Designing With AI With Designers of Figma & Codex

1. 0:00 – 1:33

   Intro

   1. AGAakash Gupta0:00

      So there's this debate that I'm seeing with the new wave of designers. What is your guys' take on this design in code versus design in the canvas debate?

   2. GSGui Seiz0:08

      And the Overton window has shifted so quickly from my teams being AI curious to, like, banging down the door, they can't move fast enough without it.

   3. EBEd Bayes0:16

      If developers have been accelerated, say, like 10X, designers have maybe been accelerated like 1.5 or 2X, so design can become the bottleneck.

   4. AGAakash Gupta0:24

      This is Ed Bayes, head of design for Codex at OpenAI, and this is Gui Seiz, head of design for AI at Figma. Where is the line now? These designer and engineer roles are clearly merging.

   5. GSGui Seiz0:37

      The empowerment that has come from AI has meant that this idea of roles and, you know, rigid org structures is, is kind of slight starting to diffuse a little bit.

   6. EBEd Bayes0:45

      Yeah, I would agree with a lot of that. I can then copy a link to this component, and then I can just paste it in here and say, "Update my code with change I made here." I also think it's, like, a really fun time to be a designer because your imagination really is the only upper limit.

   7. GSGui Seiz1:01

      Now I don't need to. Now I can just ask a question and be like, "Can you build this?" And if it starts to do something, I can ask-

   8. AGAakash Gupta1:11

      Before we go any further, do me a favor and check that you are subscribed on YouTube and following on Apple and Spotify podcasts. And if you want to get access to amazing AI tools, check out my bundle, where if you become an annual subscriber to my newsletter, you get a full year free of the paid plans of Mobbin, Arise, Relay app, Dovetail, Linear, Magic Patterns, DeepSky, Reforge Build,

2. 1:33 – 6:00

   Code vs canvas debate

   1. AGAakash Gupta1:33

      Descript, and Speechify. So be sure to check that out at bundle.aakashg.com. And now into today's episode. When should you start in code and when should you start in canvas? What does the step-by-step roadmap look like for designing in the new AI design workflow? And how should PMs be working with designers and engineers in 2026? They share everything. There's a million people talking about design with AI, but few sit at the frontier like Ed Bayes and Gui Seiz. Ed is the head of design for Codex at OpenAI, and Gui Seiz is the head of design for AI at Figma. These are the two companies defining the future of AI design, and today they're answering how you should design with AI, how you should move from Codex to Figma and back again, how you should start in code and when you should start in canvas. They're giving you the roadmap so that as you are designing any product, you can design it with the latest and greatest tools. This is never-before-seen access, and they rarely do podcasts. You're getting a once in a lifetime opportunity to learn design from the best. Gui, Ed, welcome to the podcast.

   2. GSGui Seiz2:38

      Thank you for having me.

   3. AGAakash Gupta2:39

      So there's this debate that I'm seeing with the new wave of designers. One prefers the flexibility of designing with code, using Codex to actually lead the design process and maybe bring Figma in later. The other still prefers the freeform of the canvas, the more traditional design process. What is your guys' take on this design in code versus design in the canvas debate?

   4. GSGui Seiz3:03

      Yeah, I, uh, I think it's such an interesting time right now where finally we don't really have to choose. Um, and so the, this, the, the, the two camps to me are, are really interesting because I've often found the, the canvas to be a place where I can just quickly riff on ideas and do a lot of lateral exploration. Um, and I, I still feel that's the gold standard for that. Whereas code has just enabled me to go super deep and really make my ideas real very quickly. Um, and before I used to have to choose, like which kind of mode do I have to engage in to get to the thing that I want to do, and now I kind of don't. And I think the, the new kind of shift that we have to embrace, that we should embrace is, like, navigating seamlessly between both, which is why I'm really excited to be here and talking, you know, with Ed, uh, about how we enable designers and PMs and teams just to move fluidly between AI coding and agents like Codex, um, to, to kind of move between exploration and convergence. And so, yeah.

   5. EBEd Bayes4:03

      Yeah, it's a good question, and I think it's a little bit of a false dichotomy in a way because as a designer, as a creative, you have a toolkit, and there's the right tool for the job. So, you know, if you can code, um, and you want to test out some interactions, hopping into a code base, you know, is maybe the easier option, right? You can just even create a branch off main, and you can kind of play with a prod version of what you're jamming around with, and you can test stuff out, and it can all be live. But, you know, just like there are still architects out there who still work, you know, by pencil, like th- there's a sliding scale of, like, how you represent ideas. And I think less than kind of, you know, getting into this kind of like deeper, like philosophical debate around, you know, w- what is a designer today or what should they use? I think I just come at it like really practically, and I think most of the team here do that as well, which is, you know, if you're being super exploratory, you want to sketch stuff out. If you want multiplayer, you want to be able to collaborate and comment and show work and print it out and put it up on a wall and, you know, tear it down. Um, but you know, the role of a designer is pretty expansive, and if we're shipping a product and we want to get the last, you know, mile of the product up to this, you know, the bar that the kind of designer aspires to, then it just opens up the opportunity for designers to hop in and ship PRs and do that side of stuff. So yeah, I think, I think less of it in some grand philosophical take and more around like what's the right tool for the job. And if I was a young designer today, I would be like so excited to hop in and like have these insane new kind of like box of toys, frankly, to just like build all this fun stuff. So I also think it's like a really, really fun time to be a designer because your imagination really is the only upper limit. And now you have this like really broad range of flexible tools that can, that can help you bring your idea to life.

   6. AGAakash Gupta5:52

      So I'm hearing from both of you that it's a false dichotomy. Really, it sounds like designers need to be able to use both tools effectively

3. 6:00 – 10:06

   When to use which

   1. AGAakash Gupta6:00

      with each other. What are the tactical ways one should think about this? When do you start in code versus start in a freeform canvas? How do you go back and forth? How do I think about when to use which?

