All-In

Dylan Field on Why Design Is the New Code

Dylan Field· Co-founder and CEO of Figma at Figma
·~40 min·English·All-In Podcast
AI CompanyBusiness StrategyAgents
TL;DR

Figma's CEO argues that as AI commoditizes code and production, the scarce and defensible layer becomes design, point of view, and the human choice to create rather than consume.

01The Core Thesis

Design Is the New Code

When AI can generate code on demand, the scarce, defensible layer moves up to design, brand, and point of view.

It's not code cuz anyone can create code now. That's not the hard part anymore. So, my view is design is the new code. It's the layer above.

— Dylan Field, All-In
Key Insight
Field is describing a margin shift, not a layoff: engineers still matter at Figma, but the differentiation premium migrates to whatever the model can't commoditize. His evidence is a supply-and-demand squeeze — the App Store is on the same exponential as everything else while app usage stays flat, so when everyone can ship software, the tiebreaker is taste.
02Origin Story

The $20B Deal That Died

Adobe agreed to buy Figma for about $20 billion, 16 months of regulatory review ended the deal, and the hardest part was the ambiguity, not the lost money.

the mood was relief because it wasn't about relief of not Adobe, it was relief of not having ambiguity. To have ambiguity that long is really intense.

— Dylan Field, All-In
Key Insight
The leadership move hidden in this story is the all-hands offer: anyone who had joined expecting Adobe could leave with a few months' pay and a standing option to reapply. Field treated attrition as a feature, not a wound — letting the mismatched self-select out is how he converted a demoralized workforce back into a committed one without pretending the last 16 months hadn't happened.
03Business Reality

SaaS Isn't Dead — It Splits

The SaaS-pocalypse story is too tidy; software companies are bifurcating into the specialized and founder-led survivors versus the undifferentiated rest.

it's not true that all SaaS companies are fine and it's also not true that all SaaS companies are dead.

— Dylan Field, All-In
Key Insight
Field's counter-example is pointed: Cursor out-grew frontier labs that were chasing it with their own models, before Cursor had a model of any kind, by obsessing over the user. The lesson he draws is that owning the model is not the moat — the moat is a specialized workflow built for an audience you actually know, which is exactly why Figma accelerated to 48% year-over-year growth even after a general-purpose rival launched a design product.
04The Mental Model

AI Lowers the Floor; Humans Raise the Ceiling

AI democratizes production by pulling the floor up, but it does not lift the ceiling of what's possible — only humans who choose to be bold do that.

AI is not raising the ceiling. Humans are raising the ceiling, though.

— Dylan Field, All-In
Key Insight
This is a public correction: Field says he told his own Config conference years ago that AI would raise the ceiling, and he now calls that wrong. The distinction matters for how you spend your time — if the floor is rising for free, competing on floor-level output is a losing game, and the durable work is the ceiling-raising kind Field says only humans do, when they choose to focus, invest their time, and be bold.
05How Newness Works

Create By Leaving the Distribution

Making something genuinely new means operating outside the model's training distribution, not somewhere near its statistical mean.

you are operating outside of this distribution of training.

— Dylan Field, All-In
Key Insight
Field's warning is about authorship: if you hand AI your vision and trust it to implement, you get the result through the lens of the model's own distribution, so you never quite get what you wanted. Prompting without scaffolding lands you near the mean — competent and average — while the work that gets rewarded by the market is the deliberate move out beyond that usual distribution, where a model has the least to go on.
06The Product Answer

Make Code a Material You Can Mold

Generating with AI is fine as a starting point, but you need a way to keep shaping the output by hand rather than accepting it.

it's not bad to generate with AI, but you need a way to craft that clay. You need a way to mold it.

— Dylan Field, All-In
Key Insight
The real value Field assigns to generation is narrow and honest: it solves the blank-canvas problem, giving you something concrete to react against. Figma's design decision follows from that — treat code as a material on the canvas (and build node-based tools like Weave) so the creator can oscillate between letting the machine draft and taking over by hand, rather than choosing one mode for the whole job.
07Why Progress Is Lopsided

Math Sprints, Design Crawls

Models improve fastest where answers can be checked, and slowest in creative domains because there is no clean grade to optimize against.

And then, for creativity and for design, it's very incremental progress overall. Uh, maybe that will change at some point, but I think the fact that these are non-verifiable domains has something to do with it.

— Dylan Field, All-In
Key Insight
Field is naming the same reward-signal asymmetry that drives reinforcement learning: math and cybersecurity have crisp right-or-wrong checks, so capability stacks quickly, while design has no verifier to grade against, so gains stay incremental. That asymmetry is also his business thesis in disguise — the domains where AI crawls are precisely the ones where human taste keeps its value longest.
08The Human Choice

Consume, or Create

The real fork in an AI world is whether you let the tools think for you or use them to reach things you could never have reached before.

Do you want to consume or create? Do you want to take the easy path or do you want to like climb the mountain?

— Dylan Field, All-In
Key Insight
Field refuses to frame this as elite creatives versus everyone else — his claim is that anyone can be a maker, so the dividing line is willpower and curiosity, not credentials. The closing note is unsentimental: he watched OpenAI's 700-plus-paper math drop and felt both awe and grief, and he argues it is not wrong to mourn a little lost mystery even while choosing to keep creating.