The SaaS Apocalypse Is a Goldmine With Figma’s Matt Colyer

SPEAKER_02

The SaaS-pocalypse, or the next era of software, if you will, I'm really excited about it, and I think Figma and a lot of other SaaS businesses are too, because I've worked in developer tools for a long time, and maybe five, ten years ago, the estimate of number of developers worldwide was like 25 million, 30 million, 40 million, give or take. I think what's most exciting about this time is that I think it's going to be like a billion, maybe even more than that. I think there's this incredible time that we're moving through of product development and really the democratization of technology. I think the end result is that there is dramatically more software out there in the world. If you're in that space, it means it's a goldmine, right?

SPEAKER_02

Matt, welcome to the show. [SPEAKER_01] Thanks for having me, Dan.

SPEAKER_02

So for people who don't know you, you are the Director of Product Management for developers at Figma. And I want to start with, I think, the big question on everyone's mind. I should probably say, I bought a bunch of Figma, like probably two months ago, because there's this whole SaaS-pocalypse narrative. And what I want to get into detail with you, I think you have a lot of stuff to share about AI, AI and product management, all the stuff that you've been doing yourself. But also, I'd love to start with what is going to happen to SaaS tools in AI? And I think Figma is a really interesting example where there are all these people who are saying, oh, I don't have to use Figma anymore. There are people who just launched an agent in your product. You also have Figma MCP. So if you're transitioning from a world where there was no AI when Figma started to now you're a big scaled product, and now there is AI. Like, how does that work? And how are you thinking about, do we open the product up to agents? Do we build our own agent? What's working? What's not? All that kind of stuff. I'm really interested in that.

SPEAKER_02

[SPEAKER_01] Yeah, I'd love to talk about that today. I think for me, it comes from a couple different angles. I think the first thing is the SaaS apocalypse or the next era of software, if you will, maybe is more of a positive framing on that. I'm really excited about it because I've worked in developer tools for a long time. And maybe five, ten years ago, the estimate of the number of developers worldwide was 25 million, 30 million, 40 million, give or take. I think what's most exciting about this time is that I think it's going to be a billion, right? Maybe even more than that, right? And so I think there's this incredible time that we're moving through of product development, and really the democratization of technology. There's a lot of catchphrases around homegrown software. And we can talk about some of that stuff later. But I think the end result is that there is dramatically more software out there in the world. And so coming back to your point about the SaaS apocalypse, and what does that mean for companies that have an established product? If you're in that space, it means it's a goldmine, right? That there's all this opportunity out there. And I'm really excited about it. And I think Figma and a lot of other SaaS businesses are too. And so the other part, some of the negative sentiments of the discussions you see online is around, what if I could vibe code every app, right? And I think what's really interesting about this time is, for whatever reason, January of this year was the point at which it became the larger narrative. I've been doing this stuff for probably 18 months or two years. So I was already saying, let's go build everything. But I feel like the whole world caught up in January of this year. And people are. And I'm excited to see what happens. Because I know my own personal journey through that is, it's really fun to build the initial version of it, right? And I actually built one of my own agents two years ago. And the very first one was an email agent. And I had to look back at how it started. And it was literally this terrible Python script. And it was rickety. And sometimes the replies didn't work. And I was like, okay, but the larger narrative here is, software companies build more than just code, right? There's a reason that I pay for Gmail to operate my email. It turns out it stinks when the SMTP version needs upgraded. You're like, I don't care. I just want to receive email. And so as I've had to run my own agents for my personal life, I've experienced that pain of the product I want doesn't exist. I built it. And now I get the ongoing cost of it. And I'll be honest, I'm buying more software these days than I ever did before. Because I'm, you know what, that tool seems cool. I'm just going to pay somebody else to run my agent for me.

