Key Takeaways

  • Meta's Muse team found that word of mouth is the single most effective driver for activating personal AI agents, beating traditional product marketing.
  • New users freeze when facing an open input box because an infinite action space causes blank-slate paralysis.
  • Sarantakos added an ideas tab filled with pre-packaged starter packs for tasks like setting cron notifications and building dashboards.
  • Onboarding flows built for early adopters fail with later cohorts, who require structured rails rather than open-ended discovery.

Users Freeze When the Action Space Is Infinite

When you hand someone a blank text box backed by a frontier model, you think you are giving them limitless agency. In practice, you are handing them an exam question with no prompt. Lenny Rachitsky pointed out the core tension: “It feels like in AI assistance, this is maybe the biggest lever for people to understand what this is capable of. It there's like so much power that nobody even understands what to do with it.” (play from 46:51)

Mona Sarantakos hit this exact friction while leading product for Muse at Meta Superintelligence Labs. The team watched new signups drop off immediately at the main input bar. Sarantakos saw that users froze before typing a single character. “I think the second is like getting people to get their first prompt out,” (play from 48:51) she explained. “Like people freeze on the first prompt.” (play from 48:55)

The issue is cognitive overload. Builders know the system can run scripts or structure unstructured data. The user only sees a blinking cursor. Without an immediate reference point, they bounce. Sarantakos found that peer demonstrations beat internal marketing: “So the most effective lever out there is word of mouth. It's like basically hearing other people's stories. Like that is number one.” (play from 47:26) Seeing a friend extract value gives users a specific mental model to copy.

The Ideas Tab Replaces Prompt Engineering With Starter Packs

Relying on external word of mouth leaves in-product conversion to luck. To break the freeze directly inside the software, Sarantakos and the Muse team built concrete scaffolding into the user interface: an ideas tab filled with ready-to-use recipes.

Instead of expecting users to formulate prompts from scratch, Muse presented pre-built actions. “And in the ideas tab, it's basically pre-packaged like it's like packs of like starter packs, right? Um to inspire you,” (play from 49:15) Sarantakos explained. “And so it's like Muse can help you, you know, do X. And like in that X, it can create a cron. So it's like a notification. It can build you a dashboard. It can whatever, right?” (play from 49:22)

These starter packs turned open-ended agent tasks into one-click buttons. A user did not have to write a custom prompt to schedule recurring checks. They clicked a card to set up a cron notification or assemble a personal dashboard. Packaging agent power into discrete units of work lowered the hurdle from creative composition to straightforward selection.

Early Adopter Tactics Fail the Next Cohort

The trap in AI software is assuming your first wave of users resembles the next one. Technical early adopters enjoy probing an interface, testing edge cases, and typing long instructions. Mainstream users want immediate results with minimal effort.

Sarantakos saw this shift firsthand as Muse grew. “What we're finding is what worked for the first cohort of users, which was like the early adopters, isn't going to work for later adopters,” (play from 50:11) she observed. When the product expanded beyond enthusiasts, the patience for open text prompts disappeared.

Later cohorts want their job done without experimentation. If your activation flow depends on a user asking clever questions, your retention will drop as your audience widens. The product must evolve from an open playground into a structured system that leads users by the hand.

What to Do With This

Audit your onboarding flow by recording five non-technical users opening your product for the first time. If anyone pauses for more than ten seconds at an empty input field, replace that blank prompt with three clickable starter packs that execute high-value tasks immediately.