Week in voice AI#12: Granola scramble
Would you like yours crunchy, local, or open source?
I am Ivan Mehta, a consumer tech reporter at TechCrunch. For more than a year, I have covered different aspects of voice AI. This newsletter is an experimental attempt at covering what is happening in this industry, which is growing at a rapid pace.
Top News
Granola’s database change and many flavors of meeting notetakers
Granola has been one of the most popular meeting notetakers around, especially in the tech community. The reason behind this popularity is that the app runs on your desktop, uses system audio to transcribe meetings, and doesn’t need a bot to join your meeting. This means it can run in the background, transcribe meetings, and then generate notes without any intervention.
Last week, suddenly, there was a lot of discussion around a technical change in Granola’s local database storage, breaking workflows for people who might have created AI agents to access local transcripts.
Granola’s founder, Chris Pedregal, first pointed out to users that they can use the app’s Model Context Protocol (MCP), to connect to other tools. However, some users pointed out that for mass extraction and processing, the API is a better method.
A day later, Pedregal clarified that the company is not trying to lock down Granola. He said that while people had worked out how to access the local cache of the app for the transcript, it wasn’t designed to be used in that way. To mitigate that, the company built an MCP:
“A couple of months ago, we noticed that some folks had reverse-engineered our local cache so they could access their meeting data.
Our cache was not built for this (it can change at any point), so we launched our MCP to serve this need. The MCP gives full access to your notes and transcripts (all the time for paid users, time-restricted for free users). MCP usage has exploded since launch, so we felt good about it.
A week ago, we updated how we store data in our cache and broke the workarounds. This is on us. Stupidly, we thought we had solved these use cases well enough with our MCP.”
Pedregal also mentioned that the company is going to launch a public API to let users pull data. He added that Granola is going to figure out how users can build workflows for locally-run agents.
Alternatives galore
Thanks to AI-powered coding tools, it is easy to spin up an app. While the saga around Granola’s local data access was going on, a few people spawned meeting notetakers that worked well locally. A lot of these tools were open-sourced from the get-go, to let developers build a custom version if they wanted.
Some of them even added basic features for AI summarization or dictation.
VC point of view
Granola’s new workflow issues came to light more when a16z investor Guido Appenzeller posted on X about it.
Lightspeed partner and Granola investor Micahel Mignano sided with the company and also said it was “a shameful move by a VC trying to slam a startup based on a lazy assumption.”
Granola is not going to lose too many customers over this, but this goes to show that there are a lot of moving parts in personal or enterprise AI workflows, and even small changes from companies can impact them deeply. We might see products get bashed over these kind of issues more often in the future.
Quick Bytes
Sony said that it has requested different streaming services to remove over 135,000 tracks that impersonate the label’s artists, such as Beyoncé, Queen, and Harry Styles.
A man from the U.S. pleaded guilty to pumping streaming services with thousands of AI-generated songs, and using bots to artificially boost streams to get more royalties. He earned more than $10 million in royalties by clocking over 600,000 daily streams through bots.
Instagram releases eight AI-powered filters for voice notes.
Amazon is planning a comeback in smartphones with Alexa as a core part of the product roadmap.
The UK backtracked on the rules around AI being able to train on copyrighted material with an opt-out option after artists like Sir Elton John and Dua Lipa pushed back.
Signals and Experiments
I think about noise a lot because I live in India, and my house is on a busy street. There are only rare days when everything is quiet around me. I can always hear animals, people, vehicles, and machines humming around me. I have always been fascinated to learn how loud are my surroundings?
Recently, I started testing a desk device from Deep Care. The offline device’s main idea is to inform you about better desk posture and keep you active with movement prompts. However, one of my favorite features is that the device also shows me the decibel levels around me.
The strange thing is that the mic often doesn’t pick up sounds that are coming from a distance, such as a vehicle honking. But picks up the levels of someone speaking around you. I compared this with the iPhone’s live listen feature with AirPods, and I did get a higher decibel level than the Deep Care device.
Thank you for tuning in. Keep listening.
Email: voiceaiweek@gmail.com or im@ivanmehta.com