What this covers

This is a conversation between host Patrick and Dylan Field, Figma's co-founder and CEO, who last appeared on the show in 2020. It starts with Figma's origins, including how WebGL made complex graphics applications possible in the browser and led the team to rip out earlier cross-platform desktop work. From there it moves to what Field would do differently. He wishes Figma had found the smallest inspiring increment of its idea sooner, and he describes the trouble of building in private, where teams can't viscerally believe users will care until a launch draws a reaction.

The middle of the episode covers design craft. Field explains why digital products differ from physical ones, since software can ship A/B tests and roll back failures. He names common UX mistakes, such as designing isolated states rather than flows, and he argues for direct manipulation over interfaces that act on one screen to change another. He also describes playing recorded user-research sessions for whole teams, and he explains how he thinks about delegating across a company, staying closest to design because losing quality there could lose the business.

The AI section is the most speculative. Field says the jump to o1-Pro was the first time something felt like intelligence on the other end, and he calls such a system functionally AGI, a claim he concedes many would dispute. He forecasts that the first superintelligence will appear in math research. His biggest safety worry is removing the human in the loop in military settings, and he is less concerned about wireheading. He also describes Figma's plan to lower the floor and raise the ceiling through first drafts, and he expects design literacy to become as ordinary as knowing Word.

Sharpest takeaway

Field argues that AI reasoning is underhyped in capability but overhyped as a finished product, so design will become more accessible while human craft, taste, and judgment remain the scarce differentiator.

  • Current AI reasoning models produce markedly better analysis, but still require human guidance and cannot yet handle full end-to-end work reliably
  • Lowering the barrier to making software increases the volume of output, which raises the value of craft and context rather than eliminating it
  • Figma's AI strategy focuses on low-hanging product improvements and first drafts that users can refine, not fully autonomous design

The claims · ranked55 claims · weighted by value

This asset isn't compiled yet

You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.

0.81

WebGL made complex graphics applications possible in the browser for the first time, which is what allowed Figma to exist, and Figma's early cross-platform desktop-targeting work was ripped out once WebGL proved production-ready.

causalhigh valueestablishednovelty 3/4durability 3/4· Dylan

“With WebGL, something like Figma became possible for the first time. In the past, you had to be desktop in order to do complex graphics, an application”

0.80

The first mistakes people make in software UX are deciding when to reuse familiar patterns versus invent new ones, and failing to consider user flows and where users come from, which leads them to design for isolated states.

normativehigh valueestablishednovelty 2/4durability 4/4· Dylan

“the first things I see people mess up are when to reuse versus when to be inventive. ... I would say that people fail to consider flows.”

0.80

Entrepreneurs often find a novel opportunity and build it for money without real motivation, which almost always leads to burnout, so people should focus on work they are truly driven by, especially since high-agency people deliver the best results.

normativehigh valueestablishednovelty 2/4durability 4/4· Dylan

“a lot of entrepreneurs fall into the trap of they find something and they go and they like go build that thing and they're making some money from it and they just keep going.”

0.80

Good design should make simple things simple and complex things possible, and the main risk for Figma is becoming too power-user oriented, adding complexity that is powerful but not approachable or useful.

normativehigh valueestablishednovelty 2/4durability 4/4· Dylan

“Make the simple things simple, the complex things possible. And that's not original Figma's that goes back a long ways.”

0.75

Physical products require iterating upfront because they ship once and remain fixed, whereas digital products can ship A/B tests constantly and roll back failures, so good design in software comes largely after the initial draft.

causalhigh valueestablishednovelty 2/4durability 3/4· Dylan

“In digital design, we ship AB tests all the time. We're constantly trying new things, seeing if they work. If they don't, you roll them back. No big deal.”

0.75

Playing back recorded user research sessions to the team in group settings, with people struggling in the product for minutes, motivated engineers to solve problems far more effectively than written summaries of customer feedback.

causalhigh valueestablishednovelty 2/4durability 3/4· Dylan

“we just sit there and awkwardly watch people struggle with the product for five minutes. That was the way to get people motivated”

0.73

Interfaces with indirect manipulation, where the user must act on one screen to see the effect on another, are worse than direct manipulation, and the closer the manipulation is to its visible effect, the better the experience.

normativehigh valueestablishednovelty 2/4durability 4/4· Dylan

“the closer you can get the manipulation to the effect I think is better.”

