicted in films like WALL-E. The future, it is argued, depends on the tools we build and the rules we establish.

Late to a Crowded Market: A Strategic Approach

Granola entered a market with existing AI notetakers and large players like Zoom offering similar features. However, the company's genesis was not to build another meeting recorder. Instead, the founders aimed to create a "tool for thought."

The realization that AI is only as effective as the context it has access to led Granola to focus on key areas of user context: email and meetings. Recognizing the difficulty of changing email clients, they opted for meetings as their entry point. While this meant entering a saturated space, Granola differentiated itself by focusing on the individual user experience, aiming to be a personal tool optimized for productivity, rather than a generic meeting repository. This deliberate focus on user experience, rather than growth hacks, contributed to its surprising breakout success.

Two Product Founders, Zero ML PhDs

Interestingly, Granola was founded by individuals with strong product and design backgrounds, rather than deep AI research expertise. This approach challenges the notion that building cutting-edge AI companies requires an immediate team of ML PhDs.

The founders' philosophy is that for companies building "wrappers" around existing LLMs, the initial focus should be on product-market fit and user experience. Technical expertise in AI can be brought in later, once the core product is validated and the limitations of off-the-shelf models are reached. This allows for rapid prototyping and learning without the immediate need for highly specialized and often scarce talent.

London vs. SF: Building Outside the Valley

Granola's decision to build its company in London, rather than the traditional Silicon Valley hub, was driven by personal reasons but has proven to be a strategic advantage. While acknowledging the talent concentration in Silicon Valley, London offers a rich ecosystem of engineering talent, strong AI research institutions, and a growing community of product and design professionals.

The company consciously adopted an "American company in London" persona, aiming to resonate with the Silicon Valley market. This involved meticulous attention to detail, such as using American spelling in all communications. Leveraging prior experience building a US-based company and maintaining strong relationships with American investors provided a crucial foundation.

Being outside the immediate "eye of the storm" of Silicon Valley's intense AI hype cycle also offers a degree of insulation, allowing for a more focused and less emotionally draining development process.

One Year in Stealth: Learning Before Launch

Granola spent approximately a year in stealth mode before its public launch. This period was dedicated to intensive user onboarding and iteration. The philosophy was to learn as much as possible from a controlled group of users before exposing the product to a wider audience.

This approach allowed the team to identify and fix critical issues without the pressure of public scrutiny. The decision to launch a more polished product was a deliberate strategy to stand out in a crowded market, aiming to "wow" users upon first interaction. While MVPs are often encouraged, Granola prioritized a high-quality initial experience, believing it was essential for capturing attention in a noisy landscape.

"Building For Us" & Finding First Users

Initially, Granola was built for the founders themselves, targeting knowledge workers who frequently used computers and participated in Zoom calls. This internal focus naturally led to friends and family becoming early adopters.

As the product evolved, the team identified a need to focus on a specific user type to refine the product. They strategically chose VCs and then founders, recognizing their high meeting volume and need for structured notes. This approach allowed them to build a robust product for demanding users, which could then be adapted for broader audiences.

Key Design Choices: No Meeting Bot, No Stored Audio

Granola's design is characterized by deliberate choices that prioritize user experience and privacy. A key differentiator is its "hidden" presence in meetings, avoiding the often awkward and intrusive "bot" experience common in other AI notetakers. This decision, while potentially limiting viral growth, was rooted in the principle of creating a tool that feels personal and non-invasive.

The product also deliberately avoids storing audio recordings. This decision shifts Granola's perception from a "meeting recorder" to an "enhanced notepad," emphasizing the capture of actionable notes rather than comprehensive recordings. This focus on essential information and user privacy is central to its design philosophy.

Simplicity is Hard: Cutting 50% of Features

Achieving simplicity in product design was a monumental effort for Granola. During its stealth period, the team accumulated numerous features and views in response to user feedback. However, before launching, they undertook a rigorous process of cutting approximately 50% of these features, essentially redesigning the product from the ground up.

This was a difficult but necessary step to maintain the product's core simplicity and magical feel. The founders recognized that user requests often focus on adding functionality, rarely on removing it. The responsibility for championing simplicity, therefore, fell to product and design leadership, a often solitary role.

Intuition vs. Data in Making Product Decisions

Granola's product development philosophy blends intuition with data. While the core vision guides decisions, the team actively seeks to fill their "context" with user feedback, both qualitative and quantitative. Regular user calls, aiming for four to six per week for the founders, are crucial for maintaining a connection to the user and avoiding abstraction.

This constant dialogue helps prevent the temptation to over-engineer features based on assumptions. The team understands that users are busy and may not even notice subtle changes, reinforcing the importance of a clean and intuitive interface.

