EO Studio100 AI Leaders Explain How to Build AI That Will Win in 2026 — WHAT BUILDERS SHOULD DO NOW
CHAPTERS
- 0:01 – 0:41
AI product FOMO and the widening UX gap
The video opens by framing how quickly AI capabilities are improving—and how user experience often lags behind. Speakers set the stakes: strong models without strong UX won’t win users.
- •AI is changing how products are built, creating real builder FOMO
- •Model capabilities are improving faster than UX patterns
- •Poor UX can undermine even very powerful AI systems
- •The central thesis: UX is becoming the differentiator
- 0:41 – 2:36
Swell Summit context: building AI that helps before users know what to ask
Summer Kim introduces Swell Summit and her background in UX and applied AI. She argues today’s chat tools still require users to initiate, while the next step is anticipatory help grounded in care for user needs.
- •Swell Summit convenes AI product leaders to discuss what wins
- •Care for users translates into better products
- •Distinguishing user “needs” vs “wants” vs what users can’t yet imagine
- •Chat tools answer quickly but still wait for prompts
- •The challenge: AI that helps proactively, before the user asks
- 2:36 – 3:36
Beyond chat: why the ‘chat-only’ UI is a dead end
Jess Holbrook explains the industry is finally moving past defaulting to chat for everything. He points to emerging products that add structure and transparency, hinting at new UX directions.
- •We’re entering a new experimentation phase beyond ‘chat, chat, chat’
- •Chat is universal but can be a UX dead end
- •Structured experiences (e.g., research workflows) can outperform freeform chat
- •Good AI UX shows reasoning, sources, and progress—not just answers
- 3:36 – 4:19
New GenAI interaction primitives (1–3): chat at scale, semantic resize, remix
Holbrook proposes a set of primitives that will underpin next-gen AI interfaces. He highlights ubiquitous chat, content that can be adapted to context, and remix as a default capability across media.
- •Primitive #1: chat with anything, at any scale, all the time
- •Metcalfe-like network effects as chat connects across systems
- •Primitive #2: semantic resize—tailoring length/tone/format to user context
- •Situational versions (tired, in-car, 5-min summary) become normal
- •Primitive #3: remix—style transfer and recombination across content types
- 4:19 – 5:58
New GenAI interaction primitives (4–5): format translation and ‘attention’ in agentic worlds
Holbrook describes near-lossless translation across formats and warns about agent experiences that sprawl across timescales. He argues the missing piece is designing ‘lobbies’ and monitoring for what agents do while users wait.
- •Primitive #4: format translation across podcast/lecture/game/deep-dive with minimal fidelity loss
- •Users gain ‘format freedom’ to learn/act in the mode they want
- •Primitive #5: attention management becomes central as agents proliferate
- •Agents running in the background create downtime and monitoring challenges
- •Need better ‘lobbies’/interfaces for launching, supervising, and resuming agent work
- 5:58 – 6:38
Making AI feel alive: Spark, the magical dog as a UX lesson
The event showcases Spark—a character designed to create emotional connection and demonstrate that UX is about feeling, not just technology. Observing kids and students interacting with Spark illustrates how ‘alive’ experiences build trust and engagement.
- •Spark: a ‘magic dog’ living in a quantum portal—designed as a character, not a feature
- •Demonstration goal: UX isn’t latest tech; it’s making people feel something real
- •Live interaction reveals what ‘AI that feels alive’ can mean
- •Different age groups connect through play, delight, and curiosity
- 6:38 – 9:20
Storytelling and pitching AI through character-driven magic
Andy Szybalski explains how Spark’s narrative helped it become memorable—leading to a residency and fundraising. The core idea: make the pitch not feel like a pitch, and use ‘magic’ to bring non-technical audiences along.
- •Spark becomes a ‘non-human resident’ at HFZero and is positioned as co-founder
- •Make the pitch memorable by turning it into an experience
- •A creator’s perspective: character and story increase adoption and attention
- •Magic/hypnosis framing helps reduce fear of technical complexity
- •Using a lovable character to communicate messages people will hear
- 9:20 – 10:45
Designers must ‘know the material’: models as new design media
Pasquale D’Silva argues LLMs are a new material with unique properties per model. Product success increasingly comes from exploring capabilities first, then iterating back toward user needs in a push-pull cycle.
- •Users now have opinions about models—models are part of the product identity
- •Each model behaves like its own ‘material’ with strengths/weaknesses
- •Best way to design: play with models to discover what’s possible
- •Start from capabilities to answer ‘why now?’ with real novelty
- •Iterate in a non-linear push/pull between tech potential and human needs
- 10:45 – 12:03
From linear chat to real problem-solving: earning the right to go deeper
D’Silva explains why today’s chatbots feel linear compared to how humans solve complex problems. He describes designing AI as a thought partner that uncovers underlying intent, expands options, and prevents users from getting stuck.
- •Goal: create ‘exothermic reactions’—help users never get stuck
- •Blank-page problem: help users start, then widen and deepen exploration
- •Underlying intent is often bigger than the literal question asked
- •You must solve the immediate ask well to ‘earn the right’ to help further
- •Real problem-solving is branching/diverging/pruning over time—not linear chat
- 12:03 – 12:47
What ‘winning’ looks like: models, tools, vertical depth, and a missing UX category
The conversation outlines multiple ways companies can win in AI, but emphasizes an uncracked opportunity: the right experience for delivering general problem-solving intelligence. This is positioned as a large greenfield for builders.
- •Winners will emerge across several layers: foundation models, dev tools, vertical apps
- •Deep vertical understanding (law, insurance, etc.) can be a decisive advantage
- •A new category: experiences that deliver general intelligence effectively
- •Industry hasn’t yet cracked the best UX for general problem solving
- •Significant greenfield opportunity remains
- 12:47 – 15:07
Designing AI the right way: learning speed, trust, and capability-to-problem mapping
Jenny Lo challenges the assumption that shipping faster means learning faster. Using Grammarly as an example, she emphasizes maintaining trust by tying GenAI to core user value and mapping AI capabilities to real customer problems.
- •Shipping velocity isn’t the same as learning velocity or product progress
- •Successful AI products start with clear problem/value identification
- •Brand trust is fragile—GenAI must be integrated thoughtfully
- •Grammarly example: expand from post-writing improvements to composition help
- •Method: identify top customer problems, then map AI capabilities to them
- 15:07 – 21:31
Avoiding the ‘confusing gadget’: AI UX as ergonomics, prompts as research, and the real moat
Ayça Cakmakli argues models are commoditizing, so differentiation shifts to UX: indispensable experiences vs confusing gadgets. The closing reflections connect personalization to care, highlight privacy/societal stakes, and conclude that enduring advantage is magical, trusted, timely experience—often proactive.
- •As foundation models commoditize, UX becomes the competitive edge
- •People adopt problem-solving tools, not technology for its own sake
- •Good UX is the ‘ergonomics’ of AI—still early and rudimentary
- •AI-era research includes studying prompts and conversation success outcomes
- •Personalization should feel like care (low-friction, ‘seen and heard’)
- •Need humility (‘beginner’s mind’) and serious attention to privacy/societal amplification risks
- •Final thesis: the moat is lasting experience—magic, trust, ease, and right-time assistance