   2. GSGui Seiz6:11

      It's a really interesting question because it's kind of grounded on this like dogmatic approach that we've had to, to the workflow historically because of tool constraints. We've had to like-Start in low fidelity and work our way up into high fidelity because it's expensive to do high fidelity. And so we, we've created this very standardized, uh, way of working that started the, this, that started from that, from like paper doodles and low, low-fi prototypes all the way to code. And now, like our low fidelity start can just be jumping into Codex and, and, and doing a functional wireframe. Like the, the, the point is, can I start working from something that gets the team talking and gets us thinking about the, the shape of the material, the shape of the solution, and then from then I can start bringing that into Figma and working into like, okay, well my, my wireframe no longer needs to be a doodle that I made with a marker pen. Like it can just start being something that's functional and gives me a lot more dimensionality of, to, to the problem. And so it really just depends on the level and the scale and the type of problem. If it's like you're working on how to finesse a certain component, maybe you want to start straight in code. If you're trying to figure out how you might lay out a whole flow, it might help you visualize the whole flow in front of you and try to understand where it breaks, uh, versus building into a flow. If you're trying to go wide and really try to break a paradigm of interaction, uh, you might wanna just do a lot looser fidelity rather than try to wrangle a model to get to what you're trying to do. So it's, it really just now, like I said, you get to pick whatever tool helps you get further with your team, and I think that's, uh, yeah, it's, it's a really interesting and challenging time for teams that have been, uh, reliant on a specific type of workflow that now you kind of get to invent whatever workflow you want because we're making it easy for you to, to, to extend that workflow, uh, however you want it.

   3. EBEd Bayes7:57

      Yeah. I think just to add, you know, I think it feels like we're going through this kind of like rapid pace of change where like weekly there's a new thing, right? There's some very cool and new things from Figma out this week, which, which we're gonna chat about. But like the underlying design process hasn't changed as well, right? You're solving problems, you're representing things in different ways, right? You're, you know, rallying teams, and you're like building a vision of a future, and you're kind of bringing people into that future. So it really depends on, you know, what your objective is when you're designing, uh, might be. Maybe you're, you know, designing a hero image to rally people around it, then you probably want it pix perfect. So maybe you wanna, you know, you wanna stop in, you know, step into Figma or inspect stuff out there. Or maybe you're getting really crisp on design systems and you're thinking about how different components interact with each other and how they might look in different states. Um, the, again, maybe you wanna be in Figma and you wanna have all of your tokens, or you wanna be able to do all the color tests and like the really kind of like in the weeds detail, kind of, you know, draftsmanship type stuff. Um, but then say res- you know, you're doing something that's more responsive, so you're like testing a product on like how does it look on mobile web or like across different break points or like you wanna really test like, you know, a bunch of different, um, different kind of like approaches to something or some really interact- in- interesting like interaction design paradigm. And there it's like, well, you kind of wanna get into the messiness of code and be able to kind of, you know, play around with it, shape-shift things. Um, so, uh, yeah, it's funny. It, it, it's, it's not as simple as kind of like, okay, let's start in f- in say Figma and then we end up in code. It is this often back and forth and I, you know, I see this on the team, you know, when we bring people in, they're like, "You know, so like what's the current state of play for the product?" And it's like, "Where's the Figma?" And it's like, "Okay, well, it's really well specified here at this early stage, and then there's other bit you probably just wanna like bring up the product and look there." So, um, yeah, it's not as simple as like start in one place, end up in another. You kind of like, you know, weave in and out, um, and it becomes way more fluid, which is why I think a lot of these, these cool new integrations that we're gonna chat about today, um, are really amazing because they... Then the handoff between these tools is a, is a little softer and it becomes like way more fluid and just like easier to work across these different,

4. 10:06 – 13:01

   Live demo begins

   1. EBEd Bayes10:06

      uh, materials.

   2. AGAakash Gupta10:06

      I think that's where a lot of people's heads are gonna go is can't you have a lot of lossiness as you translate between these tools? How do I do this seamlessly? So instead of talking about it conceptually, can you show us, Ed, how you actually practically day-to-day are bouncing between these tools, what this design workflow actually looks like?

   3. EBEd Bayes10:24

      Yeah, totally. Cool. So I've opened up, um, Co- Codex, which is our desktop app, right? We have... You might have seen there's a terminal product, there's a, you know, ID extension, um, but this is our s- kind of standalone app. And the way that it works is you basically just open up, um, you add a new project. So you can open up your local, local folder. It could be anywhere. It could be desktop, could be downloads, could be like a, you know, like a repo that you al- already have open, and you can kind of just get jamming. So I've opened this thing, it's called Codex Composer System. So this is a good example actually about what we're talking about. So I've already specced out a composer in, in Figma, but I wanna c- test out some interactions, and this is where I think like working in code works really well, right? So if anyone's used our product, we have this kind of dynamic, um, composer system. So, uh, if the model is asking you for permission to like run a command, for example, um, it'll pop up this like permissions, uh, kind of special tr- composer treatment like this. Or if you're in plan mode, it might ask you some questions and you can kind of toggle through and go back, and it's this like dynamic system. So I built this out in code because I wanted to test these interactions, right, how the buttons morphed into each other, how the size works, like, you know, where are the different, you know, tap targets and things. But say I wanna like go deep on a particular part of this and I wanna maybe go, "Okay, I've sketched this out. Now I just wanna get really pixel perfect. I wanna make sure the icons are my icons. I wanna change maybe some of the labels," right? I can then basically bring up Codex and I'll use this fun thing that I use a lot, which is this pop-out window, and I can just drag it over the top and I can say, um, "Okay, I want to import my homepage and a few composer states." What should we do? Let's do kind of like composer permissions and plan, um, into a new Figma file. Um-In my, you know, default account. And then I'll just type in @ and I'll tag in Figma here. You actually don't need to type in Figma. It will just, um, you know, i- if you, if you say Figma, it'll probably just trigger it yourself, but this is another entry point. We also have this, like, plus menu here, and you can install a plugin, um, which you install from the app itself. Uh, and then you get going, and, you know, because it's getting to know the file system first, it might take a, you know, a few moments just to, just to get going and to read everything. But maybe as it's doing that, I can just talk through what's going on here, right? So first it's getting to know

