CHAPTERS
- 0:00 – 2:00
Muse’s positioning: a personal agent designed for parents (and “the olds”)
Claire introduces Muse, Meta’s new personal AI agent, and explains why its marketing and feature set are clearly aimed at millennial/Gen X parents. She previews what she tested, what surprised her, and why Muse stands out as one of the best-designed agents she’s tried recently.
- •Muse is framed as a personal agent that “gets things done,” not a work assistant
- •Target audience is heavily parent-centric (permission slips, strollers, kid logistics)
- •Claire’s thesis: design and UX are Muse’s differentiator
- •Plan for the episode: onboarding + real tasks + UX/permissions deep dive
- 2:00 – 4:32
What’s familiar vs. different: secure VM, Meta ecosystem access, and consumer-first “apps”
She compares Muse to other agent products: it runs on a secure dedicated computer/VM, supports desktop/mobile, and integrates tightly with Meta channels like WhatsApp. Muse emphasizes safety, credential storage, and consumer-friendly language (apps vs. plugins/connectors).
- •Muse uses a persistent isolated Linux VM with a browser (agent “has a computer”)
- •Multi-surface access: desktop, mobile, WhatsApp messaging
- •Safety/commerce stack: permissions, credential storage, future 1Password, Stripe Link
- •Consumer framing: “apps” like Facebook/Instagram/Gmail/Shopify rather than developer jargon
- 4:32 – 7:05
Sign-in and first-run trust building: identity, naming, and editable ‘Soul MD’
After quick sign-in via her already-logged-in Facebook account, Claire highlights Muse’s trust-first onboarding and its playful identity setup. She names her agent Polly and notes the OpenClaw-like structure: editable identity, “soul,” and memory.
- •One-click sign-in with existing Meta session reduces friction
- •Onboarding emphasizes control, safety, and what Muse will do
- •Agent personalization: naming the assistant (Polly) and avatar identity
- •Editable identity/soul/memory feels inspired by OpenClaw-style agent design
- 7:05 – 8:05
Activity feed as a core primitive: task history, approvals, and ‘heartbeat’
Claire calls out Muse’s activity feed as an unusually strong UX element for transparency and navigation. It shows what tasks were done, approvals granted, and a sense of ongoing agent state, making it easier to understand and revisit past work.
- •Activity feed provides persistent task lineage over time
- •Explicit list of approvals reinforces user control and auditability
- •“Heartbeat” concept suggests ongoing background work/status
- •Design takeaway: easy-to-read transparency without overwhelming users
- 8:05 – 9:06
First real task: connect Google Calendar and delete soccer practice
She tests a practical parenting workflow: managing her kids’ schedules. Muse requests Google Calendar access, asks for needed details, and performs the deletion with clear permission prompts—creating immediate trust.
- •Guided setup: connect Google Calendar with low anxiety/friction
- •Muse asks for kid info, then confirms the intended change
- •Deletion of soccer practice is executed cleanly and quickly
- •This “first win” establishes confidence in Muse’s agent workflow
- 9:06 – 14:11
From agenda to ‘morning newsletter’: requesting a printable PDF for the family
Claire pushes Muse beyond a simple agenda list by requesting a one-page, newsletter-style PDF: schedule, kid callouts, email-derived context, weather, and local news. Muse connects email, infers relevant household context, and produces a well-designed artifact that beats her prior attempts with other tools.
- •Initial agenda output is basic; Claire asks for a designed PDF artifact
- •Muse connects email and summarizes inferred family context for confirmation
- •Output: beautifully laid out weekly agenda + kid-specific highlights
- •Includes local/kid-appropriate news + discussion prompts for the table
- •Key value: agents that bring value offline (printable, family-facing)
- 14:11 – 15:42
Personalized Feed: a healthier, skimmable alternative to social feeds
Muse includes a personalized feed that Claire configures with a single directive prompt focused on clarity and avoiding clickbait. Early items feel strongly targeted to her life (sleep training, hydration/blood pressure), and she notes it’s far preferable to her default Facebook feed experience.
- •Feed setup is prompt-driven and simple to configure
- •Emphasis on skimmability and non-clickbait tone
- •Recommendations appear personalized (sleep quality, salt in water)
- •Open question: whether it will incorporate signals from Instagram/Facebook
- 15:42 – 17:14
Ideas: an out-of-the-box prompt library for common life tasks
Instead of requiring users to invent use cases from scratch, Muse offers an ‘Ideas’ section with categorized, ready-to-run tasks. Claire highlights civic, productivity, relationships, wellness, and finance/shopping ideas, praising the presentation as consumer-friendly and delightful.