0.68

Among AI safety concerns, the highest risk is removing the human in the loop in military contexts, because it reduces visibility and transparency, can escalate quickly, and creates strong temptation to do it, which could be very dangerous.

normativehigh valuecontestednovelty 2/4durability 3/4· Dylan

“the one that I think is the highest risk in terms of lack of visibility and transparency and also the potential for it to escalate quickly is if you remove a human in the loop in the military context.”

0.68

Broad segments of the population have become addicted to things in the past and society has still gotten through, so people's habituation makes Dylan less fearful of wireheading than of military AI risks, and he is optimistic because thoughtful people work on military AI.

normativehigh valuecontestednovelty 2/4durability 3/4· Dylan

“there's also a pattern though of people becoming immune to things over time.”

0.65

Teams building in private cannot viscerally believe users will care until they launch or make a large enough announcement, and even announcements are hard for unproven products to scale into reach.

causalhigh valueestablishednovelty 1/4durability 3/4· Dylan

“I think you don't know that viscerally until you get something out into the market or at least announce.”

0.59

There is no shared definition of AGI, but a system that can talk at the depth o1-Pro can is functionally AGI in Field's view, even though it is not yet doing real work, and many people would dispute that.

normativehigh valuecontestednovelty 2/4durability 1/4· Dylan

“Is it doing work yet? Not just yet, but it feels like to me it's functionally AGI. And plenty of people would debate me on that one.”

0.54

Mobile screens have far more possible interaction patterns than are being explored, but the limiting factor is how many new gestures users can be taught; Apple had to teach long-press text selection, and attention to learning is finite.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Dylan

“there's so many more patterns and interactions that could be explored than are being explored. Is really limited by how much people can learn.”

0.53

Reasoning-based products should expose the tree of possibilities to users rather than hide it, because much of the value in design lies in seeing the branching options, though this depends on the domain and use case.

normativehigh valuespeaker onlynovelty 3/4durability 2/4· Dylan

“I actually think it could be even more interesting honestly to expose the tree of possibilities rather than try to hide away from somebody.”

0.53

The first superintelligence is likely to show up in research domains, especially math, where novel theorems will be proven first in inelegant ways, after which humans will find elegant proofs once they know it is possible, as with the four-minute mile.

forecasthigh valuespeaker onlynovelty 3/4durability 2/4· Dylan

“I think we'll see a lot of superintelligence first is actually in research domains, I think math in particular. We're going to see novel theorems proven.”

0.52

Figma's AI strategy is to lower the floor and raise the ceiling: generate a first draft that users can build on, iterate on, and explore as a tree of possibilities, so they can deep dive on trade-offs and add more craft, rather than having AI produce the final output.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Dylan

“we really thought about our AI efforts as lowering the floor, but raising the ceiling. One example of that is how do we generate a first draft for you?”

0.52

The amount of software created will keep rising exponentially and become more vertical, so the differentiator in the market becomes design and craft rather than the ability to produce software.

forecasthigh valuespeaker onlynovelty 2/4durability 3/4· Dylan

“My sense is that the amount of software that will be created will continue to go up exponentially. We've been on an exponential curve, but it's going to go more vertical. And so then what's the differentiator? It's design, it's craft.”

0.52

Watching users struggle in recorded research sessions motivates teams to solve problems, but the team does not viscerally believe those solutions will matter until the product is launched or announced in a big way that draws real reaction.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Dylan

“I think you don't know that viscerally until you get something out into the market or at least announce. If you make an announcement and it's big enough, then people will react to it.”

0.52

A CEO must own everything, but should check infrequently on areas where leaders are trusted and remain deeply involved in areas where the stakes are high, such as design, where losing design quality could lose the business.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Dylan

“In other areas, whether you have very high confidence in your leaders or not, the stakes are so high that you have to be always involved.”

0.52

Even as design becomes more accessible to non-designers, craft does not become more ubiquitous; the gap between a new designer and a deeply experienced one stays large, and thinking through complex multi-year trade-offs remains non-trivial.

causalhigh valuespeaker onlynovelty 2/4durability 3/4· Dylan

“even as design is made more accessible, it doesn't mean that craft is made more ubiquitous.”

