AI Interfaces Of The Future | Design Review

1. Raphael Schaad on Designing AI Verbs Over Nouns

Timestamp: 00:00:26 to 00:03:07

Raphael Schaad argues that traditional 2D software consists of static nouns on a screen like text, forms, dropdowns, and buttons, whereas AI software centers on verbs like autocomplete, auto-suggest, and autonomous information gathering. Designers do not yet have established tooling to draw these verbs on screen, making emerging AI interfaces a fascinating frontier.

Significance (High): Reframes AI product design around representing autonomous actions and workflows rather than static visual controls.

Sources in support: Raphael Schaad (Founder and Designer of Cron (Notion Calendar)), Aaron Epstein (Y Combinator General Partner and Host)

2. Latency as the Core Interface in Vapi

Timestamp: 00:03:07 to 00:05:14

Aaron Epstein notes that in voice AI, response latency acts as the interface itself because any delay breaks the illusion of conversing with a human. When testing Vapi's interruption handling, Raphael Schaad points out that the voice agent fails to pause when spoken over and completely misses Aaron's weather question.

Significance (High): Highlights that millisecond response times and graceful interruption handling matter more to voice UX than visual polish alone.

Sources in support: Aaron Epstein (Y Combinator General Partner and Host), Raphael Schaad (Founder and Designer of Cron (Notion Calendar))

3. Testing Retell AI with a Live Phone Curveball

Timestamp: 00:05:14 to 00:07:56

During a live simulated $800 debt collection call from Retell AI, Aaron Epstein tells the agent that his name is actually Steve rather than Aaron, and the agent adapts immediately to call him Steve. However, Raphael Schaad observes that noticeable latency and the agent abruptly ending the call when told Aaron is on vacation reveal its robotic limits, making it best suited as a first-line filter before human handoff.

Significance (High): Shows how conversational voice agents can handle mid-call context shifts to automate first-line call center grunt work before escalating to humans.

Sources in support: Aaron Epstein (Y Combinator General Partner and Host), Raphael Schaad (Founder and Designer of Cron (Notion Calendar))

4. Visualizing Agent Workflows on Gumloop's Canvas

Timestamp: 00:07:56 to 00:10:46

Aaron Epstein introduces autonomous AI agents using Gumloop, demonstrating how an open-ended, pan-and-zoom canvas lets humans oversee each step of a Y Combinator directory web-scraping flow. Because autonomous agents make multi-step decisions behind the scenes, visual workflow blocks give users transparency and step-by-step control.

Significance (High): Establishes the infinite node canvas as a primary paradigm for monitoring and steering autonomous multi-step AI agents.

Sources in support: Aaron Epstein (Y Combinator General Partner and Host), Raphael Schaad (Founder and Designer of Cron (Notion Calendar))

5. Semantic Zoom and Branching on AI Canvases

Timestamp: 00:10:46 to 00:13:13

Critiquing Gumloop's node canvas, Raphael Schaad recommends adding a color legend, collapsing unreadable small text into solid color blocks when zoomed out, and showcasing multi-dimensional branching templates rather than linear flows. Aaron Epstein predicts that interactive canvases could become the standard interface ten years from now for controlling autonomous agents, reviving 50-year-old chip design flowcharts in an interactive form.

Significance (Medium): Offers concrete UI improvements like semantic zoom and non-linear branching templates to make node-based agent builders usable at scale.

Sources in support: Raphael Schaad (Founder and Designer of Cron (Notion Calendar)), Aaron Epstein (Y Combinator General Partner and Host)

6. Clickable Prompt Suggestions and Agent Spreadsheets

Timestamp: 00:13:13 to 00:16:20

Reviewing AnswerGrid, Raphael Schaad praises turning example prompts into single-click buttons to overcome blank-canvas paralysis and suggests dynamically inferring clickable prompts from app context. Aaron Epstein then demonstrates how clicking 'AI companies in San Francisco' generates a structured spreadsheet where adding a 'funding raised' column dispatches an AI agent for every individual cell.

Significance (High): Demonstrates how combining one-click prompt starters with cell-level web-scraping agents turns a blank prompt into structured, expandable research.

Sources in support: Raphael Schaad (Founder and Designer of Cron (Notion Calendar)), Aaron Epstein (Y Combinator General Partner and Host)

7. Inline Source Citations Build Trust in AnswerGrid

Timestamp: 00:16:20 to 00:18:57

When AnswerGrid returns funding figures like 470 million, 73.6, and 6.6 without consistent currency units, Aaron Epstein clicks into the cell popover to inspect the attached source link confirming OpenAI raised 6.6 billion. He and Raphael Schaad note that attaching inline footnotes, a pattern popularized by Perplexity and rooted in academic papers, lets users immediately audit AI agent outputs against hallucinations.