SPEAKER_02

I totally agree. As someone who has vibe coded my fair share of tools, A, yes, the personal maintenance, but B, I vibe coded tools that we have re-released into production. And let me tell you, it's not as simple as saying, fix this bug. And I do think that that's really missed in the SaaS apocalypse discourse. I got to say though, if one of the first things you did was an email agent, I'm super curious how you're doing your email right now, because I feel like things just got to a point where you can just do your email without doing your email. And I'm so excited about it.

SPEAKER_02

[SPEAKER_01] Yeah, I can tell a little bit more about the story. So the problem that started two years ago is I was using chatbots at work, because at that point that was the primary interface. Agent usage was not really a thing yet. And so my personal life, I have kids in three schools and if there are any parents out there listening, you know what it's like to get the PTO emails and what is the theme for today? And the feeling of a missed email. This is the worst parent feeling in the world. But if you miss the spirit day because your kid didn't do crazy hair day, you feel like you have failed at life. I will tell you that, and having done it more than once. And so I was like, I can't miss another one. And I was like, you know what I have, and it was because I had to track 15 emails a day. You think we produce a lot of email in corporate America, wait till you get to the PTO emails you get at school. And so I was like, I can't read all these, but who can? Agents. And I was like, why can't I just do this? And one of the major agent platforms out there. And I was like, the missing part here is I just want to hook it up to email. So the first version of it was just literally grab an email inbox, look for the top email and literally paste it to one of the LLMs and dump the response back. And I know you like to talk a lot about prompts and my favorite prompts in those days was basically forward the email and it would just extract the facts. And it was always shocking to me that I would send a multi-page email and get three bullet points back.

SPEAKER_02

That is, yeah, I remember those days. The wiring up and the copy and pasting. And I feel that's so far away, but it's only a year or two ago. That is, yeah, I remember those days. The wiring up and the copy and pasting. And I feel that's so far away, but it's only a year or two ago.

SPEAKER_01

Yeah. Well, and then to your point about agents. So I've added a memory system, right? Because to your point, haven't fully automated and not sure that I trust it to reply on my behalf, but having the memory system was a total unlock. And now how I've evolved it is I have a daily. So this is an interesting thing. Open Claw, I feel, hit on this, but the proactive part is what really set it on fire. And my version of that was I would have my agent take a summary of all that stuff and send me an email every day at a certain time. And the unlock for me was instead of having to go to a tool and ask for the thing, it would just show up now. Not that it was particularly smart, it would just do it at the same time every day. But I think where agents are going is much more proactive. And then thinking about do I need to reach out and contact my owner and let them know what's going on?

SPEAKER_01

Yeah. So if that was where you were a couple of years ago, what are the workflow things that you have now that you rely on that you're excited about? So it seems like there's some sort of brief functionality you're using in Open Claw, but what are a couple of things that you've been tinkering with that you like?

SPEAKER_01

I think one of the things I'm trying to figure out in my work life is around summarization. I think part of the job is understanding an immense amount of information and then figuring out how do I filter that information? And how do I imbue my agents with that skill of, this is the thing that matters and this doesn't. And then it's actually a really hard problem because there's a lot of stuff that you read and on the first pass, you're like, oh, that doesn't matter. And then it will matter three days later. And it's like, how do you describe which things matter and which things don't?

SPEAKER_01

It also feels like the agents are a little bit like, one of the things that I will have it do is go through all my meetings and all the meetings in the company. We recorded all in Notion. So I can have Codex just go through all the meetings and be like, here's all the stuff that you might be interested in, which is really cool. Because I can be in all these meetings that I'm not in. But then it feels like if it gives me stuff that is not quite right, and I tell it it's not quite right, it overcorrects. And it gives me all the things that I said I wanted, but way too much and way too literally. And it's just like, you're never, it's never quite right somehow in this weird way.

SPEAKER_01

Yeah. I was curious to see where you're at on that. Because I feel this is one of the unsolved problems at this point. It's like we're all grasping for it. Like you said, with the email and your inbox, have you fully automated it? Does it reply on your behalf or do you approve every reply? Or what does that look like?