0.52

Founders should find and ship the smallest increment of their idea that can be released in an inspiring form, because Field wishes Figma had found that smaller version sooner, though it is unknowable whether an earlier, more basic launch would have led to today's Figma.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Dylan

“what's the smallest increment of the idea that you could get out in an inspiring form. I definitely look back and wish that we had found that smaller bit sooner.”

0.49

AI is overhyped in the sense that it still operates inside an outer loop of human maneuvering and guidance; it predicts intent better but is far from fully doing so, and this imperfection will persist for a very long time.

causalhigh valuespeaker onlynovelty 2/4durability 2/4· Dylan

“It feels like there's still very much this outer loop that exists around AI. It's like it can do things for us much faster, and yet we have to help it maneuver, we have to give it guidance.”

0.49

The jump from o1-Preview to o1-Pro, likely through far more compute or a different training approach, produced the first time Dylan felt there was something resembling intelligence on the other end, because answers show analysis, thoughtfulness, and detection of intent without requiring mechanical prompting.

factualhigh valuespeaker onlynovelty 2/4durability 2/4· Dylan

“It's the first time I've felt like there's something that resembles intelligence on the other end.”

0.49

About one-third of people who come into Figma label themselves as designers, while the other two-thirds do not, and Field believes those non-designers will still get involved in the design process in various ways.

factualhigh valuespeaker onlynovelty 2/4durability 2/4· Dylan

“it's like one-third of the people that come into Figma label themselves as a designer, the other two-thirds do not.”

0.49

Design literacy is becoming a skill that everyone should have, and Field hopes that listing Figma on a resume will eventually be as unremarkable as listing Microsoft Word, which is currently a red flag to omit.

forecasthigh valuespeaker onlynovelty 2/4durability 2/4· Dylan

“I hope one day it'll be more like listing Microsoft Word on your resume. Right now, people put Figma on the resumes. If you put Microsoft Word in your resume, it's kind of like a red flag.”

0.48

Historically, society tends to suffer bad outcomes first and then build protective mechanisms in response to public outcry, rather than designing safeguards thoughtfully in advance, though humanity also tends to become desensitized to harmful technologies over time.

causalestablishednovelty 1/4durability 4/4· Patrick

“Seems like the historical lesson is really bad things will happen and then we'll backfill protective mechanisms.”

0.48

The obvious dark pattern for brain-computer interfaces is wireheading, in which users keep pressing a reward button, which is an old idea that has been discussed for a long time.

definitionestablishednovelty 1/4durability 4/4· Dylan

“the obvious dark pattern for BCIs, if you are like a true believer, is wireheading where you basically have the fentanyl button.”

0.46

Reasoning and LLMs are different things, though the market currently conflates them; reasoning models that think for extended periods could drive progress in math and science.

definitionhigh valuespeaker onlynovelty 2/4durability 2/4· Dylan

“Reasoning is a little bit different than LLMs. We're conflating them right now, but they seem to be two different things.”

0.46

Few products use reasoning at the product level, and those that try struggle to productize it, because reasoning branches in ways even less deterministic than an LLM, making it harder to craft a great experience.

causalhigh valuespeaker onlynovelty 2/4durability 2/4· Dylan

“I think it's harder to probably create a great experience around given just the nature of where that tree can branch out into”

0.45

AI capabilities are both underhyped and overhyped at once, but as of January 2025 the reasoning capabilities of o1-Pro mode still seem underhyped, because the depth of analysis and intent detection in its answers is a large step beyond o1-Preview.

normativehigh valuespeaker onlynovelty 2/4durability 1/4· Dylan

“I think that it's both like underhyped and overhyped at the same time. Somehow I feel like the reasoning capabilities in like o1-Pro mode right now.”

0.45

Figma found substantial low-hanging fruit for AI in simpler features, such as improving FigJam diagrams, auto-naming layers, generating first drafts, and rewriting slide text to match tone, and these should be pursued before attempting reasoning-engine products.

normativehigh valuespeaker onlynovelty 1/4durability 2/4· Dylan

“there's so much low-hanging fruit that we can explore and that's super obvious before we get to fancy stuff like reasoning engines.”