- •/ideas acts as a starter kit of suggested automations and workflows
- •Examples: reporting potholes/streetlights, managing returns/inbox
- •Relationship ideas: trivia night, planning dinner with friends
- •Claire predicts wellness-focused Muses could become a killer use case
- •Strong UI design reduces “blank page” prompting burden
- 17:14 – 21:16
Goals as a personal tracking primitive: water, shopping, and sleep training
Claire argues Muse’s ‘Goals’ is a major UX unlock: it frames ongoing life changes as goals rather than dev-y tasks or chat threads. She demos setting a gentle sleep-training goal, praising Muse’s supportive tone, structured questioning, plan creation, and reminder system.
- •Goals categories (health, relationships, finance, productivity, etc.) feel consumer-native
- •Sleep training setup: Muse asks clarifying questions and proposes a plan
- •Tone is empathetic without feeling overly sycophantic
- •Goals persist beyond chat, with scheduled reminders to nudge progress
- 21:16 – 23:54
Library: managing generated artifacts, including surprising AI podcasts
Muse’s Library aggregates outputs—documents, websites, images, videos, and even podcasts. Claire tests podcast generation (NotebookLM-style) for weekend AI news and is impressed by the quality and the option to publish/share, calling it a ‘moment of delight.’
- •Library is the repository for everything Muse creates
- •Supports multiple artifact types beyond PDFs (sites, media)
- •Podcast generation is a standout unexpected capability
- •Generated show includes hosts/format and feels publishable
- 23:54 – 25:25
Design deep dive: step-by-step tool-call lineage and progressive disclosure
Claire returns to the activity feed to highlight a power-user layer: detailed tool calls/scripts per task, presented in a friendly UI. She frames this as the right kind of transparency—useful for developers, optional for consumers—and pairs it with praise for Muse’s progressive disclosure patterns.
- •Activity feed expands into granular execution steps (tool calls, scripts)
- •Transparency is available without forcing it on casual users
- •Claire wants similar lineage UX in Codex, Claude Code, and other agents
- •Overall design thinking: reveal complexity only when users opt in
- 25:25 – 25:56
Permissions done right: trust through staged consent and confirmation
Muse’s permission system earns Claire’s trust by asking at the right moments and reflecting back what it learned before using it. She contrasts this with agents that either assume broad access or overwhelm users with developer toggles like ‘YOLO mode.’
- •Muse avoids confusing modes like auto-approve/ask-every-time for consumers
- •Staged consent: request access → ask to retrieve info → confirm findings → ask to use
- •Reduces anxiety while staying efficient and not “annoying”
- •Permissions UX is positioned as a key differentiator for mainstream adoption
- 25:56 – 29:28
Avatar UX as a ‘delight engine’: animations, working states, and generative makeovers
Claire calls the animated avatar her favorite UI element: it visibly “works” with a laptop, shows a different state for media generation, and makes task progress feel friendly. She demonstrates generating a new teal dragon avatar named Slime, emphasizing how generative UI enables richer loading/interaction patterns.
- •Animated avatar communicates activity (laptop animation while working)
- •Media generation uses a distinct visual state (orb) for feedback
- •Avatar editing is integrated: prompt → generate options → pick with approval UI
- •Claire’s broader point: great design is new interactive affordances, not just rounded buttons
- 29:28 – 33:32
Browser-use tests: shoe shopping struggles vs. IMAX tickets success (and Stripe Link checkout)
Muse underperforms on shopping for New Balance 9060s (wrong details, gets stuck), revealing limitations in web navigation and trust signals. A second test—buying IMAX tickets—goes much better, and she highlights Muse’s approach of showing just the browser (not the full VM) plus the Stripe Link flow for payment.
- •Shopping test failure: difficulty finding correct product/colorway and details
- •Muse’s UI exposes only the browser (more consumer-friendly than full VM)
- •Ticket purchase flow succeeds: finds showtimes and proceeds to checkout
- •Stripe Link integration enables agent-driven payments with user control
- •Trust dips when sources/results seem mismatched during browsing
- 33:32 – 37:11
TL;DR: what Claire will use Muse for—and why designers should study it
Claire concludes that Muse nails consumer agent UX: onboarding, permissions, transparency, and delightful primitives (feed/ideas/goals/library). She plans to use it for wellness tracking, kid logistics, morning briefs, shopping, and home/civic errands—while noting browser-use still needs improvement and Meta privacy is a ‘devil you know’ tradeoff.
- •Muse’s differentiator is UX polish and consumer-friendly primitives
- •Planned ongoing uses: wellness, kids’ schedules, morning newspaper, shopping, home/city tasks
- •Design lessons for PMs/designers: what to expose vs. hide; transparency with control
- •Weakness: browser-use reliability, especially shopping
- •Curiosity about Meta’s underlying models due to strong artifact design
