Devoured - April 21, 2026
I do Design Innovation. I barely open Figma anymore. (5 minute read)

I do Design Innovation. I barely open Figma anymore. (5 minute read)

Design aifrontend Read original

A designer at an AI-native startup explains how their work shifted from Figma mockups to writing code prototypes and behavioral specifications that both engineers and AI systems can execute.

What: A first-person account from a designer at Swoop, a 20-person AI-native team, describing how AI features—which center on timing, conversation, and dynamic behavior rather than static screens—require new design artifacts like code prototypes and pull requests instead of traditional Figma files.
Why it matters: This represents an emerging model of design work where traditional role boundaries blur: designers prototype in code, pull requests function as interactive specifications, and design systems are written as markdown files consumed by both humans and AI models, though the author acknowledges this pattern works best in small, code-fluent teams.
Takeaway: If you're building AI-heavy features, consider whether behavioral specifications (timing logic, decision thresholds, edge cases) communicated through runnable code prototypes might work better than static mockups for design-engineering collaboration.
Deep dive
  • AI-native features often have minimal traditional interface—short conversations, voice prompts, background agents—making static Figma frames inadequate for capturing timing, behavior thresholds, and intervention logic
  • Three working modes emerged: AI-logic-heavy (designer writes behavioral descriptions, engineers implement), UX-heavy (designer builds end-to-end in code), and innovation (designer vibe-codes prototypes in Cursor or Claude Code)
  • Pull requests function as executable behavioral specifications that engineers can run, modify parameters on, and observe effects—unlike Figma prototypes that only approximate behavior
  • A fourth mode involves designing frameworks and interaction patterns that constrain entire feature families, requiring cross-team context and Director-level authority
  • Design artifacts now serve two readers: human engineers who build from them and AI systems that execute them, eliminating ambiguity that humans could resolve by asking questions
  • The team's design system lives in a single DESIGN.md file consumed by both engineers and the AI model, with no parallel Figma library
  • This workflow succeeds because the team is small, AI-native from inception, and has senior engineers comfortable treating designer PRs as reference material rather than production code
  • The author frames this as a "second shape" of design work coexisting with traditional Figma-centered workflows, not a replacement—enterprise teams with rigid roles still use established processes
  • The shift isn't about designers becoming engineers but about design judgments shipping through code-adjacent artifacts when designing behavior rather than surfaces
  • Management overhead is distributed across the team rather than concentrated, enabled by everyone shipping code and working across traditional role boundaries
Decoder
  • IC: Individual Contributor, a non-management role focused on hands-on execution rather than people management
  • Vibe-code: Informal prototyping in code to explore how a feature should feel, typically not production-ready but executable enough to demonstrate behavior
  • DESIGN.md: A markdown file containing design system specifications readable by both human developers and AI models, replacing traditional visual design libraries
  • Behavioral spec: A description of how a system should act—timing, thresholds, decision logic, edge cases—rather than how it should look
  • AI-native team: A team built from inception around AI capabilities where workflows, roles, and tooling assume AI as a core part of the stack rather than an add-on
Original article

In AI-native teams, design is shifting away from static tools like Figma toward shaping behavior—timing, logic, and interaction—which can't be fully captured in traditional mockups. Designers increasingly work across coding, research, and strategy, using prototypes and pull requests as “behavioral specs” that engineers and even AI systems can interact with directly. This creates a new model of design work: faster, more code-adjacent, and focused on systems and frameworks rather than screens, while traditional Figma-based workflows still coexist in more structured environments.