5. 13:01 – 16:43

   Figma MCP deep dive

   1. EBEd Bayes13:01

      what's going on. So I just have like a React app, and then it's gonna dig into the Figma MCP, which has a few really cool things which, uh, you know, Gui's probably better, better placed to, to speak to, to than I am. Um, but there's some really cool functionalities that the team, team has used, which is, one is it will go into file and it will basically kind of pop open a window, which it should happen, should happen any, any, any minute now. And this vertical interaction happens and it basically like takes a, takes a kind of snapshot of what's going on, and then it will take it and then it'll... You just kind of click open in Figma and it will open a Figma file and it will kind of basically copy anything that you want. So you can do interactions, you can do, you know, whole, um, you know, whole websites and then it, it kind of specs it out, um, in kind of full flow. Might just-

   2. GSGui Seiz13:47

      The other thing that you can do, uh, especially given your prototype, is you're able to also select specific components. So, so in, in that example, for example, you only had the text box and you probably don't need the big white rectangle behind it if you're wanting to work on that. So it also allows you the ability to go down your, uh, kind of nodes and select the one you actually want, um, and just iterate on that. Um, so we wanna give you as much flexibility as possible to just work with the data that you want.

   3. EBEd Bayes14:13

      Cool. So as I say, it's just popped open and it's pretty cool. It's like... So yeah, um, to your point, right, you can go in, you can choose to select an element. Um, there's a whole bunch of other fun stuff that you can do here. You can do the entire screen. Um, and then I just click open file, and it will just open up a new file and it will have that composer state in there. And I think that's the thing that has been very cool for me as a designer, okay, it's doing a few different screens now. So in the background. Yeah, so it's pasted in these files here, and as you can see, it's pasted in the screens that I mentioned. And the very cool thing is as you click in, everything is like responsive. So all of the exact pixel perfect things, um, right, the padding, the, the border radius, it's exactly one to one with the file. Whereas previously, you know, engineers might have to go in and they have to like look at the separate numbers and like, you know, pull out the different like, oh, what's the, you know, what's the, the shadow of this? And like, what are these like X, Y values? Like it just does that out of the box, which is I think very cool. And then you as a designer can go in and like, maybe I wanna spec this out, maybe I wanna play with color, right? And I can in here, you know, test out a few different, um, you know, shades of, shades of green or like, you know, change some colors. Um, and then the very cool thing, so like, you know, this is just a small change, but let's say I end up changing the modeling. We've recently like released another model. Um, I can then copy, um, a link to this, uh, component, um, copy link selection, and then I can just paste it in here and say, you know, update my code with the change I made here. And now this will go the opposite way. So if I bring up Codex again and I just run, um, run localhost so I can get it running. Um, okay, sorry. Demos, you know, something always... But yeah, you make some local changes and then you can just paste that link into, back into Codex, and then it will make that change in, in your code. So yeah, this handoff is kind of seamless. So like maybe you're a designer who can code and you can kind of weave between these, but maybe you're not, but this doesn't block you right? Now you, you and your engineer can kind of collaborate in a more seamless way, um, and it no longer becomes these like siloed areas. It's like

6. 16:43 – 18:10

   Ads

   1. EBEd Bayes16:43

      way more fluid and the handoff is like much simpler.

   2. AGAakash Gupta16:45

      Here's the dirty secret about prototyping. You spend two weeks building a prototype. You validate your assumptions. Engineering loves the direction. Then what happens? You throw the whole thing away. Bolt changes this completely. When you prototype in Bolt, you're not building throwaway mockup. You're building real front-end code that integrates with your existing design system. So when you hand it to engineering, they don't throw it away. They ship on top of what you've built. I use Bolt every single day. I host my LAN PM job cohort on it, and honestly, I'm up till 2:00 AM some days just vibing in the tool, having fun, and building. That's when you know a product is good, when you're using it past midnight, not because you need to, but because you want to. Check out Bolt at bolt.new/aakash. That's B-O-L-T.N-E-W/A-A-K-A-S-H. Link in the show notes. Today's episode is brought to you by Amplitude. Replays of mobile user engagement are critical to building better products and experiences. But many session replay tools don't capture the full picture. Some tools take screenshots every second, leading to choppy replays and high storage costs from enormous capture sizes. Others use wireframes, but key moments go missing, creating gaps in your understanding. Neither approach gives you a truly mobile experience. Amplitude does things differently. Their mobile replays capture the full experience, every tap, every scroll, and every gesture with no lag and no performance hit. It's the most accurate way to understand mobile behavior. See the full story with Amplitude.

7. 18:10 – 23:59

   Lossiness and fidelity limits

   1. GSGui Seiz18:10

      And on the lustiness thing, it's really interesting because like it only gets better from here. It's already like very good, but as you know, OpenAI releases new models, the better AI gets, the better all of this gets. And so we've just created the pipeline for, for this to work, and it can only get better. And so that to me is also a really interesting thing because it, it feels like we're already at such a point where this is so much easier to do than have to... For example, likeUh, you know, in the past I've inherited files from, or, or surfaces from other people that have since left the company. And I'm trying to figure out where is the Figma file [chuckles] for this thing. Some of that, those decisions got made in engineering. They didn't even get recorded in the Figma file. They got all added in to solve some edge cases, and I just wanna, like, recreate all that so I can work on top of it. And just being able to capture all of that in the highest fidelity possible, um, is, is incredible. And I think now with some of the, the new releases that we've made with, um, the Use Figma tool, um, we're able to do that in relation to your actual design system. And so when you copy, right now, um, the stuff that Ed copied, because it's just a snapshot of a website, it doesn't yet bring in your, uh, design tokens. With this new tool, you're now able to actually reference your design library and have that pull in those design tokens and local styles. So you're not just working on a, a facsimile of the thing. You're working on a representation of the thing that is using your style guide. And I think that just unlocks a whole new level of, uh, fidelity where you can start having these, uh, aligned sources of truth. Uh, I saw an engineer actually not that long ago on Twitter showing their process of aligning their storybook to their GitHub to Figma with, uh, you know, with these AI kind of like running the k- the kinda loop and finding the, the dif- the, the diffs between them. And so now you don't have this kind of overhead of maintenance of like, "Oh, but we haven't updated that component in here, and it only exists in there." You can just run a loop and have anyone, depending on i- irrespective of their, uh, preferred modality, whether it's their engineers or their designers or their PMs, be able to, like, have the latest materials available to them. Um, and I think this is kind of where, why we're talking about it doesn't really matter whether you wanna start in code. [chuckles] Like, what do you wanna do? And the tools will be able to all, like, in- interoperate between them.