Significance (High): Shows why inline source citations are essential for users to catch formatting omissions and trust data gathered by autonomous agents.

Sources in support: Aaron Epstein (Y Combinator General Partner and Host), Raphael Schaad (Founder and Designer of Cron (Notion Calendar))

8. Handling Long Generation Waits in Polymet

Timestamp: 00:18:57 to 00:21:49

While testing Polymet's multimodal prompt-to-UI generator, Aaron Epstein and Raphael Schaad wait with no clear time estimate as humorous status messages flash by. They propose exposing an under-the-hood progress log for technical users or borrowing from flight meta-search engines by rendering low-resolution early results that users can filter while full generation completes.

Significance (Medium): Addresses a major UX bottleneck in generative AI by proposing progressive rendering and persistent progress logs over opaque loading spinners.

Sources in support: Aaron Epstein (Y Combinator General Partner and Host), Raphael Schaad (Founder and Designer of Cron (Notion Calendar))

9. Prompt Feedback Loops and Sub-Element Diffs

Timestamp: 00:21:49 to 00:25:17

After Polymet generates a glassmorphic treasury dashboard, Raphael Schaad suggests offering draggable design-term pills for users unfamiliar with jargon like 'glassmorphic' or 'skeuomorphic', as well as visually highlighting which words in a prompt the AI actually respected or ignored. Aaron Epstein then selects the sidebar module and prompts Polymet to turn just that sub-element blue while preserving the rest of the layout.

Significance (High): Identifies modular diff prompting and prompt-execution highlighting as frontier techniques for iterative human-AI design collaboration.

Sources in support: Raphael Schaad (Founder and Designer of Cron (Notion Calendar)), Aaron Epstein (Y Combinator General Partner and Host)

10. Adaptive Email Actions and Hotkeys in Zuni

Timestamp: 00:25:17 to 00:30:20

Examining adaptive AI interfaces where content dynamically dictates the UI controls, Aaron Epstein and Raphael Schaad review Zuni's founder email app, which generates context-specific reply actions and custom follow-up inputs like call-time pickers for each email. While single-letter hotkeys like pressing 'Y' keep keyboard users fast and consistent even as action buttons change, Raphael warns that unmodified hotkeys risk misfiring if users think a text input is focused.

Significance (High): Balances the efficiency of context-adaptive controls against the usability risks of shifting buttons and unmodified single-key shortcuts.

Sources in support: Raphael Schaad (Founder and Designer of Cron (Notion Calendar)), Aaron Epstein (Y Combinator General Partner and Host)

11. Trading Fidelity for Immediacy in Argil's AI Video

Timestamp: 00:30:20 to 00:34:41

Aaron Epstein demonstrates Argil's AI video studio using a deepfake trained on a few minutes of his own footage, showing how users type a script and manually pick camera angles and body language gestures. Because rendering lip-synced video takes 12 minutes, Argil immediately plays a blurred video preview with newly generated audio, a trick Raphael Schaad praises for trading fidelity for immediacy so humans can iterate quickly.

Significance (High): Showcases a clever design pattern for heavy generative media: pairing fast audio generation with blurred visuals to enable rapid script iteration before committing to a 12-minute render.

Sources in support: Aaron Epstein (Y Combinator General Partner and Host), Raphael Schaad (Founder and Designer of Cron (Notion Calendar))

12. Reimagining Software Components Like the Touch Era

Timestamp: 00:34:41 to 00:36:20

Wrapping up the review, Raphael Schaad compares the current explosion of native AI components to the 2010 shift when touchscreens forced designers to reinvent every software primitive from scratch. Aaron Epstein concludes that across voice, canvas, adaptive, and video tools, the central design challenge is keeping the human in the loop and in control while AI executes complex tasks.

Significance (Medium): Synthesizes the episode's thesis: AI is triggering a foundational UI paradigm shift comparable to mobile touchscreens, centered on human-in-the-loop control.

Sources in support: Raphael Schaad (Founder and Designer of Cron (Notion Calendar)), Aaron Epstein (Y Combinator General Partner and Host)

Y Combinator Community Showcase (Medium severity)

Type: Commercial

Host Aaron Epstein is a General Partner at Y Combinator, which produces the show, solicits accelerator applications in the description, and reviews products submitted by the Y Combinator community.

Significance: The institutional tie incentivizes selecting promising community startups and framing their interface shortcomings as constructive growth opportunities rather than disqualifying flaws.