AI Interfaces Of The Future | Design Review

YC GP Aaron Epstein and Notion Calendar founder Raphael Schaad critique seven live AI interfaces and extract design principles for the next decade of software.

• Raphael frames the core shift: traditional software is nouns (buttons, forms), AI software is verbs (autocomplete, go-fetch, execute) — and UI tooling for verbs barely exists yet.
• Voice AI latency is the interface: every extra millisecond breaks the human illusion, making it feel robotic regardless of voice quality.
• Argil deepfakes require only a few minutes of training video; it shows a blurry audio-first preview so users can iterate on script before committing to a 12-minute full-video render.
• AnswerGrid treats every spreadsheet cell as its own AI agent dispatched in parallel — structured data output from free-form prompts with inline sourced citations per cell.
• Zuni’s adaptive email UI changes suggested-response buttons per email but keeps keyboard shortcuts constant, solving the predictability problem of shifting UI elements.
• Canvas interfaces (Gumloop) are predicted to become the standard control panel for autonomous agents over the next decade, reviving flowchart paradigms made interactive.
• Raphael compares this moment to 2010 touchscreen emergence: all software components are being fundamentally reimagined, not just incrementally updated.

2025-02-27 · Watch on YouTube