Key Takeaways
- Meta incubated its Muse agent by eliminating formal slide decks for executive reviews, relying instead on live testing on physical devices.
- Mark Zuckerberg pushed the team away from basic task execution toward long-term alignment, which produced the Goals tab and background reasoning loops called dreaming and study.
- Nat Friedman worked directly with the core team to architect critical security systems, including the agent sentinel.
- Alex Wang fast-tracked the underlying Muse Spark model and made the call to launch when the team wanted more time.
Executive Reviews Without Decks
Big tech product reviews usually require weeks of slide preparation, endless pre-meetings, and defensive positioning. When Mona Sarantakos and the Superintelligence Labs team built Muse at Meta, they threw that process away.
Instead of walking leadership through slides, Sarantakos handed Mark Zuckerberg a phone. “We would give Mark our phone and be like, what do you think? And he'd play around with it and give us feedback,” (play from 0:05) Sarantakos said. If a feature felt slow, awkward, or confusing in Zuckerberg's hands, the team knew immediately. They did not debate theoretical UX trade-offs across thirty slides. They watched the chief executive tap through the interface, saw where he stalled, and fixed the code.
This operating rhythm removed the layer of political theater that slows large teams down. It forced engineers and product managers to keep the mobile build stable enough to put in front of the company founder on any given morning.
Aligning to Goals Instead of Tasks
Most agent builders build glorified chatbots that check boxes. You ask them to book a flight, summarize a PDF, or run a search. Zuckerberg pushed Muse in a different direction.
Years before Muse launched, Zuckerberg wrote about personal superintelligence. When the project began, he refused to let Muse settle into a passive helper that waits for commands. He wanted an agent aligned with a person's life goals over months and years. That vision forced the team to build the Goals tab, alongside background reasoning loops they called dreaming and study. Instead of shutting down between user prompts, Muse runs background routines to organize information, track active projects, and plan future check-ins.
Nat Friedman jumped into the trenches to help shape that architecture. “Nat was obviously like I mean he was basically part of the team, right? I mean he was the team, right?” (play from 18:46) Sarantakos explained. Friedman worked on technical guardrails, including the agent sentinel, a system designed to keep autonomous background execution secure without crippling its ability to act.
Alex Wang and the Push to Launch
Technical breakthroughs do not matter if teams sit on them forever. The team got an early boost when Alex Wang helped fast-track access to early research models. The results caught everyone off guard.
“And what was cool about that story was that the first Muse Spark model that we got was way more agentic than we thought it was going to be. So it was actually kind of a surprise,” (play from 19:51) Sarantakos said. The model showed unexpected capability to link actions together and anticipate needs without explicit prompting.
Yet even with strong models, product teams create reasons to delay. They want cleaner code, another testing pass, or more polished visual assets. Wang intervened to end the indecision. Sarantakos recalled his verdict: “And like he loves Muse. Like, in the end, he was kind of the one that, like, pushed us out the door. He was like, guys, I think it's ready. Like, just go.” (play from 20:33)
Inside an organization with billions of users, shipping requires an external force willing to absorb the discomfort of an imperfect first version. Wang supplied that pressure.
What to Do With This
Cancel your next deck-based product review. Replace it with a rule: attendees only evaluate the live build on a real phone or in a live browser tab. If the build breaks during the meeting, spend the remaining thirty minutes watching the engineer fix the bug instead of talking through a roadmap.