   2. EBEd Bayes20:30

      Totally. And, you know, while you were chatting, I made a very dummy change. I just changed the string there, changed it to our, to our newest model, um, and ran it and, you know, here it is, um, in the model picker here. So a dummy example, but yeah, this point of, like, you know, you can basically go deep. Um, you know, you can build out the component in, you know, in code, and then in parallel, your designers can go deep, and then you can just, you know, one click, um, kind of update the, the master from there.

   3. AGAakash Gupta20:57

      So practically, when you guys are using this day to day, how much do you have to update either the Figma or the code? How often does it get things wrong where things stand just technologically today?

   4. GSGui Seiz21:09

      There are limits to the kinds of things that it's able to infer, right? There's gonna be a bunch of stuff that just doesn't translate to Figma. Like, you know, maybe you have some shader effects, and we're not set up to run that kind of stuff. Maybe you have specific transitions that you're trying to dial in and, you know, a static canvas yet doesn't support for that. But the reality is you can get to a lot through even annotations. Well, a trick that we use a lot is actually being able to read annotations, uh, through, um, into the MCP. And so we're able to communicate a lot of this stuff back. Um, but it is that point of, uh, you know, the designer's job and that labor of love of being able to go in and, and do an absolutely, like, um, leveraged amount of work, uh, just from a design and straight into code, and then be able to, uh, also do all these things that you used to have to, like, manually spec out and stuff, just be able to do with natural language to say, "Hey, by the way, I want this to kind of feel like that." Uh, because, you know, it's hard sometimes for a model, uh, to infer kind of a lot of intention there. And so it often, to us, doesn't get a lot of the things wrong. Once you've set it up with, uh, your design systems and everything, uh, you tend to be okay. It's gonna be on the more kind of web specific kind of effects that, you know. And that's kind of like, uh, an opportunity there for us to kind of evolve some of the tooling to be able to, like, support you there.

   5. EBEd Bayes22:32

      Yeah. I think my mental model is kind of like working with a colleague as well. Um, to your point earlier about, you know, you bring a new colleague on, and you have to kind of bring them up to scratch. And if your, you know, maybe Figma files are a bit of a mess or, like, you know, things aren't named correctly, then it's a little bit harder for your colleague to get up to speed. That kind of stuff, I think, is, like, basically is the same for agents. So, you know, like, in terms of kind of like hot, you know, hot tips, I think things like, um, you know, making sure that you, like, name your components well and, like, really thinking through. There's a lot of great stuff, um, from the Figma MCP about, like, aligning your design system with, like, the CSS tokens in your, um, you know, in your code base. And, like, once you get those foundations, which are good for working with humans or agents, then I think it, like, makes things much simpler. But with our most recent Model 5.4 as well, I have had, like, a number of designers internally reach out to me, and they're like, "Has, like... Is it... This model is, like, way, just, like, way better at, like, working with these tools." And, like, obviously there are limitations, but I do think if people haven't tried 5.4, um, give it a go with the Figma MCP, because I think it's, like, materially better than, than things that, that we've, you know, released before. And I see more and more folks internally now using these tools in, like, a day-to-day way, which means I think that the reliability has improved and, and the interoperability and the kind of, you know, new MCP, um, features that, like, Figma have released, um,

8. 23:59 – 27:52

   Behind the scenes

   1. EBEd Bayes23:59

      uh, really starting to help.

   2. AGAakash Gupta24:00

      So you guys have been involved in shipping some pretty important AI features used by tons of people. I'm curious if we can get a behind-the-scenes of what it looked like when you or someone on your team was working on one of these features, what the back and forth looked like between code and the canvas.

   3. EBEd Bayes24:16

      Well, it's funny. I think, you know, on the Codex team, we're building products for developers, so the designers who work on the team are pretty developer forward already. So, you know, many of us code. Um, you know, I probably spend, um, around, I don't know, 70, 80% of my time coding, so, like, you know, the nature of the role has changed. So-There is less of this kind of handoff process on, on my team specifically, but, you know, we're also, we've been building these zero to one products and we've been moving extremely fast, so it's kind of all hands on deck and hopping in. From speaking with folks across the team working on, you know, other products or, like, products that have been around for longer or, you know, maybe folks who are, like, newer to working in code, what I've seen is a kind of, like, gradual onboarding and a gradual change. So, you know, we have some, like, work in progress, um, like Slack channels, uh, and design reviews, and I just see more and more people arriving with, like, a hybrid of, like, there'll be a link to a Figma, but there'll also be, like, a URL link, and people will just, like, hop in and we'll click through and we'll show interactions. Or, you know, the other day I was speaking to one of our, um, you know, content designers who typically, you know, works through all of the, like, UX copy and, and, and, and, like, text across our products. You know, and she was telling me that she's been shipping to prod and she's been, like, submitting PRs herself. So I think it, it, it, it, it really, um, varies, but I see a lot of people more and more, like, prototyping early stage using some hybrid of kind of AI tools in Figma, and then a lot of people at that last mile of, like, helping bring the polish up and just, you know, getting in and doing, like, very small, maybe it's a styling tw- tweak, maybe it's, like, changing a string. Um, but yeah, I mean, if, you know, even if you look at how design has changed on Codex over the past year, I would say it's, it's, it's kind of radically changed, and I think particularly with the app, right? We released it, I think, gosh, I don't know, maybe like six to eight weeks ago. Um, but we've been dogfooding it internally, and I think, like, something changed in December when we were dogfooding the app. The models hit, like, a certain capability threshold and just felt like anyone could hop in and change and, like, the, you know, the, the pace of PRs and development was, you know, was really, was really moving fast. And I think there was... I forget where I read it, but someone had an interesting quote recently, which is, you know, I think the challenge now for design is, like, choosing when to go slow if you're building, um, AI products. And, like, if developers have been accelerated, say, like, 10X, which I don't think is an over- overstatement, designers have maybe been accelerated, like, 1.5 or 2X. So, you know, design can become the bottleneck, um, if you're kind of not coding yourself. So then the trade-off is, like, when do you just kind of, you know, meet the speed of engineers and make sure that you, you know, keeping up with the velocity of shipping that's kind of required in this, like, crazy competitive market with when you choose to go slow and actually go through, you know, deep design process, um, you know, deep in Figma, deep in collaboration in credits to make sure that you get those fundamentals right so that you can have some velocity rather than speed, and you kind of can set a direction where you're going.

