All right. Next, we're gonna talk about how products are getting designed in the age of AI. Dylan Field, CEO of Figma. Give him a big round of applause.
Thank you so much.
Good to see you.
Good to see you. Thank you for having us.
It's a pleasure to have you here. Firstly, congratulations on Figma Make.
Thank you.
It seems like you're one of those companies that fundamentally adopted AI in a way that was very expedient and very natural, and it didn't seem like it was an artificially, you know, jammed-in thing. So how'd you go about doing that? Talk to us about it.
Yeah, I mean, even before Figma Make, our point of view has just been we have to be able to add AI in ways that solves real problems. And I think it's so tempting to kinda, you know, sprinkle some AI fairy dust on stuff-
Yeah
say AI a lot, and that's just not gonna actually solve any user problems. So early on, that meant things like renaming layers, and users loved little features like that-
Mm.
Because it takes hours to rename all your layers if you're a designer.
Right.
If you're able to just do it with one command, then you save so much time. But then if you go bigger, Figma Make obviously is something that we're really excited about, the ability to go right from design to prototype to back to design. They really should be two sides of the same coin.
Yep.
As we look forward, bringing all the efficiency of experiences you can get in code to the designing canvas, we've talked about this publicly already, and it's something that we're really excited about what we can do there.
Do you think, like, the role of a designer and the PM starts to fuse a little bit also because of what you folks are doing?
Well-
Like, does prototyping get to be a core part of product design or, or product management?
Yeah, it's so interesting 'cause I think for, maybe 10 years or so, 2015 to 2025, I was, like, very vocally saying, "I think that all these role- these roles are gonna blur: engineer, design, PM, research, they're all gonna blur." And I was kind of extrapolating out from some of the early signs I was seeing, totally disconnected to anything AI. And then I think as models started to really get powerful and the scaling laws kicked in fully, I sort of revised my opinion just as everyone started to say the roles would blur. And now what I think is that the roles actually aren't blurring. The roles are sort of staying what they are, but the responsibilities are blurring, and I think AI is this pull to be more generalist in the first place.
It's not an excuse if I'm blocked as a designer by my PM or vice versa anymore, and it definitely won't be, you know, 6 months, 12 months from now. You shouldn't be blocked. You should just actually go solve the problem. That said, there are people in your org who have expertise and are spiky in certain areas, and that's why we have roles. But yeah, I think more people are going to prototype-
Mm.
More people are going to go design, and I also think more people will be hopefully talking with customers and, you know, sort of doing all of it.
Yeah. So okay, so I'm super excited to talk to you because if I were not doing what I'm doing right now, I'd be a designer.
I love it.
I am obsessed with design. But-
But roles, you know, responsibilities blurring.
They're blurred, so now I'm a designer.
Yeah, so you can still also be a designer.
Yep, my design team is gonna be really upset, but otherwise, it's great. So-
We're excited
... so I'm gonna talk to you about a range of issues, but let's start with the future of UI. If agents become the prominent worker bee, and if agents are gonna interact with other agents, and if we evolve these systems to be speaking human language rather than us learning machine language, does the fundamental first principle construct of design change, where you're not gonna build a bunch of UIs with browsing ability, where you're gonna go out and try to surface different things, but it's just gonna be you're gonna ask it something, maybe with voice, maybe with text, and it's just gonna give you an answer? And does that fundamentally change design in the future?
Yeah, I think there's two ways to approach that question. I guess first, I would just say that the current paradigm of how people are thinking about agents is limited. Right now, it's very much, I go kick off something, the agent goes and does a thing-
Yep
... and it comes back, and it's done. And I think that we will see that. We will see prompting through text-
And by the way, that will change, right? Because the duration for autonomous execution will just get longer and longer.
Absolutely.
It'll be-
There'll be long-running... You know, maybe you'll have an agent running for a year.
Right.
But also, I think what will happen is the prompt box, while it will persist for a while, I think we're gonna see an evolution there. I've been saying for a bit now, I believe that we're sort of in this MS-DOS era of AI. Brian Chesky apparently agrees. He said it recently too, and it's exciting that, like, this is catching on, and people are really starting to imagine what can we do if we think beyond the prompt box? Because this is just one way to navigate latent space, and ultimately, you're just trying to figure out a way to get to this hyperdimensional space and create the right activations. And so I think prompting can evolve. And I would just say for the long-running agents or short-running agents, auditability matters.
I mean, if you're somebody that is just blindly trusting an agent right now, then you're a very brave person. And I think as AI gets better, perhaps you build more trust. But I still think that you need, just like you need for humans, the ability to audit. And then how do you audit? Well, you need to be able to actually look at something that you can understand. Finally, I really believe that where we're headed is not towards a world where you just have agents working independently in a sandbox but rather alongside you. I mean, I'm coming from Figma, so we have an infinite canvas where humans can come together and work together in that canvas. We're very inspired by Google Docs, where you can edit multiplayer at the same time.