Prioritizing the Future: Build for Tomorrow's Workflows

Granola is being built with a long-term vision, anticipating the future of work rather than optimizing solely for today's needs. The product is positioned not just as a meeting note-taker but as a "tool for thought" that will evolve to handle a wide range of work tasks.

The current product serves as a "Trojan horse" to collect user context, which will be leveraged for future capabilities. This involves investing in deep research that may not be immediately requested by enterprise customers but is seen as crucial for future value creation.

Tech Stack Tour: Model Routing & Evals

Granola utilizes a multi-model strategy, employing the best third-party models available on the market and fine-tuning them when necessary. The focus is on product experience, leveraging the rapid advancements in base models.

The company has developed sophisticated model routing, where different models are selected for specific tasks within the app, ensuring consistency and quality. Prompts are meticulously crafted and updated to maintain a cohesive Granola "feel" across various AI models.

The system takes in numerous signals about the user and their meetings to generate tailored notes. For instance, a VC and a founder in the same pitch meeting will receive distinct notes reflecting their different priorities and perspectives.

Context Windows, Costs & Inference Economics

While context window limitations have significantly improved, the challenge now lies in managing large corpuses of information across numerous meetings. Granola prioritizes putting extensive context into model windows, even at a higher cost, to enable intelligent queries that go beyond simple keyword searches. The philosophy is to build for the future, anticipating that costs will decrease as capabilities mature.

The most expensive component of Granola's business is currently transcription, though its cost has fallen dramatically. While inference costs are expected to rise with more complex queries, the company aims to balance this with the increasing value users derive from advanced features.

Audio Stack: Transcription, Noise Cancellation & Diarization Limits

Granola handles echo cancellation internally, building on open-source frameworks. For transcription, they partner with leading providers like Deepgram and AssemblyAI, always utilizing the latest models.

Real-time diarization (identifying speakers) remains a challenge, with current solutions still in their infancy. The team is closely monitoring this area, recognizing the potential for incorrect diarization to confuse AI models.

Guardrails & Citations: Building Trust in AI

To mitigate the risks of AI errors and hallucinations, Granola emphasizes transparency and user control. While striving to minimize mistakes, the product design allows users to "view source," examine citations, and refer to original transcripts and quotes. This "human in the loop" approach builds trust by enabling users to verify information.

Growth Loops Without Virality Hacks

Granola's growth has been largely organic, driven by word-of-mouth and a focus on product quality. The absence of a visible bot in meetings, while a trade-off for direct virality, has led to a unique growth loop: users are prompted to discuss Granola when encountering other AI bots.

The product also facilitates sharing notes via links, allowing recipients to interact with the transcript and ask questions, effectively unlocking AI capabilities for them. The future roadmap includes building a "second brain" for teams and companies, which brings its own set of challenges regarding data sharing and privacy.

Enterprise Compliance, Data Footprint & Liability Risk

As Granola expands into the enterprise, questions around recording conversations, data privacy, and legal compliance become paramount. The company is committed to transparency, with features allowing users to announce their use of Granola.

The broader challenge for AI in the enterprise is balancing the desire for AI-driven efficiency with the need to manage liability footprints. The tension between retaining historical data for AI insights and limiting data retention for liability reasons is a significant area of exploration.

Retention & Habit Formation: The "500 Millisecond Window"

Building user habits is a core challenge in product development. Granola's success in retention is attributed to its usefulness combined with timely triggers. Meetings, being calendar-based events, provide a natural trigger point.

The "500 millisecond window" refers to the critical moment during a meeting when a user realizes they need to take a note. Granola aims to be the tool that is readily available and effective in that instant, preventing users from forgetting or reverting to less efficient methods.

Competing with OpenAI and Legacy Suites

The question of why giants like OpenAI or Google wouldn't replicate Granola's functionality is a common one. Granola's strategy is to focus on doing "something way better for a specific use case and a specific type of user." While acknowledging the broad ambitions of AI-native companies, Granola believes that deep tailoring and a focus on user experience will win.

The company envisions a future where AI tools are not just note-takers but integral to a new medium of work, offering capabilities far beyond current offerings.

The Future: Deep Research Across Meetings & Roadmap

Granola's roadmap is focused on transforming its platform into a comprehensive "tool for thought." This includes enabling deep research across a user's entire meeting history, allowing for complex queries like identifying potential investors for a Series C round.

The company is also exploring dynamically generating and sharing artifacts, such as up-to-date memos on customer feedback, accessible via URLs. The ultimate vision is to leverage extensive context to unlock new workflows and use cases that users are not yet imagining.

Granola as Career Coach?

A fascinating potential future for Granola is its evolution into a personal "coach." By analyzing a user's entire history of conversations and work, Granola could offer insights into strengths, weaknesses, and areas for improvement, akin to a human coach or therapist. This would leverage the vast context the platform collects to provide personalized guidance and enhance professional development.