   4. GSGui Seiz27:24

      Yeah. It's, it's such a, an interesting moment. Even if you look back to even maybe just mid last year, I would've given you a completely different answer, and the Overton window has shifted so quickly from my teams being kind of, you know, AI curious to, like, banging down the door, they can't move fast enough without it. And it's, it's just this really kind of interesting kind of challenge because I have teams now that are working directly

9. 27:52 – 30:53

   Designers in staging

   1. GSGui Seiz27:52

      in staging, right? And they're designers and, you know, like, they, they have now given you superpowers that are able to do this. And to, to Ed's point, it's like, okay, well, come back a bit. Like, when are you doing the exploration? Like, you're, you're, you're able to do stuff and you're, like, almost, like, just so empowered by this that you have to think about, like, "Okay, this is cool that I can, but then what should I?" And I think that's really interesting. And so for a lot of our, um, work, even just, like, mid last year, we were still working from designs to prototypes. We did prototypes in Make. We would then kind of, uh, you work with engineers to kind of put that into production. Now, you know, we might, uh, like, put something, a, a, a, a flag into production, and then we kind of finesse some prototypes in Make and just try to get, like, little details right, and then we will, uh, bring that back out to, to, to GitHub and then use Codex to kind of, uh, work, um, into that. And so it, it's this kind of constant jumping around between the, the right tool for the right job. Um, that being said, it's... it-- the thing that surprised me the most is how kind of prevalent across Figma this became. Um, when I started, it was very much the, uh, you know, I'm in this, um, area called the, the build tools area, and so it has a lot of, like, very technical-leaning, uh, folk. Uh, whereas other teams, uh, you know, they work on kind of the editor problems and they work on, uh, trying to help you, give you better tools, more deterministic tools. Um, and so suddenly now I look around and I have people, uh, from, uh, monetization building these incredible, uh, you know, prototypes, uh, that, that are just so incredibly technically accomplished, uh, in a way that, like, just as a designer that in the past might never have, have given themselves permission to do so. And so what we're starting to see is a sprouting of interesting concepts because people now have not just the vocabulary to do it, but the, uh, ability to get something in front of people without having to task an engineering team to help them bring that to life. And so we, we see projects starting at all different stages. Like lots of new zero to one projects, lots of, like, revisiting old projects that used to be P2s, where we joked about, like, there's no more P2s. Like, now everything, [chuckles] everything can just, like, be shipped. That cutoff line is almost arbitrary because you can just, like... You have polish stuff that you wanna do, a designer can now just go and work on that polish stuff. Um, and that's become just really empowering for, for the team, where they have this kind of sense of self-determination that they're able, they're able to, uh, ship PRs and stuff. And I think, uh, as a designer, as a builder, as someone that, you know, values agency, I, I... it-- there's never been a more exciting time to be a designer.

   2. AGAakash Gupta30:37

      I think both of those points are massively illuminating. On the OpenAI side, people might say, "Okay, well, OpenAI, especially Codex, they would be operating at the frontier," so Ed spending 70, 80% of his time in code, maybe people would expect that, although I still think it's mind-blowing.

10. 30:53 – 33:13

    Ads

    1. AGAakash Gupta30:53

       And then today's podcast is brought to you by Pendo. Welcome to the SaaS plus AI agent era, where every product team, no matter the size, can build world-class experiences. Meet Pendo AI agents, virtual teammates that read your product data, chat with users, and take smart in-app action so you don't have to. With agent analytics, Pendo shows exactly how those agents drive adoption, complete tasks, and even cut churn.No extra engineering lift, fully SOC 2 and HIPAA ready. And because Pendo's behavioral insights sync straight back into your BI stack, you get AI-ready data to fine-tune models and prove ROI in one place. Smaller team, bigger ambitions. You no longer need an army to deliver software your customers love. Grab early access at pendo.com/aakash. That's P-E-N-D-O.com/A-A-K-A-S-H. Today's episode is brought to you by NayaOne. In tech buying, speed is survival. How fast you can get a product in front of customers decides if you will win. If it takes you nine months to buy one piece of tech, you're dead in the water. Right now, financial services are under pressure to get AI live. But in a regulated industry, the roadblocks are real. NayaOne changes that. Their air-gapped cloud-agnostic sandbox lets you find, test, and validate new AI tools much faster, from months to weeks, from stuck to ship. If you're ready to accelerate AI adoption, check out NayaOne at nayaone.com/aakash. That's N-A-Y-A-O-N-E.com/A-A-K-A-S-H. I hope you're enjoying today's episode. Are you interested in becoming an AI product manager, making hundreds of thousands of dollars more joining OpenAI and Anthropic? Then you might wanna do a course that I've taken myself, the AI PM certificate ran by OpenAI product leader Miqdad Jaffer. If you use my code and my link, you get a special discount on this course. It is a course that I highly recommend. We have done a lot of collaborations together on things like AI product strategy, so check out our newsletter articles if you want to see the quality of the type of thinking you'll get. One of my frequent collaborators, Pavel Hern, is the Build Labs leader, so you're gonna live build an AI product with Pavel's feedback if you take this AI PM certificate. So be sure to check that out. Be sure to use my code and my link

11. 33:13 – 39:15

    Roadmap for traditional teams

    1. AGAakash Gupta33:13

       in order to get a special discount. And now back into today's episode. Gui, you building upon that, that even at Figma we're seeing the new wave of designer, they are putting things in staging. They might even be shipping polish on their own. That is what the new wave of design looks like. But the funny thing is I talk to designers every day, PMs every day, engineers every day, and a lot of them are maybe at a more traditional company, and especially true in compliance or regulatory heavy industries. So I'm thinking specifically about our colleagues over in healthcare and financial services. You know, if you're working at like a UnitedHealth or a Ginkgo Bioworks where there's huge design teams, you don't, you aren't working in this way yet. You may not even have access to these tools yet. But if you're listening, what is the step-by-step roadmap to get to this new way of working? If you're kind of still stuck in the 2024, 2025 way of working at your guys' companies, how do you come into this new way of working?