0.42

WebGPU is valuable but represents an extension of what the GPU can already do in the browser, rather than a shift from no GPU access to GPU access as WebGL was.

definitionestablishednovelty 2/4durability 3/4· Dylan

“Whereas WebGPU is awesome, will create all sorts of benefits. But I'd say it's more of an extension of what you can do with the GPU”

0.39

Design systems solve consistency at scale as more people design a product, and the central challenge is balancing standardization with creative room, since these two objectives are in tension but both must be held.

normativeestablishednovelty 1/4durability 3/4· Dylan

“And how do you both help people standardize while also giving them the room to innovate and be creative? And those two things are somewhat in tension sometimes”

0.34

Brain-computer interfaces are the ultimate interface because intent itself becomes the input, so that a user creates something directly from thought; this is the grail of interaction design.

forecastspeaker onlynovelty 2/4durability 3/4· Dylan

“BCIs are kind of like the ultimate interface, and if you can get to the point where your intent is an input and then you're creating something, it's kind of the grail.”

0.31

Figma's next chapter should make design bigger than pixel pushing, so that design is considered in every part of building a company and teams have tools to be intentional about it.

normativespeaker onlynovelty 2/4durability 2/4· Dylan

“make design bigger than design. How do you take it so that design is not thought of just as pixel pushing but also is something that people are really considering and doing at all parts of the process of building a company?”

0.31

Just as AI lets a non-cook create recipes from stated intent, context, and constraints, non-designers with a point of view can be empowered by design tools to create and make great work, which may expand the market.

forecastspeaker onlynovelty 2/4durability 2/4· Patrick

“Is that the right way to think about it? That like, maybe Figma might make someone like me expand the market by making me a designer?”

0.31

Non-designers will eventually realize that getting what is in their head at full fidelity is not enough, and they will want to improve their ideas, which leads them to appreciate professional designers and begin their own journey toward becoming designers.

forecastspeaker onlynovelty 2/4durability 2/4· Dylan

“people will realize that it's not enough just to get out what's in your head at full fidelity, but you'll want to improve what's in your head too.”

0.31

People starting companies today on average think more long-term, focused on what happens over the next six years rather than the next six months.

factualspeaker onlynovelty 2/4durability 2/4· Dylan

“It's less about what happens over the next six months and more about what happens over the next six years. That'd be my macro observation.”

0.31

Figma's founding vision of eliminating the gap between imagination and reality was too abstract for customers, so the company made it concrete by focusing on design accessibility, and now believes it is time to make it more abstract again.

factualspeaker onlynovelty 2/4durability 2/4· Dylan

“And the feedback we got was this is so abstract. What the heck are you guys talking about? At some point, we made it more concrete, and now it feels like it's time to make it a bit more abstract again.”

0.31

Predictions from essays written in the early 1990s, and even 30 years earlier, about technology have proven accurate, which Dylan takes as a reason people feel bolder about long-term technology bets.

factualcontestednovelty 1/4durability 2/4· Dylan

“It's wild how some of these essays from like early '90s got it right on the money.”

0.30

The most enduring motivation for Field is helping users achieve their goals, including both constructive complaints that drive improvements and examples of what users accomplished with Figma, and he loves interacting with customers.

factualspeaker onlynovelty 1/4durability 3/4· Dylan

“anytime that we can help a user achieve their goals, anytime I get user feedback, whether it's constructive”

0.30

The most serious setbacks in Figma's history trace back to people problems, such as hiring incorrectly or failing to set the standard, which forces a painful reset; hiring for self-awareness helps recover from such mistakes.

causalspeaker onlynovelty 1/4durability 3/4· Dylan

“it all comes back to people when you're talking about that. And I think when you maybe don't hire correctly or don't set the standard the right way and you have to reset a bit”

0.28

Designers are not worried about AI replacing them, because design integrates culture, user behavior, alignment, craft, and knowledge of user flows, and product managers cannot yet prompt AI well enough to specify it.

causalspeaker onlynovelty 2/4durability 2/4· Dylan

“as a designer I'm not worried about AI replacing my job because a product manager can't even prompt me the right way.”

