CHAPTERS
- 0:00 – 3:37
Latency, status, and progressive disclosure in agent UX
Claire and Alex dig into a core consumer-AI design problem: how to make waiting feel acceptable without dumping verbose reasoning into the main experience. Muse uses a layered status system—simple “working” signals up top, with deeper logs available only when the user wants them.
- •Designing for perceived latency: show progress without overwhelming users
- •Progressive disclosure: from “it’s working” → what it’s doing → detailed audit trail
- •Different needs for engineers vs. everyday users
- •Avoiding loud chain-of-thought style narration in the primary chat
- 3:37 – 5:34
What Muse is optimizing for: a conversational exchange with an ‘entity’
Alex explains the foundational framing behind Muse: treat the interface like messaging with a distinct being rather than a traditional AI tool. That framing drives UI choices like message bubbles, a persistent embodied avatar, and clear separation between messages across time.
- •Messaging patterns over ‘AI app’ conventions
- •Embodiment at the top to reinforce an ongoing relationship
- •Designing for proactive agent messages and time-separated interactions
- •Status visibility becomes critical when the agent works in the background
- 5:34 – 9:14
Choosing consumer-friendly primitives: character, ideas, feed (not files/projects)
Claire highlights how Muse elevates different core primitives than enterprise/coding agents. Alex ties this to accessibility for “normal” users and to helping people discover what the agent can do without requiring software-centric mental models.
- •Muse foregrounds character/persona as a primary object
- •Ideas and feed surfaces help capability discovery
- •Progressive disclosure makes advanced details optional
- •Consumer orientation: intuitive over tool-centric abstractions
- 9:14 – 11:01
Speed and responsiveness: acknowledging requests like a human assistant
They discuss why Muse’s fast model responses materially change the experience, and how the agent balances instant acknowledgment with longer background work. Alex emphasizes the human-like pattern of confirming a request, sharing a plan, and then executing.
- •Latency is a key adoption lever for consumer agents
- •Fast responses can create trust questions (“did it really do it?”)
- •Quick acknowledgments + background execution reduce anxiety
- •Status UI helps users verify progress and debug wrong paths
- 11:01 – 14:25
Generative UI done ‘on rails’: personalization within stable surfaces
Claire points out that Muse’s generative UI is subtle and legible: stable primitives with personalized content inside them. Alex explains how avatars and their animated states (typing, making) create delight and ownership without making the whole UI unpredictable.
- •Generative UI as personalization, not total layout randomness
- •Avatar transforms across animated states (e.g., ‘tomato typing’)
- •Ideas are personalized to conversation history
- •Feed is configurable but remains a designed native surface
- 14:25 – 15:33
Demo setup: meeting Slime and Marlo (and why avatars matter)
Claire introduces her Muse “Slime,” while Alex introduces “Marlo,” his cactus avatar, and the origin of Jollibot. They underline how the character layer makes the agent feel personal and approachable.
- •Avatar choice as a bonding and identity mechanism
- •Jollibot’s personal origin story (named after Alex’s daughter)
- •Personalization increases comfort for non-technical users
- •Delight comes from small, consistent animated touches
- 15:33 – 17:05
SkyMall-style shopping catalog from Instagram Saves
Alex demonstrates turning Instagram saves into organized categories and then into a nostalgic SkyMall/Sharper Image-inspired shopping magazine. The point isn’t just utility—it’s rendering personal data into a fun, browsable artifact that refreshes over time.
- •Instagram saves → folders/categories → shopping suggestions
- •Rendering text suggestions into a designed ‘catalog’ artifact
- •Daily refresh makes it feel like a living, personalized magazine
- •Reframing mundane personal data into something enjoyable
- 17:05 – 19:07
Shopping workflows: monitoring, research, and price/discount alerts
They expand from the demo into practical shopping use cases: continuous monitoring, alerts, and research flows. Claire describes weekly checks for discounts on a wishlist and price comparisons, while Alex notes how Muse replaces manual search-heavy research.