SPEAKER_01

I have to approve every reply, but basically what I have is in Codex, I have a little app that I open in the Codex in-app browser that runs locally. And basically every day it sweeps through all my emails and gives me a list. Every email is on that page. And there's a draft that it has said, here's what I'm going to probably try to reply to. And because it has access to my computer and everything, if it's an email from my lawyers or whatever, it can go and search and be like, here's basically what I think I should say. And then I just scroll through and talk to it. We have this tool called Monologue. So I just monologue into it and just say, no, go fix this. Or yes, you can send that draft or no or whatever. And I've been at inbox zero for four weeks straight. And that is huge. I've never, it's never happened before. My assistant is like, what the fuck is going on here?

SPEAKER_01

[SPEAKER_02] I will fully admit I am part of the religion of inbox zero. So I've been running it for many years and I believe in it, but sure does take a lot of work. I'd be curious about the Monologue thing. [SPEAKER_02] I will fully admit I am part of the religion of inbox zero. So I've been running it for many years and I believe in it, but it sure does take a lot of work. I'd be curious about the Monologue thing.

SPEAKER_01

[SPEAKER_02] And then I just scroll through and talk to it. We have this tool called monologue. So I just monologue into it and just say, no, go fix this. Or yes, you can send that draft or no or whatever. And I've been at inbox zero for four weeks straight. And that is a huge thing. I've never, it's never happened before. My assistant is like, what the fuck is going on here? I will fully admit I am part of the religion of inbox zero. So I've been running it for many years and I believe in it, but it sure does take a lot of work. I'd be curious about the monologue thing. Do you actually talk to it or do you type to it?

SPEAKER_01

Yeah. Does it know video or is it only just the audio? It's only audio right now. Yeah. That is one of the hidden tips for most people. The audio unlock is huge. And one of the things that I've learned about it is it's weird to talk to your computer. And so my trick for that is I actually use Loom a lot because it feels less weird to pretend I'm screen sharing to somebody. And it allows me to actually talk through the problem. That's so funny. Huh? Like in the office? Yeah. I feel, well, I guess I do it from home mostly. So people don't hear me talking to myself, but even in the office, I feel like people aren't like, they'll just think you're on a Zoom or something.

SPEAKER_01

Yeah. Yeah. There's always, there was this barrier at some point. Now everyone in the office, I just assume they're not talking to me. I assume they're talking to their computer. So it's weird when they're talking to me. What a world we live in.

SPEAKER_01

What a world. But I think it became this social thing where you can see, are they looking at their computer? Are they looking up? Or there's also the whisper, they're just getting close to their computer and they're like, I want you to do this little thing. And you're like, okay, I know what's going on here. But it's twice as fast. I forget what it is. I don't know. It's roughly twice or three times as fast to talk. And I've got carpal tunnel and shit. So it doesn't aggravate my hands as much. It's much more ergonomic. So yeah, huge unlock. Use voice.

SPEAKER_01

I do want to get back though to the original thing that we were talking about, which is, let's just say, I think we're on the same page. SaaSpocalypse, not a thing. Actually making a piece of SaaS software that works all the time is a gigantic effort that only some people want to do and other people just want to pay for. But then let's dive more distinctly down into Figma land. So there are questions about in a design world, what do I want to just chat with my landing page and move things around that way? Or do I want to be on the infinite canvas? And I know internally, all of our designers, they're super AI pulled, super early adopters. And they're all like, yeah, typing is good for a first pass, but to actually get the details right, I need to be able to move stuff around. So in the design world, how does that change? How do you think that changes the product strategy when the possibilities for how you might design something have changed so radically?

