Lenny's Podcast · Nov 15, 2026 · 1h 27m · 77m saved
With Mona Sarantakos, VP of Product at Meta Superintelligence Labs
Meta's Muse went from a weekend experiment on a Raspberry Pi to a launched personal agent in about seven months. Mona Sarantakos, VP of Product at Meta Superintelligence Labs and head of product for Muse, told Lenny Rachitsky on Lenny's Podcast that most of that time went to safety and reliability, and that the biggest surprise came after launch planning was done: shoppers.
Other personal agents reached the market while Muse was still in testing. Mona's team held it back until it cleared an evaluation bar the team wrote for itself, because no industry benchmark covered an agent with access to a person's email, calendar and credit card.
"If we launch too early and we have a bunch of issues, people will not trust this agent. And if you don't trust the agent, you're never going to use it." — Mona Sarantakos
Mona was one of the few people Nat Friedman messaged about the idea in late January, and she worked directly with Friedman, Alexandr Wang and Mark Zuckerberg on what Muse became.
The full interview is covered here so you can skip it. 87 minutes of audio, 10 minutes of reading.
Here are the 13 lessons that matter.
Key Takeaways
Muse spent most of its 7 months on reliability, safety and security, against a bar the team set itself
Each new capability, such as a browser, opened a new attack surface
Shopping was not in the plan: the team expected productivity, health and personal finance to lead
The first Muse Spark model was far more agentic than expected, which reshaped what the product could do
Zuckerberg pushed the team from task completion toward alignment with what the user is trying to do and why
Jolly, the avatar, survived repeated calls to cut it because a personal agent should not look like a corporate logo
Two employees' agents in one group chat leaked a doctor's visit, the team's first lesson in agent discretion
Word of mouth beat advertising at getting new users past their first prompt
1. A Raspberry Pi Weekend
Muse started with a message, not a pitch. Around the end of January, Nat Friedman wrote to a few colleagues that OpenClaw, an AI agent he had been using, was changing his view of personal superintelligence.
Nat Friedman's message to a few colleagues started the project
"Nat Friedman pinged a few of us... he said to us, I'm using OpenClaw. It's changing my whole view of what's possible in terms of personal superintelligence." — Mona Sarantakos
Mona could not find a Mac Mini in stores that weekend, so her husband dug up a spare Raspberry Pi. She spent the weekend talking to the agent running on it and gave it her email, her calendar and her home automation.
"I spent all weekend essentially talking to it... I gave it access to my email, I gave it access to my calendar. And he was like, what are you doing? That is crazy. That is so not safe." — Mona Sarantakos
What convinced her was a Monday-morning newsletter the agent built from her children's scattered school communications, something she had never kept on top of herself. In hindsight, she noted, "a lot of people can do that." There was still no formal team. A few engineers peeled off from the main Meta AI product to prototype, and the work stayed informal until the group decided "this is a thing" and staffed it properly.



