08 Feb 2025

Granola App - def a better note taker but not because of it’s AI

Lots of love recently for Granola app (founders of Intercom, Linear, Vercel below)!

I have been a user for ~4 months now and my tl;dr: it’s a really good meeting notes app, but only because of it’s seamless design choices and not due to AI. Let’s dive in!

granola-tweetsgranola-tweets2granola-tweets3

What is Granola and how does it work?

Granola calls itself as “the AI notepad for people in back-to-back meetings”. With a really small 17 member team, they have absolutely nailed virality with an amazing product that a LOT of people LOVE.

The workflow is pretty seamless:

granola-workflow

Product decisions that really ticked for them

  1. Bringing the human front-and-center of the note taking Unlike other AI note takers, Granola understands that the human notes still contain the highest signal. We can take notes as BAU on the minimalist notepad which are then used to finalize the “AI summary”. Doesn’t hurt that it directly connects to the Mac audio so there are no annoying bots “joining” the virtual meet. granola-notes

  2. Reducing any friction in taking notes 1 min before any meeting, a Mac notification pops up that allows you to join a meeting as well as open a Granola note for that meeting. No separate effort needed to start a note. It also “discovers” ad-hoc meetings like Slack huddles/adhoc scheduled google meets and gives a notification to start a Granola note for that call. Pretty smart! granola-notification

  3. One central repository for most notes So refreshing to have all of my individual notes reside in Granola & not go have to remember which slack thread, google doc, apple notes, obsidian vault did I make a note for which particular call. Ofcourse, multi-user notes still end up in a Google Doc auto-linked to the GCal invite (more on this later)

What is still not working for me?

As a PM at a big tech company, Granola’s core use case of AI in note taking still doesn’t work for me personally.

Specifically,

  1. The 2 core AI features are absolutely useless to my persona
    1. AI enhanced notes still add a lot of slop. So much so that I don’t EVER see the AI enhanced summary of the call anymore. Sure it’s good but it is still easily 5x longer than the actual takeaway of the call for me. Having 5-10 meetings a day with a lot of AIs, it’s impossible to keep going through the slop of AI and actually get to the valuable AIs.
    2. Chat with your notes: I didn’t even know this existed till my research for this article, and it’s def not a good fit for my persona. Maybe it is for a VC persona taking a lot of calls with different founders where they are the ones getting info.
  2. The act of note taking is not about actually taking notes. It’s so much more, it’s about having the context to understand who does what, why, when. It’s also about ensuring others know what is expected of them. It’s about juggling multiple balls and ensuring nothing gets dropped. Which is why I like Granola’s philosophy of keeping the human in the center of the loop. But with that same logic, I can’t outsource this core act to the AI and sit back “concentrating” on the call. Outsourcing this core act seems counterproductive to me right now.

What would I change in the product?

Granola’s stated next steps in public blog posts and podcasts has been about allowing more context-sharing (e.g. show me all the calls I had with this person prior to today’s call) and automating post-call workflows (slack, email follow ups etc.)

I have a few thoughts:

  1. Personalize the AI from my behaviour and make it actually useful!

    If I am never looking at the AI generated summaries, there is a problem. And with Granola burning incremental money with each incremental call, it’s important that the output of the AI is actually useful. Instead of having generic templates, add a “just like you” template which can learn from all my notes so far and just recreate notes at that depth. The gold standard here would be me generating 5 AIs but AI actually catching the one I missed, and only add one additional line - instead of adding like a hundred more lines.

  2. I don’t think this is a one-fit-for-all solution. Granola has to define the persona they want to own and go deep

    Granola states that they see a lot of love in senior leaderships in an org and typically flow down to junior levels. This persona typically doesn’t use “call notes” as Action Items but more like “context” of what happened in the call. The correct next step for this persona is to spread this context further - further in the org and further in future calls. E.g. if in a call it was discussed that xyz function is already working on an initiative that will be impacted by the learnings of this call, then a slack thread on a relevant channel or a follow up conversation, initiated on behalf of the senior leader seems like an ideal next step. If Granola can aid connecting the dots across multiple meetings and maintain context throughout, it becomes central to the core act of knowledge management.

    One moonshot idea here was a potential Glean integration where meeting notes actually become part of the company’s internal knowledge base. This allows a potentially infinite context dissipation across the org on a need-to-know basis.

  3. Make it Multi-user!

    Very tactically Google Docs is the king here because it gets auto-attached in the meeting invite. However, Granola should either allow the invite creator to link a web view of the granola meeting note OR drop a personalized msg on my behalf at the start of a Google Meet call. They are already dropping this generic msg so it’s just about personalizing the msg with a live multi-user editable link granola-meet

  4. (already in publicly shared existing plans) Post-call workflow automation - super exciting! E.g. me writing a “paste x on y slack channel”, “send an email with this msg” and Granola being able to quickly generate drafts for them, ready to go would be super helpful. All clicks saved are always appreciated!

Intro into how Granola potentially works under the hood

Evals. Training. Corrections. Potential system design - trying to figure it out with first principles. granola-system-design


References