SPEAKER_01

Yeah. I mean, I think it's a lot to unpack and I think we're in the early innings here and we're figuring it out. I think we're still in this hangover of the text box rules. I think so much of us are defaulting to chat as the experience for generative UI. And I feel like we're starting to enter the second chapter of that, of what does it mean? And that's part of the reason I'm so excited about our agents launched. We've had it internally for a while. And for those who haven't seen it yet, it's the ability to use an agent directly on the infinite canvas. And I think going back, it's funny about what's old is new again. There's a lot of this happening throughout LLM and ML land of we've reinvented evals. It's okay, well, we had unit tests before. And we've reinvented prompting and we had user input before. And it's okay. And design in the new AI era, the principles still matter, right. And so one of the core principles of design for me is the diamond. There's this idea of divergent thinking and then convergent thinking. And most design problems are like this. And this is the idea behind brainstorms. When people tell you don't ever shoot somebody's idea down, brainstorming is all about generating ideas. And one of the things that I don't think we fully unlocked yet from these new capabilities is the ability to supercharge generative thinking. I think oftentimes we get stuck in our own life stories and lived experience. And we approach a problem from a certain angle. And this is what's so valuable about having teammates, right? You go talk to your teammate and they have a totally different starting point. And the answer to this problem is different. And the creativity comes between the conversation between the two of you of, oh, I hadn't thought about it from that angle. That's interesting. Let me take it and build on top of that, right. And so going back to what does this mean in this new AI world? If we get outside of the text boxes, because I think text boxes are super limiting, and it's very much a linear, well, this and then that and then this. If we get to the canvas and you have the ability to have some of those same concepts, but the agents allow you to do divergent thinking. It's like, hey, I have this frame. Let me start here. It's like, hey, I think it should be grayscale. And then you have another frame. You're like, well, let me try SEPIA. And then it's like, oh, the SEPIA thing is cool. But the type is wrong. And it's like, oh, let me try that. Now it's like the accessibility is off. Let me duplicate the frame and try again. And so I think, and even that is still, I think early innings, that's very much the human driving the input, right. And you kind of have talked about proactive flow, right? And it's like, I think we're starting to figure out what if we had an agent that's like, here's a bunch of frames that I have on the canvas, your job is to push them, try different directions, or don't just double down. But then I think there's a separate set of agents that's like, okay, we have 25 frames on this canvas of concepts for a new marketing page, right? And then it's like, how do I channel them down, right? So then there's a convergent agent that's like, okay, these three are kind of like this, and these are clustered around this. And I think, and you can ask it for its opinion. You're like, pretend you're a customer clicking through this, which one makes the most sense, right? And so I don't think we've really tapped all of that stuff yet. And so I think, even the best agents, like the command line agents, don't have the ability to do those workflows. And so that's kind of where I see the future of design and product thinking.

SPEAKER_02

[SPEAKER_01] I think that makes total sense. And yeah, it seems like, from what I can tell so far, the agents are really good for, I already have a design system, I need a new landing page, get me the landing page in the design system I already have, which, to be honest, a lot of designers don't really want to have to spend time doing the nth landing page or the nth graphic for this post or whatever, which is more convergent and a little bit less divergent. What about the future of, you know, maybe Figma design tools, or just generally software with allowing external agents in versus building your own agent, or having both, which you all do have? How do you think that works?

SPEAKER_02

[SPEAKER_01] I think that makes total sense. And from what I can tell so far, the agents are really good for: I already have a design system, I need a new landing page, get me the landing page in the design system I already have, which, to be honest, a lot of designers don't really want to have to spend time doing the nth landing page or the nth graphic for this post or whatever, which is more convergent and a little bit less divergent. What about the future of Figma design tools, or just generally software with allowing external agents in versus building your own agent, or having both, which you all do have? How do you think that works?