0.28

VR is a fertile place to experiment with new interaction patterns because its early adopters are mostly gamers who can be taught new mechanics through games, so many experiments fail but some will prove as foundational as early computing interfaces.

forecastspeaker onlynovelty 2/4durability 2/4· Dylan

“when you have something like VR where you have a bunch of early adopters and you can teach them through game mechanics because it's mostly a gaming platform at this point”

0.28

Digital products can be built to anything that does not violate the laws of physics given enough time, and digital time-to-build is shrinking, which makes digital possibilities feel unlimited.

forecastspeaker onlynovelty 2/4durability 2/4· Dylan

“I've kind of always felt like the approach should be you can build anything that's not violating the laws of physics given enough time. And now the time portion of that even is decreasing on the digital side.”

0.27

The best way to get started with design tools is to make anything, including recreating something you already know to get comfortable, and with AI the best use is learning subjects you have wondered about with a personal tutor.

normativespeaker onlynovelty 1/4durability 3/4· Dylan

“If you don't have an original idea, literally go recreate something that you already know to get comfortable and familiar with the tools.”

0.26

Current AI agents do not have agency, even though the words agent and agency sound similar, so high-agency humans will keep delivering the best results.

factualspeaker onlynovelty 1/4durability 2/4· Dylan

“In the age of AI, yes, currently agents don't have agency, despite the words being very similar.”

0.26

Future designers will need to articulate a point of view, since tools can empower people with intent and context to create and refine work, and making great work requires understanding craft beyond the initial output.

normativespeaker onlynovelty 1/4durability 2/4· Dylan

“if you've got a point of view in the world and you're able to articulate it, how can we empower you with the tools”

0.25

Field expects humanity to harness the energy of a sun, perhaps a neighboring one, by 2100, and says setting it as a goal could bring the timeline sooner.

forecastspeaker onlynovelty 2/4durability 1/4· Dylan

“Dyson sphere by 2100. ... I think if you set that as the goal, maybe it becomes a lot sooner.”

0.21

Many founders start designing in Figma themselves before having a designer on the team, using the digital canvas to work out how their product should function.

factualspeaker onlynovelty 1/4durability 2/4· Dylan

“I hear so many founders that tell me they start in Figma and they're like in Figma design a lot of the time, they might not even have a designer on the team”

0.21

Figma has tried to map the whole software development process, from FigJam brainstorming and ideation, to Figma Slides for alignment, to Figma design, and to Dev Mode for translating designs into developer work, and there is much more to do to speed that loop.

factualspeaker onlynovelty 1/4durability 2/4· Dylan

“So we start with FigJam brainstorming, ideation. Then we go to Figma slides like alignment.”

0.20

Using o1-Pro, Dylan can anticipate the important topics to cover before a legal call, which gives him a shorter billable hour because he short-circuits preparation to the matters that count.

factualspeaker onlynovelty 1/4durability 1/4· Dylan

“with o1-Pro mode, I can often get a sense of what are the important things to cover in a legal call before I have the call. So I can have like a shorter billable hour.”

0.20

A friend's prompt asking o1-Pro to write a dinosaur story for a six-year-old using only the first ten letters of the alphabet produced a very good result after long reasoning, which Dylan found wildly difficult to do himself.

factualspeaker onlynovelty 1/4durability 1/4· Dylan Field

“Write a story about a dinosaur for a 6-year-old using only the first 10 letters of the alphabet.”

0.19

Dylan's competitive drive is strongest when he is on a team, where he enjoys trying to figure out how to win together, which he sees as making the team and its outcomes the source of motivation.

factualspeaker onlynovelty 0/4durability 2/4· Dylan

“I think it's fun to have that gamer mentality, especially when you're on a team.”

0.18

Within three years, a non-designer should be able to use Figma with AI to create what is in their head with near-complete fidelity, and Field would be surprised if it happened in one year.

forecastspeaker onlynovelty 1/4durability 1/4· Dylan

“I would hope within three years, and if we can do it within one year, then we should do another podcast on that.”

0.17

Dylan has been interested in design since childhood, when he drew notebook plans for the best bathroom, and found automatic sinks and foot-operated doors to be clever affordances.

factualspeaker onlynovelty 0/4durability 2/4· Dylan

“I have notebooks from when I was a little kid about how would you go make the best bathroom?”