---
title: "Figma Founder Dylan Field: AI Will Not Replace Human Designers"
author: "Unique Research"
sourcePublication: "Unique Research (FFCAP English library)"
originalPublishedAt: "2023-10-19T12:45:00+08:00"
canonical: "https://ffcap.cn/en/research/20231019a04lau00"
source: "https://ffcap.cn/en/research/20231019a04lau00"
language: "en"
cnSourceUrl: "https://view.inews.qq.com/a/20231019A04LAU00"
---

# Figma Founder Dylan Field: AI Will Not Replace Human Designers

_Unique Research · Insight · October 19, 2023 · From a16z AI Revolution series with David George · Original: [a16z.com/democratizing-design](https://a16z.com/democratizing-design/)_

## Key points

-   AI will not replace designers. It may draft or suggest elements; humans still iterate the final product.
-   AI enlarges the design job: lower floor to participate, higher ceiling on quality and range.
-   FigJam (plus Jambot) connects brainstorming → Figma Design → Dev Mode, with graph-linked prompts.
-   Further AI progress should give designers better prompts and option sets — not remove them.

Figma went from ~$100M valuation in early 2018 as a niche design tool to ~$10B by 2021 — a 100× jump — becoming a must-have alongside Adobe in product and design circles.

## Will AI replace human designers?

**David:** Will AI really take designers’ jobs?

**Dylan:** I don’t think so. The question hides two claims: “less to design” and “AI does all design.” Every prior platform shift — print, mobile — created _more_ design work, not less. OpenAI is hiring product and design hard for consumer excellence. Design matters.

We’re at “AI can do a first draft.” Getting from draft to ship still needs a team. If AI starts suggesting UI elements in a meaningful way, we get a new era of contextual, intent-responsive design — fascinating for designers, not a replacement.

AI first lets everyone do more design work: lower barrier to enter the process, higher ceiling on what you can achieve. It helps climb the abstraction ladder — from high-level prompts to pixels to motion curves — and jump across solution spaces before deep iteration. That’s what Figma has been exploring in FigJam.

## FigJam + AI: Jambot

FigJam is Figma’s whiteboard for brainstorming; Design for craft; Dev Mode bridges design to code. Dylan demoed **Jambot**: connect sticky notes or screen regions to prompt agents. Example: feed talk topics → generate novel titles → switch to haiku variants. History stays as a graph, so exploration is nonlinear versus a linear chat. Another path: “AI startups worth watching” → company list → “rabbit hole” related topics → rewrite stickies in a Twitter voice with fewer emojis. The demo came from a hackathon.

## Human + AI creativity

David recalled Peter’s thesis when a16z first backed Figma: a “decade of design” where interfaces lead code, not the reverse. Dylan’s modern Turing-test framing: GPT-4 can draft a business plan and pitch — but if you must choose between a human-run and AI-run competing company to invest in, we’re not yet at parity.

Figma’s vision is everyone contributing to design via systems, kits, and components — design-system teams at scale already make that real. Role boundaries are blurring: best designers think more about code; best engineers think more about design; PMs ship mocks to communicate ideas faster. You still need humans to choose what to explore, refine drafts into finals, and — for now — even generate a decent open-source README. Code autopilot is still far.

## Applying AI to design

Figma’s document model is closer to an AST than an image. Copilot-like approaches may beat pure diffusion for design generation. GPT-4 can still emit simple XML layouts that spark ideas. Figma’s approach: don’t put all eggs in one basket; spread AI across the product; then ask which uses have acceptable cost and predictable usefulness.

## Startups vs incumbents

For startups, now is a good time to chase domains that truly benefit from the tech — much of the near term sits in foundations / infra / developer tools. David frames systems of record evolving into systems of prediction and, with AI, systems of _action_. Dylan still expects a human in the loop for a long time — and likes action systems that clear obvious tasks so people can do higher-level work.

Advice against cargo-culting “build for years like early Figma then talk to users”: Figma didn’t work that way — ship fast. Between today’s LLMs and AGI there may be a plateau; use that window. If Dylan founded again, he’d chase underexplored scientific applications of deep learning — accelerating approximation and human progress.

---

Editorial English edition published on UniqueCapital: https://ffcap.cn/en/research/20231019a04lau00
On-site reading page: https://ffcap.cn/en/research/20231019a04lau00