SPEAKER_02

I mean, I think we embrace both, right? And I think design workflows are different than engineering workflows, but the lines are blurring. And so I think in the future, we're going to be all builders, right? And that comes from which perspective are you coming at the problem from? And so we definitely very much support third party agents today. And our answer for that is our MCP server, right? And so I think one of the nice things about MCP is it allows a standardized interface across all these different kinds of tools, right? And so we think about the problem in two directions. We think about it as code to design. So it's like, okay, in that scenario you just said, that's a pretty common thing. You're like, hey, I have a signup page, but it doesn't support GDPR, right? Most people are not going to be like, you know what I should do, I'm going to start with a greenfield page and re-imagine what our signup flow should be to add GDPR. Most people are trying to get their job done. It's like, on Monday morning, you're like, okay, I just got to get this thing done, right? And let me add the checkbox here, right? And so for that workflow, it's like, if you are comfortable in Codex or Cloud or Windsurf or Cursor, you pull up your code base, you fire up the MCP server, and you ask it like, hey, can you go to this page, fire up the dev server, go to this page and copy it into Figma Canvas, and it will actually do it. That's one of the releases we had earlier this year, which is a little bit mind blowing that agents can do this faithfully, but it turns out they can. And now you have just removed all that drudgery, right? And you've got it into a medium where you can actually interact with it. It's like, okay, let me go move things around very precisely with the direct manipulation tools that most people are comfortable with. And then the other kind of workflow that we think about is okay, now that we've got the design, take that design and bring it back to code, right? And so we've got a tool called Git Design Context, which takes a Figma design, wraps up all the different properties and components that you're using, as well as any other kind of guidelines you've provided in your design library and provides it to the agent. And again, it's kind of magical. It's like the agent will be like, okay, cool. Let me look at your current code base. I'll make a branch, create a PR, make the changes. And then you can ask the agent, be like, okay, take me a screenshot and put on the PR. And then your job is, kind of what you're talking about your workflow earlier with email, it's like, you don't merge it, but you're like, okay, well, I've got a good starting spot, right? And then you can come in there and riff.

SPEAKER_02

[SPEAKER_01] What have you learned about what makes for a good internal agent experience internal to a product that you may not have known before, before Figma Agent?

SPEAKER_02

That's an interesting question. What would make it a good product? I think specifically, this is very specific to Figma, but the context and personalization matters. In other products I've worked on in the past for AI companies, personalization is often the last thing. You get it just working for everybody first. But I think the thing that really differentiates an okay agent to one that people really love is the personalization aspect. We talked about memory as a form of it in these third-party chat agents. I think for Figma's version of that, it's the design system. If you have an assistant, but without the concept of how we structure our designs and how we put them together, the designs that it creates just aren't usable from that perspective. I don't know what your plans are for Figma and the proactive agent being a proactive agent. But I'm curious to the extent you have those plans and those experiments, how that's working. Obviously, we've talked about that being hard to get right.

SPEAKER_02

Yeah. I mean, I think that's where the future is going. If you look at how agents have evolved, I will say we've got a lot of things cooking internally. I don't know. I can't talk about specifics too much. I definitely think this is an area we want to head. Right. And so I think I can talk about the problems that we see today. If the amount of software is really exploding in the world, one of the bigger challenges then becomes how do you make sure that it's consistent with your values? Right. And so I think we become the bottleneck then, right. It's we only have so many human eyes to review all of this work. And so it's how do we provide a solution that allows people to make sure that they can continue to innovate at the speed that these agents create, but also maintains their values.

SPEAKER_02

[SPEAKER_01] What has been the transition like internally in Figma in terms of your own workflows in the engineering org, in the product org, in the design org from a pre-AI world to now? So I joined in January and I would say even in that period of time, it's been night and day. I think folks in January were experimenting with these new ways of working and across all the functions, right? I think probably engineering was leading the way as they do in most of these cases, but I'll give you an example on the product org. So we had an offsite and I think you might have actually come by. Yeah, weirdly enough, small world.

SPEAKER_02

[SPEAKER_01] What has been the transition like internally in Figma in terms of your own workflows in the engineering org, in the product org, in the design org from a pre-AI world to now? [SPEAKER_01] So I joined in January and I would say even in that period of time, it's been night and day. I think folks in January were experimenting with these new ways of working and across all the functions, right? I think probably engineering was leading the way as they do in most of these cases, but I'll give you an example on the product org. So we had an offsite and I think you might have actually come by. Yeah, weirdly enough, small world.