I think that where we're probably headed for software in general is more towards that multiplayer philosophy, where agents, humans work together alongside each other, as well as, you know, a human or an agent going off on a long-term task.
And so now, let's fast-forward—I mean, you're already there today, but let's say that up until now, really good creative thinking and design took time. You know, and it was like, we would... When I was at Box, Aaron and I would, like, literally iterate on designs for, you know, very, very long time to make sure that the screen was perfect. Now, you have this kind of notion of infinite design options, where you could just ask a prompt, and you'd get, you know, 50 different design options. How do you train for judgment at scale? And does that become the scarce commodity? Because it doesn't seem like getting renditions of different design options is the scarce commodity anymore.
Like, you will have unlocked that pretty substantially over the course of the next, even year, I would say.
Yeah, I think getting an output to a prompt is only a first step. You know, you might go and get a bunch of variations-
Yes
... but even the way you prompt that, the way you set that up with the first prompt, is actually quite challenging to do right. And then, actually thinking about the system that exists in, we're headed towards a world, we're in a world where you're existing across so many surfaces, so many modalities, and you have to think about how all that connects, stays consistent, solves a real problem, and most importantly, I think, has a point of view.
Mm-hmm.
And if you don't have that point of view, I think that in the increasingly competitive landscape of software, you might be in a place where actually, you're just not gonna make it. So I really think the design craft point of view is the differentiator. And what's exciting right now is that you can plant flags in the option space more easily. What I found personally was, you know, I've been trying to make sure that every part of Figma's aligned to models getting better.
Mm.
You know, it's obviously, as the models get better, we gotta all get better. And as models get better, Figma's gotta get better.
Do you see yourself writing models as well, or just using foundation models that might exist?
Well, we experiment with all sorts of stuff-
Yeah
... but the foundation models are increasing their capabilities at such speed that it's, I think, foolish to believe that you can outpace them.
Right.
The thing that's really important is that you're able to ride that wave and be able to add something to it. And what's been exciting for me is that, as these models have gotten so much better, whether it be in diffusion or it be in code generation, my sort of, like, hope all along was that we would get better, too, and I aligned it strategically the way that I think we needed to. Now, we're in a place where the models have gotten substantially better. They're getting better at a rapid clip, and I think that is even more true because as I look ahead, I see a world where we can go from design and in Figma, and, you know, in 6 months, 12 months, it's hard to say exact time frames because we don't know.
But you could get to a world where not only you're prototyping a new thing, but you're actually getting to a place where you can actually update existing code base with design, and designers can do so much more. And I think also, being able to react to designs that are opinionated is really important. Like, if you get AI slop generated from your prompt, it's just the case that, like, you don't have much of a reaction. It's like, yeah, it kinda works. If you have a really opinionated output, then you have a real reaction to that. And you might have a reaction of like, "I don't like this," or, "I really like this direction," but undoubtedly, you'll wanna change a bunch of stuff. So I think the outputs are sort of like clay, that then you have to shape and put through a process.
Mm.
And that's what we're doing with our a company we acquired called Weavy, which we're now calling Figma Weave. And this is something where you're able to then really take that process, manipulate media outputs and get to a final result that's outstanding. And that process is, it really, really helps create great assets, and it creates a workflow that can be repeatable.
Can I ask you a personal question?
Yeah.
Okay. I'm assuming, because you built Figma, you like design.
I'm a big fan of design.
Do you ever get really annoyed when people have poor taste? Like, how does that psychologically affect you? I'm asking for a friend.
Yeah. Look, I try to empathize. You know, there's always constraints you're not aware of, but yeah, there are, there are some decisions that are just bad. I will say that. And I think it's also something that, like, you can aspire to, to really have better taste and judgment around these areas. Like, everyone can build a skill.
Do you feel like in your-
I truly believe everyone can build a skill. I think it takes a very long time to build.
Do you think taste is teachable?
I think it's learnable.
Learnable.
I think it's something that everyone has their own journey on. Like, you get 10 designers in a room-
Right
... and you ask them what taste is, you get 50 responses.
Right.
You know, like, no one can agree on this stuff. But I think it's... In terms of, like, what is good software design, I mean, you have to be able to manage how do you make it simple, but also expose complexity in the right ways and still make it powerful? That's extremely important. You have to have a point of view, like we talked about, and I think it's really important to be connecting emotionally with the user in some way as well, and managing the experience and connecting with the brand as well. So I just think there's so many constraints. Some of them are cultural.
Right.
-some of them are more self-imposed or business realities, and navigating that solution space is actually really tough. You have to really think through all the system trees and, yes, sometimes I get annoyed.
You get annoyed. Okay, so like, is your ultimate fantasy simulation of the future where someone with really poor taste can actually build really good designed software? Like, do you feel like that's a possibility with what Figma will be able to do in the future?
Well, I think like let's, let's not mix up taste and design too much because I think they are slightly different. I think there's just like category errors you can make when you're doing design.
