Dylan Field
speaker
223 appearances
2 recordings
2 series
first heard Jan 2025
last heard 20 May
Dylan Field’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 1 in all, peaking in May 2026 with 1.
Appearances
At the same time, I feel like it's overhyped to AI in general. 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. Yes, it's getting better at predicting intent, but it's not all the way there at all.
And so, yeah, it's this interesting period where I think people see where things could go if everything was perfect, but it's just not perfect. I don't expect it to be for a very long time. I think in the context of Figma, 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?
This isn't going to be a system that's going to generate the final output. But if we can make a first draft that then you can go off of, build off of, iterate on, explore the tree of possibilities with, then hopefully you can get to the point where you're able to deep dive on one of those possibilities, understand the trade-offs, and bring even more craft to your zine.
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. And it's really important to make sure that you bring that if you want to stand out in the market.
I saw a tweet recently and I wish I remember who posted it. I can't find it for the life of me, but sorry, whoever it was, I'm not attributing you. But it was something like, as a designer, I'm not worried about AI replacing my job because a product manager can't even prompt me the right way. Yeah. And I feel like that's super true.
Design is incorporating so many different aspects of culture, of user behavior, of alignment, of craft, of just knowledge about where people are in different flows and more. It's so much about like the almost debate of how it should work as well as the pixel perfection of what it looks like. It's something about creativity meeting problem solving in an interesting way.
I feel like maybe design becomes much more accessible to people, which I think is a good thing. But I mean, if you work with a designer who is very new to the world of design versus somebody who is just incredibly experienced, knows deeply how every system works, is able to keep incredible state in their head around a complex systems problem, I mean, the latter is rare. You will see the difference.
And I just feel like even as design is made more accessible, it doesn't mean that craft is made more ubiquitous. I suspect that there will be higher levels of craft that tools like Figma can help people reach.
But the thinking around what to bring to market, when, how, and how exactly it should work in navigating these really complex systems trade-offs that could unfold over years is just non-trivial. I think those all remain constants.
Reasoning is a little bit different than LMs. We're conflating them right now, but they seem to be two different things. And I think it's pretty exciting to imagine what progress in math and science can be like with the help of reasoning models that are able to think for extended periods of time on a task. And very exciting to think about how that could create new technologies for everyone.
In the Figma context, I wouldn't say that there's necessarily like new technical breakthroughs outside of the world of AI that I think are really changing a lot for us right now. That said, I think that markets and sort of preferences are ever shifting.
And I think people are getting a lot more sophisticated about the way they build applications, the way that they think about their internal systems. And that's exciting and stuff that we can think about as we work through our strategy.
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 in application. And we were not sure where the limits of WebGL would be. Originally, Evan spent a lot of time building something that was more cross-platform, honestly, and could target desktop as well as web.
We had to rip it all out later because WebGL worked out. But we even hedged our bets initially because we were believers, but it was like, oh man, there's some part of this that's not going to work out. And it turned out that you could do everything we dreamed of on the web and do it in a really production-ready way.
Whereas web GPUs, awesome, will create all sorts of benefits, but I'd say it's more of an extension of what you can do with a GPU than going from not being able to use the GPU to using the GPU.
It's maybe counterintuitive at this point, but despite my excitement about Reason Engines, I think there's so much low-hanging fruit to dive into reasoning to make our product better with AI. And so we started off with the most basic of things, like let's go make diagrams better in FigJam.
And we went into, okay, how do we get it so that you can do stuff like name your layers automatically or create a first draft of a design? Or can we help you rewrite text in Figma slides? to match tone better and create an interface to explore that. Without talking more about where we're going next in our roadmap, I still think that 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. And I think that there's not many examples in the market as of today that are using reasoning at the product level yet. And I think the few that are trying are still struggling to productize it. doesn't mean it's not coming.
I think it's harder to probably create a great experience around given just the nature of where that tree can branch out into versus it's even less deterministic than like an LM. If you conceptualize it as like an LM thinking and talking to itself for a long time and ending up in interesting spots down the logic tree.
I actually think it could be even more interesting, honestly, to expose the tree of possibilities rather than try to hide away from somebody. I think a lot of reasoning is an LM basically traversing that tree itself and then figuring out what do I tell to the user. Personally, I think that a lot of the value in design is actually seeing the tree.
Showing 141–160 of 223 · page 8 of 12
← Previous
Next →