00:35:11 · Dylan Field
[00:35:11] S6: Yeah. I mean, look, I think that it's different context for sure when you're in a sort of code base and you're using COD and it's overclaiming and making stuff up Mhmm. Or you're using, you know, OpenAI's ChatGPT models. 5.5 is amazing, but, like, definitely overcomplexifies things. And, you know, these these teams will deal with the sort of problems that are being identified, and they'll fix them and figure them out. But, overall, I think it's a different sort of class of problem than, like, when you're in a design file, and, really, part of your mode is exploration. If we can help you with that exploration, that can be useful for the user. And in terms of semantics, I mean, the way that you represent a a component or, you know, you the way you apply auto layout even, these are all examples of things that we've had to get more rigorous around how we build evals for this Mhmm. To make sure that we're doing things in a way that is both clear and opinionated, but also doesn't get in the way of the user. Mhmm. And I think the worthy year for us has been evals and probably for the industry. And so there's so much on the eval side that, you know, we have to always be improving on. And part of the feedback that we'll get from this beta will be towards where have we missed the mark and what we need to do better. And I think there's lots of room to grow there. I'll also say that I think that when you look at products like Weave, there's so much potential to not have necessarily this pile on effect, but rather to explore more with AI. The outputs of the models are like clay you can mold.
00:36:51 · S2, Dylan Field, John Coogan
[00:36:51] S2: Mhmm. [00:36:52] S6: And, you know, it's like, how do you take stuff? I mean, that might be a one shot slop that's generated, but then actually apply it across multiple models in a pipeline in order to actually figure out a way to make it into what you want. And sometimes the best way to do that is to get on a canvas and use your hands. Sometimes the best way to do that is through a workflow you define. Other times, it's by going back and forth with an agent. I think you have to be very smart about what you do when, and it's our responsibility as a tool creator to give you all the options. [00:37:24] S0: Mhmm. Jordy?
00:37:24 · John Coogan, Jordi Hays
[00:37:24] S0: Mhmm. Jordy? [00:37:25] S1: What are you excited about in terms of using AI to help augment the creative process. Because when I think of Part of what I think is exciting about this moment is like I don't Right now right now it's hard. Like a human still has to have like the fundamental kind of idea. Right? And and and you we know that because you have to like type it into a box. Right? You need to you need to do the typing still. So we we we're still we're still necessary there. But when I think about like iconic like startup brands, think of Linear. I think of some of what like Cursor has done over the last couple years. I think of like Figma's brand where there's so much work that went into just like coming up with that coming up with like the concept and and the look and feel of the brand. And then like way way way more work in terms of like instantiating that brand across every possible surface area. And part of what I think is exciting about this moment is like we're potentially entering into a world where you can spend more time just doing that like heavy creative work coming up with like a concept that can stand out. And then once you create a system, then you can like much more rapidly scale it across, again, all these different surface areas, whether it's like digital products, web, you know, email, ads, etcetera. How are you thinking about like augmenting and giving creative superpowers on that on that, like, brand building process when you're kind of coming up with coming up with, like, the idea?
00:39:05 · Dylan Field, John Coogan
[00:39:05] S6: Yeah. I think there's so much we can explore and do there, and using design systems effectively is hard. We sure certainly have not perfectly nailed that. I think we've done much more on the maintenance side, and we have a lot more coming on the execution side of how do you put the design system you've created together so you can fully productionize that. And it's one of the things that, you know, we are looking at from so many angles, and it is critical context to be able to figure out across code, across design, how do you use this effectively. But brand, I think, is, I mean, where on the Weave side, we see even more exploration with these workflows and people being able to define workflows that are more canonical and then branch out from them. And I think we'll see the same with assistant and with Figma's design agent as well. And the I I I also think that this idea bank, content bank can be established and added to over time and by more people. And what we're seeing with the last quarter results that we announced is that more people are starting to use Pigma in the organization. Thank you. [00:40:14] S0: They're fantastic results.
00:40:14 · John Coogan, Dylan Field, Jordi Hays
[00:40:14] S0: They're fantastic results. [00:40:15] S6: Thank you. But the breadth, I think, of the usage expanding is a huge part of that. And the more that you can get it to a point where people can come in with their ideas, contribute to the conversation on a canvas, collaborate with others to refine, doesn't mean that, you know, these, like, sort of nondesigners who don't know all the conversations that are being had about the UX or the brand are going to be the ones that are bringing the idea as the idea they're gonna ship to the table right away. But I do think that bringing more voices into that conversation, more viewpoints only helps. [00:40:53] S0: How do you think about the concept of Gell Mann Amnesia in, like, the in the context of, like, say staying sane in the age of AI? Like cause there's probably nondesigners who use the Gen AI tool and are like, oh, wow. Like, designers are cooked. And then there's real designers who use the Gen AI tool and say, oh, there's, you know, yes, it's maybe 99% of the way there. But the battle is to get to 99.999, maybe a 100% of what you can do. And it feels like [00:41:22] S1: Yeah. [00:41:23] S0: This is sane. [00:41:24] S1: Yeah. All outputs. Right? If you ask if you ask AI to generate you something around a topic you know really, really well Yeah. It's a different feeling than a topic that you're just learning about.
00:41:34 · Dylan Field, John Coogan, Immad Akhund
[00:41:34] S6: Yeah. Yeah. I mean, I think that basically, folks are in a place where they learn by comparative cases. And Sure. The more that they see great examples next to sort of output that's not as good and they can learn why, that's ultimately where you see folks then push further. And, you know, they can then take it on to to do more in area and evolve their skills and their judgment and their their sort of taste or not. But I I do think it's really important to have those conversational loops going, and also, get your arms around as a design leader, what is happening in your organization? Because what we're seeing is more people prototype things, and then, you know, there's, like, this latent fear of, oh, no. What are we gonna ship? And is it the thing that is sort of like a random exploration that someone thought was real, or is it something that is, like, gone through our cycle and is ready to go and is, like, very well thought out? And in general, the other thing that I'd mention is it feels like there's a lot of tunnel vision right now. Mhmm. You know, I I think that there's certain models that kind of went, oh, let's learn from four o And Mhmm. You know, are sort of sycophantic. And doing that in especially for engineers and get very latched onto an idea, and then it's kind of like the AI psychosis sets in, and you're like, this is the best idea ever, and you're fully tunnel vision towards creating it. And you can show massive progress, but, like, are you going the right direction? [00:43:12] S0: Yeah. [00:43:12] S5: And I