    2. EBEd Bayes34:10

       Yeah, I'd just say give it a go. I mean, it's, it's kind of a, you know, it's a kind of, um, m- maybe a kind of simplistic statement, but, like, you know, I think to, to, to some of the examples that, that Gui was talking about, folks all the time kind of DM you questions about it and, and then, you know, I look at kind of what's their role and it's like you've got people in GTM, you've got people in finance, you've got people from all backgrounds, technical, non-technical, hopping in and trying these things out and it's kind of like once you start using them, you kind of realize how, um, you know, creative you can be. So, you know, just to give you a few examples, like someone on our GTM team like just built this whole iOS app that was like on her phone [chuckles] that she showed me at a, you know, at a work event and, you know, it... as... And she's like, "I don't know anything about iOS." [laughs] And then, you know, just this morning, someone from our comms team, like she wanted to design a seating plan for events, so she like designed a HTML file and you could like drag people around, you could like type in their name, and I think, you know, having chatted to folks across the org, it's, you know, it's like how did you, how did you learn to do this? And like, "Oh, I just downloaded it and gave it a go." So, you know, maybe you don't have access, um, in the company for whatever reason, but, you know, in the evenings and weekends you can still download these things yourself. Um, you know, try it in the terminal. If that's a little intimidating, like try, you know, um, try the app or, you know, if you've, if you've ever coded in like, you know, an, an IDE, like maybe try the extension. Um, but just get going and, and if, and if you're not sure what to do or, you know, it's, it's a kind of similar problem that we have with ChatGPT all the time, which is it's this kind of like blank box and you can do anything, and then the limitation becomes your imagination. So I would just say kind of like remove that limitation, and if at any point you get stuck, you know, just ask. Um, so like, you know, I'm going to Japan next week on my honeymoon, and my wife, uh, and I are like building this little app. So we're like mapping out our like, you know, the cafes we wanna go to and the restaurants, and my wife is like not technical, but she's hopping in and she's making stuff and like, you know, sending prototypes over. So I think just like really kind of like removing these ideas that like it's not for me and just like giving it a go, I think is like the best place you can start.

    3. GSGui Seiz36:19

       It's also worth pointing out that, you know, you don't have to start with code. You don't have to start with [laughs] like that has to be your thing, you know? Um, you know, we all have access to, you know, ChatGPT on our phones. You, you're able to ask it any question. I remember when I was trying to learn how to code the hard way before AI and doing all these bootcamp courses and trying out to like figure out how to keep up to pace with new frameworks, uh, in HTML and CSS and like what do I need to learn and like where do I actually invest my precious time into building a skill and a muscle and, and now, now I don't need to. Now I can just ask a question and be like, "Can you build this?" And if it starts to do something, I can ask, "How does that work? Help me figure this out." And so it's really up to me on how deep I wanna dive into. If I don't care about how it's built and I just want it to work, I can just keep asking it for new stuff. And so there's never been an easier way to get started because it's just like if you can ask a question, you can do the thing, and I think that to me is really interesting. And the, the question doesn't always have to be about code. Maybe you're asking about strategy. Maybe you're asking about how your team dynamics can be improved. Maybe you're asking about your org chart. May- you can ask-Anything, and that will build confidence to, like, ask more complex questions. And before long, you're asking questions about code. You're asking questions about, "Hey, the, I've just inherited this system. Can you help me explain the data architecture of this just so I understand where we're making calls to? Are there any systems here that are redundant?" Like, "Please look through all my code base." You can ask a bunch of stuff that before you'd have had to ask someone incredibly proficient, uh, in this to do, but you can gain a, like a, at least a surface level understanding of the problem. And it's just through the exposure of that, as Ed was saying, you're gonna build facility with this. Um, what happens at places like Figma, and I imagine OpenAI, is that you're constantly doing this as part of the thing, and so you build that facility very quickly. And I think there's gonna be moments where, you know, you- we're not asking designers in healthcare companies, "Go and start shipping," [laughs] you know, uh, updates to, you know, to main. Um, but at the same time, I think you can start interrogating the systems around you, and you can start finding where are opportunities for me to contribute or to think differently about something, uh, whether that's in code or not. Um, I think what's really interesting is that we've tried to do at Figma for a long time is try to make, like, design bigger than design, right? And, and say it's not just about designers, it's, like, the people that we interact with. There's PMs, there's researchers, there's writers, and how do we build tools for that? And I think the same is true with AI, and the way these two things are connected is all these people can ask interesting questions that can make interesting artifacts, uh, whether it's Figma, whether it's in Codex, whether it's through Figma into Codex and Codex into Figma. What's important is, like, you feel comfortable just asking a question, and I think that's a really interesting place to start.

    4. EBEd Bayes39:05

       Oh, I was gonna say, yeah, I mean, I think, I think that's a really great point. And one other, you know, maybe kind of thing I would, a kind of cherry I'd put on that cake is there's also,

12. 39:15 – 43:29

    AI as your tutor

    1. EBEd Bayes39:15

       I think, a, a great way of, of thinking about these tools as well is thinking of, of them as a tutor or a kind of partner. And, you know, even if you don't know how to code and you start coding, it- you shouldn't kind of like... I don't think the mental model should be I kind of like release control and like I don't need to know how to code, and I can just build this thing. I think it's, like, an, an amazing on-ramp, right? Like, you know, if you're a designer working in a company, you still work across the software development life cycle, and there are still a lot of, you know, really important parts of software engineering hygiene that you're gonna have to go through. You're gonna have to go through PR review. You're gonna have to understand the code base. You don't wanna introduce, you know, foot guns in the code. You wanna, like, respect the data model. And, you know, all of that actually requires knowing how to code. So it, it, it's a very empowering, um, new technology to give people an on-ramp. But, like, once you start going down this, um, this route, you know, you can also just start asking questions. You're like, "Okay, I've built this React app. Like, what is React? Like, what do these different pages do?" You know, "What is the difference between a layout page and a, and a normal page?" Like, "Okay, how does CSS interact?" And it's this way of kind of basically kind of getting over this hump, initial hump, which I think can be a bit scary for folks. Like, you know, you think, I've been to the same, you know, boot camps probably that, that Gui has and, like, you know, it was a lo- lot of time and, and investment to go to these things and, uh, l- and frankly, a lot, quite scary for a lot of people if your kind of, your background isn't in engineering. But now you can kind of have this on-ramp, and you can slowly ramp up because I do think, like, it's really important to understand the underlying systems that you're designing within, and none of this basically, like, obfuscates the need to do that. But it's a really great tool to basically bring more folks into the, into the fold.