- •Monitoring as a ‘superpower’ (albums, concerts, collaborations)
- •Wishlist scanning for weekly discounts and price drops
- •Price comparison and research as default starting point in Muse
- •Replacing repetitive web search + YouTube rabbit holes with an agent flow
- 19:07 – 21:45
Instagram + Pinterest connectors: visual data as a creativity unlock
Claire argues connectors to visual platforms are uniquely valuable because they hold rich personal context that’s historically hard to reason over. Together they describe using saves/boards for summarization, concept generation, and extracting descriptive language.
- •Instagram is a high-signal personal dataset when made queryable
- •Pinterest boards as inspiration sources for concept and language generation
- •Agents logging into websites expands usable personal archives
- •Turning buried saves into structured, actionable knowledge
- 21:45 – 25:28
Going hard on hobbies: golf round recaps and practice breakdowns
Alex shows how Muse uses golf data (round logs, shot tracking, practice machines) to generate richer narratives and visuals than traditional scorecards. He also describes goal tracking toward a single-digit handicap and why long-term goals fit agent memory well.
- •Round recap dashboards: course context, holes, and improvement deltas
- •Visual storytelling from plain stats (front nine vs. back nine trends)
- •Practice session analysis from swing/launch-monitor data
- •Long-term goal tracking (single-digit handicap) tied to ongoing habits
- 25:28 – 28:01
Main chat vs. side chats: organizing conversations in the agentic era
Claire asks how to decide between Muse’s canonical main chat and side chats. Alex explains the architectural intent: an ongoing primary relationship thread with optional side threads for scoped work—contrasting with the ‘everything is a new thread’ pattern.
- •Canonical main chat reinforces a single ongoing ‘entity’ relationship
- •Side chats are for scoping and keeping the main chat clean
- •Contrast with traditional chatbot threading defaults
- •Open design problem: making the distinction clearer for everyday users
- 28:01 – 29:36
Reviving a lost blog into a daily morning podcast via Wayback Machine
Alex shares a personal, emotional use case: retrieving a 20-year-old blog from the Wayback Machine, compiling it, and converting it into a randomized daily audio ‘podcast’ reading. Muse helps reformat forgotten archives into formats he actually consumes.
- •Wayback Machine retrieval across blog platform moves (Blogger/Tumblr/etc.)
- •Compilation into PDF, then into a blog-like artifact
- •Daily audio delivery turns text archives into a new habit
- •Theme: transforming personal history into usable, delightful formats
- 29:36 – 31:41
Chess dashboard: analysis, trends, and an unexpectedly snarky coach
They show a Muse-created dashboard pulling Chess.com data, mixing trend tracking with last-game reviews. The personality that emerges (light snark) turns coaching into something both useful and entertaining.
- •Chess.com data → game-by-game breakdowns + trends over time
- •Most recent game review as a learning loop
- •Dashboards are defined conversationally in chat
- •Emergent tone/personality can increase engagement
- 31:41 – 34:33
What designers must learn for AI: non-determinism and ‘scaffolding’ over pixels
In a lightning round, Alex explains the biggest shift for designers: building for non-deterministic outputs and highly personalized experiences. Instead of designing one fixed screen, designers must focus on systems, foundations, and acceptable ranges of output quality.
- •Non-determinism changes what ‘correct’ looks like in UI design
- •Designing scaffolding/architecture vs. single expected outcomes
- •Letting go of perfect polish for every generated artifact
- •Personalization means every user’s product is meaningfully different
- 34:33 – 36:31
When Muse is obstinate: prompting style, politeness, and course correction
Claire asks how Alex prompts when the agent won’t cooperate. Alex describes staying polite and human-like rather than aggressive; they note frustration often stems from unclear instructions, especially in expensive coding detours.
- •Polite, person-to-person prompting as a default style
- •Aggressive prompting can trigger ‘reset’ behavior, but isn’t his norm
- •Misadventures are most painful in long/expensive coding tasks
- •Often a ‘me problem’: better direction yields better outcomes
- 36:31 – 37:47
Where to find Alex and how users can influence Muse
They close with how to engage: download Muse, provide feedback inside the product, and follow Alex on Threads for behind-the-scenes design decisions. Claire wraps with standard show sign-off and ways to follow the podcast.
- •User feedback inside Muse reaches the team
- •Alex shares product/design notes on Threads
- •Encouragement to try the app and request features
- •Podcast wrap: subscribe, comment, rate, and follow the show