SPEAKER_02

I think one of my favorite memories of that offsite is that we have a product operations team and they had put together what we call PMOs. And to take a step back for a minute, one of the unlocks to me about AI is that you realize every problem becomes a context problem. And it's all then the work becomes about framing the problem with the right set of information. And so our product operations team, a lot of the work that we do is in the structured data as a PM. And it's why don't we aggregate that all together? And it's okay, let's get a copy of the org chart. We'll throw that in the SQLite table and put it in the file system. And then it's okay, why don't we create a connector to Asana and then we'll connect Slack and then we'll connect GitHub and a few other things. Right. And then the real insight was skills had really taken off at this point. And it's okay, the magic sauce here is one of the skills that I really like is this onboarding file creation. So when you add a new team member to your team, as a manager, you got to create a customized, okay, here's the channels you should know. Here's the people you should know. And it takes a lot of that knowledge that I would have previously said was entirely in my head. But then you start to look at how if you shape the context, right, that data was actually already there, right? I already had the org chart and it can walk the org chart and figure out who's the team and who's the trifecta on the product engineering design side for this new team. You just have to identify here's the new person, here's the team they're going to join. And it's okay. And then it can do a bunch of research. And then it goes into Slack and it figures out the channels. It's oh, this team is probably these three channels. And then it reads the last 30 days of content. And then it's here, and then it goes and checks the Asana board and finds all the projects. And so it comes back and you're like, oh, this is uncannily good. It's yeah, this is a pretty good starting spot.

SPEAKER_02

[SPEAKER_01] That's one of the things that I think is the thing that made CloudCode so good and is the thing that makes Codesx so good right now is everyone tried to start with agents that live in the cloud are always on, but then you have to manually connect them to everything. And CloudCode is just, yeah, no, it's just an agent on your computer has access to everything that you have access to. And then that just totally changes all the stuff it can do because it can get all the context it needs. And same thing with Codesx. I can just ask a random question to Codesx. We just published an article today and I was asking, who should I send this to? And it just went through my emails and my texts. And I didn't even know it had access to any of that. It just found five people that I probably would have forgotten about, but that I should have sent it to. And I did. And that's the magical thing that's starting to happen now. That was very, I think the technology, AI itself, if you gave it all the context would have been able to do this for a while, but it's only now that it's in the right harness and form factor, and it can do it a little bit more independently than it was able to before. Yeah. I'm going to put a plea out there. I remember WWDC. It was 24. I can't remember when it was. Apple Intelligence, whenever the headline was WWDC. Oh my God. I think that was more recent. I think it was 25.

SPEAKER_02

It was 25. Gosh, AI time, who knows what year it is. I upgraded my phone. I was all in. I upgraded my iPad. I was like I'm going to get, because I had just upgraded the last cycle and then they're like, well, you need a new processor. I was like, okay, I'm up for all my stuff. Solid WWDC. I was like, this is great. Because they had this concept of our phones have all of this personal data. And I was like, oh my gosh, this is going to be— WWDC. Oh my God. I think that was more recent. I think it was 25.

SPEAKER_02

It was 25. Gosh, AI time, who knows what year it is. I upgraded my phone. I was all in. I upgraded my iPad. I was like I'm going to get, because I had just upgraded the last cycle and then they're like, well, you need a new processor. I was like, okay, I'm up for all my stuff. Solid WWDC. I was like, this is great. Because they had this concept of our phones have all of this personal data. And I was like, oh my gosh, this is going to be WWDC. Oh my God. I think that was more recent. I think it was 25.