Yep.
It would be awesome if Figma could give you better feedback when that happens. And we can be able to help you on that journey and not repeat the mistakes others have already had. I think that when it comes to taste, that's a self journey in some ways, but also organizations have taste. You know, you're not gonna, schizophrenically change everything about the way that you present-
Right
... as a company every five minutes. Users would hate that. And so it's really, really important that you're consistent, but still creative. So I think right now we're in this world where, I mean, there is, sort of like the Flash GeoCities era of the web, and the pendulum swung back to skeuomorphism and then flat design and really Swiss. And, and now we're swinging back towards, I think, this very creative place where people are gonna be exploring, because what's the scarcest resource right now? It's attention.
Mm.
You have to stand out. And so I think we're about to see this like very creative explosion of ways folks express through software and through the web and through applications, and I'm really excited for that. It will lead to some things that are just not good ideas.
Right
... or don't work, but that's necessary. You have to be able to have a lot of those things happen too if you want to get the great results.
So like, in your quest of building this AI forward design system and design, you know, platform, what surprised you the most about dimensions on human-machine interaction patterns, generating emotional appeal in the age of AI for the user? Like, what's generally surprised you and that you did not expect, and what are the things that were far easier to get done than you thought would be the case?
Well, I think when you work in a way the models naturally are good at, that leads to just the best outcomes in general. Don't try to fight the models, in other words,
Don't try to fight the models?
Yeah. I mean, if you try to take the models and make them good at something they're not good at, that's generally a bad idea.
Mm.
If you try to lean into where they're good, you will have more success. I think that's a lesson that we intuitively knew, but, you know, like many in the room, have learned many times. What's maybe not surprised me, but has delighted me, is as you make it so that people can be more efficient, like they just love it. It's more of a state of flow that you can be in. So for Figma Make, the ability to go and really just prototype any idea rapidly, or even just, you know, doing things like taking model functionality, bringing it into the canvas, such as, adding images in, something basic like that, or text generation or, you know, copy tweaks.
Like even the small things can keep you in a state of flow rather than going from place to place and hopping and kind of getting distracted. And as I think about it, sort of in the long term, it's like: how do you extend that state of flow? And the more we can do that while also bringing that human element to the software and that emotional connection, the better. I mean, I look back to the pandemic and what made FigJam stand out, our whiteboard brainstorming tool.
Mm.
You know, yes, it was getting to market at the right time and creating a great product, but it was also the little things. It was like emoji reactions, stamps, the fact that you could wave your cursor around and it would create something that you could like high five another user with. And in a world where people were remote, that really mattered-
Mm-hmm
... to create the psychological safety to have a good generative brainstorm. And I think in general, divergence is undercounted right now. It's really important to diverge and to think about the option space in general, and in a world where LLMs are so good at code, otherwise, you might just be on this linear path, and that leads to actually less outcomes. And I think you don't get to as good of results.
And so we've got, you know, CIOs, CSOs, there's product leaders. If, if you were to start to think about metrics for success that could actually demonstrate hard ROI with, with use of AI and Figma, like, what would those be? Like, are we gonna be able to... Are you compressing the cycle time of building software? Are you making software better so that it actually gets more used? Like, what, what are your true north metrics that you look for of what the customer needs to achieve?
Yeah, I think that there's, you know, going from design to development, that time is going to shrink a lot. And I also think that it's, like we talked about, very hard to quantify craft, design, taste. These are inherently non-verifiable aspects. And I think what's most important maybe is not the quantitative metric, but the qualitative. If I was gonna like, give everyone advice in the room, is you can move a lot faster but ship worse stuff. Everyone knows this. And so what's really important in this moment, I think, is to understand for your unique culture, who are the people that actually decide when a product is shipped? And how does that decision get made?
If you're not solving for both business constraints as well as quality, I think that you're going to end up in a world where you're actually at a disadvantage. And so it's really important to actually be prioritizing quality and making sure the designer has the ability to really be part of that gate on how things get shipped. Because some cultures are already there, but some are in a place where actually that's not a gate on shipping, and I think those are the cultures that are gonna have a harder time.
Hey, you know, I would feel bad if I didn't do this, but we have two designers on my team, Travis and Jason Cyr, who you've spoken to, both of them-
They're awesome
... that absolutely love Figma, love you. And so I just wanted to make sure I gave them a shout-out, and let them know that we're gonna be partnering very closely with Figma as we actually build great software in the world, because I think the world is full of badly designed products, and we need to make sure that we get some great ones that move people emotionally. Thank you for what you do.
Thank you.
Yeah. We're-
Thank you for your partnership.
... we're lucky to have you, and thank you for making the trip.
Thank you.
Thanks. Take care.
Thank you.
Take care. All right. The next guest, the guest is someone that I have known extremely closely for many, many years. I worked for him for a while. He is, he's a freak of nature, Aaron Levie.