    2. GSGui Seiz40:54

       I, I really do think that cu- curiosity is gonna be the defining skill of people that, like, succeed in, in this kind of new era because i- A, it's changing so fast, and you just need to have enough of it to, like, ask, "Okay, what does that mean for, like, what I'm doing?" And if you're already asking that, you're already, like, in a good place. And then to, to Ed's point, like, what does that mean, this thing that I've just done? Like, how can I find out a little bit more? And if you have that curiosity, you, uh, you're adaptable, and you can... You're open to, like, new information. You're open to trying out new things, and I think that's gonna be the thing. Things are changing so drastically that if you think it's a small shift that you can adapt slightly, you might miss, like, the actual bigger picture. And so to be asking these questions, to be trying these tools, to be, uh... It's hard to keep up, and there's a lot of, uh, sometimes I feel the anxiety of like, "Oh my God, there's too many things launching at the same time, and I still have to do my daytime job." But there is just this kind of mindset, "Let's ask questions, and let's see." And now I have this incredibly patient tutor, like as Ed said, that never gets tired, never clocks out, and I can just ping, "But I don't get it. But I don't get it. Just explain it to me in simpler terms." And I can get it to work with me at whatever, like, level that I need it to work in terms of abstraction. And so I think if you can just develop that curiosity instinct, uh, it's probably to your question, like, "How do I get started?" Just that. Just, like, see yourself as a curious person and start asking questions is what I would say.

    3. EBEd Bayes42:23

       I saw this funny, uh, meme on Twitter that I'd rather hire a new grad that has been using these tools to their full extent than somebody with a ton of experience who hasn't tried them just because their company hasn't procured them. And I hope that you guys all get that message, which is like even if you don't personally have your company paying for Figma, MCP, and Codex yet, if somehow they're not doing that and allowing you to do that yet, it's not a reason to not do it on your own. Ed gave the story of how him and his wife are building an app to go on their honeymoon. You can use personal use cases to explore some of these tools, and when your company finally gets on the bleeding edge, then you will be the first person to be able to take advantage of those tools. And one of the things that's, talking to you guys, is such a privilege is you guys are basically operating at the frontier. A lot of other companies are gonna be operating at the way you guys are operating in a year, in two years. And we've been talking about how designers might be in a staging environment. Designers might even be shipping polish to production, or Ed might even be shipping whole things to production

13. 43:29 – 45:39

    Roles blurring not disappearing

    1. EBEd Bayes43:29

       that he creates on his own. Where-

    2. AGAakash Gupta43:31

       ... is the line now? These designer and engineer roles are clearly merging, and then there's always been this third leg of the trio, which is a lot of the people watching this podcast, which is PMs. I would love to know not your guys' prediction of where these roles go, but actually just your description of what it looks like inside OpenAI and Figma today. How are those roles working together? Where are the lines drawn?

    3. GSGui Seiz43:52

       I mean, the, the, the boundary between these roles have been blurring for a, for a minute, right? Um, and I think AI's really accelerated that. And so for us, the way that we're, we're thinking about this is that there, there are people with spikes, right? But there is no, like, kind of territories. There's things that people gravitate naturally to do, uh, and that's now broadened. And so I have designers shipping code. I have PMs shipping, uh, stuff, uh, and prototyping stuff and, and working into designs. And so one of the things that we found really interesting is with this advent of things like skills, um, and that you're able to kind of guide models to, to do something, you have this almost like pedagogic nature of, of writing kind of, like how do I think about this thing? And we have this kind of collective wealth of, of information the way, like some of it we're using to train our, to, to add to our tools that, you know, that we've released, but some of it we're using internally. And so what you end up having is designers, uh, writing skills on how code, how to coach AI to design well. You have PMs, uh, working through skills on how to make interesting product decisions and all this kind of stuff, which means that it accelerates that blurring even more, right? Because now as a designer, I can make use of a, a PM's mindset, uh, to, to work through some decision-making, whereas a PM is trying to like figure out how to, uh, you know, debug something, right? And it creates this really interesting thing where I have a natural spike in design, not so much in engineering, but I now have the, the, I'm standing on the shoulders of giants of a bunch of engineers that have helped me kind of have guide, you know, Codex into a certain way of, of, of deploying or something. And so for us, we're now just becoming more, you know, in, in football, soccer in, uh, in Holland in the '70s,

14. 45:39 – 52:30

    Total football explained

    1. GSGui Seiz45:39

       there was Total football, where every position played every position. You didn't have a, a specific thing, and now we're in this thing. When someone moves over here, you can cover over here, and you have this thing where you don't, you don't know how to mark, and you're a much kind of, uh, you know, um, bigger threat as, as a team. And so for us, we're starting to move into that, where teams have a natural interest and spike in things, uh, but they're able to internavigate and help each other in problems where sometimes someone goes out for a day or they're sick or whatever, and that's now no longer a bottleneck. You can start making progress here and there. Um, it doesn't happen on every team yet, but we're starting to see the signs of this becoming a, a thing where you, the empowerment that has come from AI has meant that this idea of roles and, uh, you know, s- rigid org structures is, is kind of slightly starting to diffuse a little bit.