SPEAKER_02

It was 25. Gosh, again, AI time, who knows what year it is. I upgraded my phone. I was all in. I upgraded my iPad. I was going to get it because I had just upgraded the last cycle and then they're saying you need a new processor. I was okay, up to all my stuff. Solid WWDC. I was like, this is great because they had this concept of our phones have all of this personal data. And I was like, oh my gosh, this is going to be it. And then I will say, I'm really hoping WWDC this year is actually it because this past year and whatever has not been it. Right? Because, to your point, the technology has been there. The part that's missing is the tying it all together. And the mobile phone ecosystem has all that content. And so I'm waiting for the thing where it's the always on Siri that runs in the background and is actually smart rather than the one that's like, what was that? I didn't understand you, Matt. One day. Do you think they're going to be able to get that right? And if they don't, do you think it matters? I think it still matters because I think even them being late to the game, they are still the king of context. Right. And I think that's what's been interesting to watch about Google IO too this year. Seemingly Google has also woken up to that, that they don't maybe have as much data as Apple, but they have a lot. And it seems like they're now starting to marry their AI products. If you can keep all the names straight, I saw this great tweet that was like, is it Bard or Gemini Pro or Spark or whatever the thing is, but there's some. I think Spark is the product that is supposedly going to be the always on agent that is auto connected to all of your Google content. And so I'm waiting for the day that it runs my inbox for me and I get to inbox zero. Yeah. I think I just have this feeling about Apple that when OpenAI took off, everyone started buying Mac minis and you're just like, that's such a good business. They didn't have to even be in the AI race because they win by default because it's the thing that everyone runs the AI on. And even if they're behind on Apple Intelligence, which they are, and historically their software products have been lagging behind their hardware because their hardware is so good, it doesn't matter. They have a lot of time to catch up. Yeah. Well, and I think their strategy is very smart, right? It's the privacy play, right? And I think it is scary to upload all of your information to the cloud. And so I do think they're in the game and I really am hoping that they've got something interesting this year. So if we look back over the last year, there's been this big sea change in how we build stuff, how good the tools are. And then maybe correspondingly, how software works and how we build software and all that. What do you expect or what are you planning for over the next year as the capabilities increase, both in how you make stuff and what you make? I think the big thing this year will be about how do we review better? I think that's where the bottleneck is now, right? I think we have agents that are capable of producing all this stuff. They're available enough. They're cheap enough. And now we're just being inundated with all of this new content. And I think people are getting overloaded with what do I do with all of this? It's not just summaries of stuff that's been out for a while, but now it's, okay, this is net new content. Do you want me to go or not? And I think we have to solve the problem of how do we scale our value system of how do we evaluate that this new thing that we're creating is actually, and then feel confident and have enough trust in it that it can go to some degree in auto mode, if you will. Do you have any inkling about how that will work inside of Figma or what the interesting design considerations are for that kind of flow? And I think that's one of the problems that we're really focused on is talking to customers, understanding. I think a lot of customers and us are figuring out at the same time, right? I think the industry in general is trying to understand what is the new way? Is it a video walkthrough that's recorded, right? Or is it screenshots? Or is it another agent that's with a different prompt that reviews the work? And then you trust this agent so much that you approve its decisions. I don't know. It's hard to predict the future, especially at this time. Yeah. One last question for you. So I feel like there's been a lot of back and forth over the last year or two about, is there a future for PMs? Is there a future for designers? And if you want to be a PM, how do you break into the industry now? Because maybe there are fewer PM seats or engineers don't need PMs. How do you think about the career progression for a PM and how someone who's not senior becomes and gets to where you are? Yeah, that's an interesting one. I think the fundamentals still matter, right? I think the best analogy I've seen is there were math classes in school, but you still had a calculator, right? But at the same time, we all learned long division, right? And how to do that. And I remember integrating and taking derivatives and the rest of that. And do I do that on a daily basis now? Absolutely not. But I think it's incredibly important in order to drive these systems that you have an understanding of what these concepts are, be able to do it by hand. And so I think the foundations still matter. And I would be really curious actually at this point to see what CS classes look like, what is a one-on-one CS class, right? Because I think there are two parallel worlds. There's one where it's like, oh my gosh, you can just dump your answer

