Granola deliberately minimizes interface elements and hides live transcription because the product is designed for chaotic, rushed moments between back-to-back meetings. Rather than overwhelming users with continuous text updates, Sam Stephenson argues meeting software should actively lower a user's blood pressure and keep their attention focused on the person they are speaking with.
Sam Stephenson uses the metaphor of a handbrake to frame Granola's product role. Instead of positioning an AI application as the centerpiece of an interaction, software should remain unobtrusively on the periphery. The user remains the primary actor making decisions, while the AI quietly handles context and support tasks.
Although Granola initially resisted bypassing its own interface, enterprise customer demands made API and Model Context Protocol (MCP) access inevitable. Large companies possess extensive internal toolsets and workflows; refusing to feed contextual meeting data into existing internal tools and custom AI agents makes an external productivity product a non-starter.
To prevent AI-generated code slop without forcing product designers to review every feature, Granola standardized its design system into components, tokens, and markdown instructions. Providing AI coding agents with markdown documentation on design rules enables any team member to prompt changes that automatically align with the application's design system.
Granola found that static canvas tools like Figma are ineffective for designing context-heavy AI features like pre-meeting briefs. Because evaluating an AI output relies on seeing real, personal data under emotional conditions like meeting stress, designers must rapidly prototype in live code to determine if an experience feels right.
Instead of keeping prototypes hidden until polished, Granola deploys rough, end-to-end working features internally within days of starting. Making team members confront an ugly, flawed experience in their daily routines creates strong intrinsic motivation to refine output quality and interaction details before public release.
Granola's primary internal assistant, Nacho, was initiated as an informal evening project without a dedicated owner. Its widespread company adoption stemmed from integrating broad context across Slack, Amplitude, server logs, and codebases, enabling anyone to debug issues and spin up automated pull requests directly from chat.
Because AI has dramatically lowered the barrier to writing software, simply releasing a functional version one is no longer a moat. Stephenson suggests value now lies in discerning what is genuinely worth building and persevering past shipping until users experience daily utility from thoughtful detail execution.