Beyond Figma: AI Design Tools From Asset Libraries to Workflow Infrastructure

This piece explores a recent shift in design tooling, arguing that design tools are moving from simple "asset libraries" toward becoming foundational workflow infrastructure that can be consumed by AI, teams, and developers.

1. Product Design Psychology – an online design‑psychology book

Wouter de Bres announced a ten‑year project, Product Design Psychology, an online book offering 40 free chapters. The book is organized into four sections:

The Designer's Mind – explores designers' own cognition with chapters such as Nobody Thinks Like You, You Ruin Your Own Designs, Taste Isn't Talent, and Deadlines Make You Dumb.

Minding the Design – examines how interface mechanisms affect perception and behavior, e.g., Pass the Vibe Check First, Your Design Is Rigged, Intuitive Design Is a Lie, Fake Progress Is Real Motivation.

The User's Mind – discusses user psychology entering a product, with chapters like Users React, Then Rationalize, Users Will Hate Your New Design, More Options Make Users Quit.

The Organization's Mind – looks at how organizations shape final designs, e.g., Good Design Dies in Meetings, The Metric Is Not the User, Research as Alibi.

The author highlights that the book forces designers to consider "who's brain you're dealing with" rather than merely focusing on aesthetics.

2. Vercel design.md – a design system written as AI‑readable rules

Vercel open‑sourced two markdown files, design.md (light theme) and design.dark.md (dark theme). The files contain concrete tokens for colors, fonts, spacing, corners, components, motion, and copy guidelines, often with explicit numeric values.

“Use motion only when it clarifies a change, never for decoration.”

Key observations:

Colors convey information (blue for success/focus, red for error, amber for warning) rather than decorative purposes.

Typography is split into Geist Sans for UI/text and Geist Mono for code and data.

Heading sizes include precise font‑size, weight, line‑height, and letter‑spacing values; larger headings use tighter letter‑spacing (e.g., -4.32px for 72 px headings).

Corner radii follow a family system: 6px for surface controls, 12px for menus/pop‑ups, 16px for large panels, 9999px for pills and avatars.

Interaction durations default to 0ms for instant feedback.

The author argues that these explicit rules make the design system readable by both humans and AI agents, enabling consistent UI generation.

3. serve‑sim – an AI‑friendly wrapper for Apple simulators

serve‑sim

(npm version 0.1.43) packages the Apple Simulator as a local service optimized for AI coding agents such as Cursor, Codex, and Claude Desktop. It provides unified device panels, fast device switching, streaming controls, and a simple launch command: npx serve-sim The tool bridges the gap between AI agents and real‑device contexts, allowing developers to see status bars, keyboards, safe‑areas, font scaling, and runtime errors—information that is otherwise hard for language models to infer.

4. Cowart – a Codex‑powered infinite canvas plugin

Cowart, built on tldraw, offers a local infinite canvas that persists to a canvas directory. Its capabilities include:

Opening a local tldraw canvas inside Codex.

Persisting canvas pages and images alongside the project.

Creating AI image placeholders that Codex can fill.

Uploading annotated screenshots for Codex to generate clean images.

Reading selection state and inserting images via a Cowart MCP tool.

Unlike pure image‑generation tools, Cowart focuses on the collaborative workflow: designers annotate directly on the canvas, Codex reads the annotations, and produces revised images while preserving the original and the edit history.

Overall Trend

The four items illustrate a common direction: providing richer context for AI. The book supplies cognitive context, Vercel’s design.md supplies system context, serve‑sim supplies runtime context, and Cowart supplies visual‑collaboration context. Together they answer the question, “How can AI move from merely hearing prompts to actually participating in the work site?”

The author predicts that in the next one‑to‑two years, designers will shift from curating visual assets to curating contextual rule sets that AI can consume—brand tokens, design‑system specifications, user‑research insights, canvas annotations, and real‑device states.

Practical Takeaways

Designers should read the Product Design Psychology chapters that challenge their assumptions (e.g., You Ruin Your Own Designs).

Product managers can use Vercel’s design.md as a template for writing clear design specifications.

Developers interested in AI‑assisted coding should experiment with serve‑sim and Cowart to bring visual context into their workflows.

By integrating these tools, teams can treat AI as a genuine partner in the design workflow rather than a mere code‑completion gadget.