Unlike many competitors that require a visible bot to join a call, Granola records audio directly from a user’s computer system, which allows it to function across various platforms such as Zoom, Google Meet, and Slack.
This privacy-centric approach addresses growing friction between employee autonomy and the expanding role of AI infrastructure in the workplace.
By refusing to sell surveillance capabilities to employers, Granola is attempting to navigate the ethical risks of "context exhaust"—the data naturally generated during speech—which Pedregal believes will eventually be transcribed by default in most professional settings.
He framed the current state of AI as an early technical revolution, suggesting that while transcription is currently the primary use case, it represents only the basic "word processor" stage before more complex digital interfaces emerge.
To balance long-term utility with privacy, Granola is researching technical mechanisms to automatically delete verbatim transcripts after a set period while compressing the captured knowledge into a "working memory" for AI agents.
The company is also prototyping "enhanced transcripts" designed to help AI models better understand specific project names or internal company jargon.
To address consent concerns regarding its invisible recording method, the app uses a virtual camera—a software tool that mimics a hardware webcam—to overlay an animated watermark on the user's video feed, notifying other participants that a transcription tool is active.