    2. EBEd Bayes46:28

       Yeah, I, I would, I would agree with a lot of that. I think that, so, you know, first, different roles can, can kind of, you know, blend a little, uh, a little more easily in terms of just like the day-to-day work that you're doing. But I think one, one counter perhaps I would, I would provide is, you know, I think the, the role of a designer doesn't go away. The role of a product manager doesn't go away. The role of an engineer doesn't go away because you actually play very different roles, right? Like, you know, if you're a product designer, you're, you know, arguably the voice of the user. You're trying to build products that people love and use, you know, same as product managers. If you're an engineer, you're perhaps, you know, more focused on systems and performance and, you know, designers care about these things as well. But, you know, e- even though the different tools that you use are more accessible or even if now, you know, a engineer can hand something off to Figma more easily, I don't think that, yeah, kind of impacts actually much of the day-to-day work that, that has happen, that, that kind of happens in terms of the, you know, the, the, the different value the different roles, um, prefer. And I would actually kind of look more to the past than the future on this point. So, you know, one way of thinking about it is we look to the future and these roles converge. Another is kind of you look to the past, and actually all of these roles have like very different roles within an organization and, you know, you bring each other together and maybe your toolbox is the same, but, you know, the role of product managers and, and product is, is building great products. And, you know, the mandate of an engineer isn't necessarily that. So y- you know, it, it is like great engineers are great, you know, great, great product folks as well. But, you know, I think bringing, um, bringing these different skills to bear, um, at a more kind of like conceptual level, I think like is, is even more important than the kind of low level of im- implementation side of things.

    3. AGAakash Gupta48:14

       So what I'm hearing from both of you is that we have always seen the roles converging, but you do see both of these three roles staying separate kind of conceptually or in spirit representing sort of different elements, where the engineers represent more of the quality system and engineering, the design represents the voice of the user and a great user experience, and the product manager, it sounds like, is bringing that business angle to make sure that these are all driving the business goals. Is that roughly the mental framework I should have, or how should I be thinking about, well, even if you can do some of the elements of somebody else's job, this is what a person stands for?

    4. GSGui Seiz48:51

       Yeah, I'd, I, I'd say yes. I'd say to the point about spiking is that you have, um, things that you naturally gravitate towards doing, and that kind of has, has in a way informed the trajectory that you took your career on. But there was moments there perhaps where you were arbitrarily siloed into that perspective because of the amount of, um, time it took to become proficient in a tool. So if I wanted to like learn a specific tool, um, that would kind of gate me in a specific thing unless I had so much time in the facility to also learn, let's say, code and design, and then somehow spend a lot of time thinking about like the intricacies of, of, um, you know, SQL for like, uh, querying, uh, you know, um, data sets, et cetera. And so you kind of ended up... funneling down one thing. And now that the tools are becoming easier and th- these abstraction layers are being built on it so that you can step into slightly other domains, you're able to do that. Doesn't mean I'm gonna stop becoming a designer and becoming a PM or become an engineer, but it means that I no longer am working with only, like, a third of the picture. And so what, what it means is I can start, like, working into product decisions and into shaping strategy roadmaps and s- uh, and having a higher seat at the table, but also working much more in the implementation thing if I want to. Um, but my role as a, as a designer now, I would self-identify as still, as Ed said, is still I wanna figure out, like, as the voice of the consumer, like, what is the kind of the ergonomics of a thing? What are the ways that I want someone to work through this problem? Um, and then, you know, I will put in enough of my energy into, you know, bringing some of that to life and also creating a narrative and a business narrative around it, but I have teams to help me do that. But for sometimes I'm working on my own zero to one projects, I'd say, and I'm now able to step into that and write a much more coherent executive briefing, let's say, uh, without having to have gone through the, the PM, the, you know, uh, indoctrination [chuckles] of, like, how do I, like, think about these mental frameworks, mental models, and all that kind of stuff. A lot of that can just be surfaced up for me to take advantage of, and I think that's kind of what I mean by people being able to kinda lean a little bit further into another one of the, the kind of the pillars, um, to, to ship products. And so yes, we've, we've been, we've been doing-- I don't think people are necessarily jumping ship from one end to the other 'cause they like what they like and they're good at what they're good, but they're s- now able to be better at more things.

    5. EBEd Bayes51:19

       Yeah, I think similar. You know, I, I can only really speak for myself, but, like, even though I'm spending most of my time coding, I would still self-identify as a designer. I d- and I don't think that has changed. It's just, you know, you're, you're still solving the same problems. You're still trying to, you know, uphold a, you know, high bar of craft, and that just means that you, you know, work in this different medium, and you can kinda collaborate more easily with folks. And same with PMs that I work with. It's like, you know, a lot of, lot of PMs, you know, still do write PRDs, but there's this great phrase that I've heard a little bit, which is, like, you know, prototypes, not PRDs. And, um, you know, many of the PMs on our team I think are good examples of this. And, like, they'll, you know, bring a working prot-prototype that they're thinking about or they'll, you know, ship a PR for a feature that they, that they were thinking through to, like, stress test it with some engineers before maybe someone else kind of, you know, takes it on and takes it further. Um, so I think it's about, like, how can, you know, how can code help you in your role, um, you know, better achieve the, the types of objectives that you're trying to do? And for product folks, like, that's a little bit different often actually than, than engineers, but the medium that you, that, that, that you use kind of, um, you know, c- can, might change.

15. 52:30 – 53:35

    Outro

    1. AGAakash Gupta52:30

       I could probably ask you guys another hour of questions just on this topic, but we're gonna leave it there. This has been a fascinating insight into how I think two of the most important companies shaping the future of product development, how they work in the inside. Gui, Ed, thank you so much.

    2. GSGui Seiz52:46

       Thank you. Thank you for having us.

    3. EBEd Bayes52:48

       Thanks a lot.

    4. AGAakash Gupta52:48

       All right, everyone, see you later. I hope you enjoyed that episode. If you could take a moment to double-check that you have followed on Apple and Spotify Podcasts, subscribed on YouTube, left a rating or review on Apple or Spotify, and commented on YouTube, all these things will help the algorithm distribute the show to more and more people. As we distribute the show to more people, we can grow the show, improve the quality of the content and the production to get you better insights to stay ahead in your career. Finally, do check out my bundle at bundle.aakashg.com to get access to nine AI products for an entire year for free. This includes Dovetail, Mobbin, Linear, Reforge Build, Descript, and many other amazing tools that will help you as an AI product manager or builder succeed. I'll see you in the next episode.

Episode duration: 53:35