SPEAKER_02

and the rest of that. Do I do that on a daily basis now? Absolutely not. But I think it's incredibly important in order to drive these systems that you have an understanding of what these concepts are and be able to do it by hand. So I think the foundations still matter. And I would be really curious actually at this point to see what CS classes look like, what is a one-on-one CS class, right? Because I think there are two parallel worlds. There's one where it's, oh my gosh, you can just dump your answer or question into ChatGPT. It's literally, here's the bubble sort for you, right? Like, which of the 42 ways do you want it? And then there's a version of that where it's, I'm a really curious person. I wrote the bubble sort in C, but take it, put it in assembly and then explain it line by line to me and explain what a register is and what is L1 cache and L2 cache and the rest of it. And I think being a curious person, I guess, maybe this is the answer. I think the most important thing is to be a curious person, right? Because I think with these new tools, the people who cannot leverage them are the ones that just accept the output. And the people that are able to invent the next set of tools and really drive the tools to their maximum are the ones that are pushing the boundaries and understand how it's put together. And in order to be able to do that, you have to be the curious person. You can't have been the one who's, answer this problem for me, right, and just give me the answer. You have to be that curious person to be, how is this thing put together? And how does this actually work? And help you understand the next level, right? I agree. And it's so much more fun to live that way. Totally. I mean, it's catnip for me. I remember it's, I don't know if you're a Hitchhiker's Guide to the Galaxy person, but I feel like LLMs are the book. It is literally the manifestation of it. And you can even have them on your, I have this when I go on airplanes. I don't run a lot of local LLMs, but I'll download an 8B model and run it on an airplane. It's literally that. It's like the Hitchhiker's Guide to the Galaxy. You can ask it a question, like, why is the sky blue? And it breaks it down into the refraction and all the rest of it. And you're, well, what is a squirrel? And it will give you an answer of that. And so it's, and you know, they're not perfect. And some of them are a little weird, especially at the 8B size, but I don't know. It's a magical time to be alive for curious people. I'd say.

SPEAKER_02

I totally agree. Matt, it was a pleasure. Thanks.

SPEAKER_02

Oh my gosh, folks, you absolutely positively have to smash that like button and subscribe to AI and I. Why? Because this show is the epitome of awesomeness. It's like finding a treasure chest in your backyard, but instead of gold, it's filled with pure unadulterated knowledge bombs about ChatGPT. Every episode is a roller coaster of emotions, insights, and laughter that will leave you on the edge of your seat craving for more. It's not just a show. It's a journey into the future with Dan Shipper as the captain of the spaceship. So do yourself a favor, hit like, smash subscribe, and strap in for the ride of your life. And now without any further ado, let me just say, Dan, I'm absolutely hopelessly in love with you.

SPEAKER_02

software out there in the world. And so coming back to your point about the SaaS apocalypse, and what does that mean for companies that have an established product? Is like, if you're in that space, it means it's a goldmine, right? That there's all this opportunity out there. And that I'm really excited about it. And I think, you know, Figma and a lot of other SaaS businesses are too. And so I think the other part, some more of the negative sentiments of the discussions you see online is around, well, what if I could just vibe code every app, right? And I think what's really interesting about this time is, for whatever reason, January of this year was the point at which it became the larger narrative. I've been doing this stuff for probably 18 months or two years. So I was already, yeah, let's go build everything. But I feel like the whole world caught up in January of this year. And I was, yeah, let's go build everything. And people are. And I am excited to see what happens. Because I know my own personal journey through that is, it's really fun to build the initial version of it, right? And I actually built one of my own agents two years ago. And the very first one was an email agent. And I had to look back at how it started. And it was literally this terrible Python script. And it was rickety. And it sometimes the replies didn't work. And I was, okay, but the larger narrative here is software companies build more than just code, right? Like, there's a reason that I pay for Gmail to operate my email. It's that it turns out it stinks when the SMTP version needs upgraded. You're, I don't care. I just want to receive email. And so as I've had to run my own agents for my personal life, I've had to experience that pain of the product I want doesn't exist. I built it. And now I get the ongoing cost of it. And I'll be honest, I'm buying more software these days than I ever did before. Because I'm, you know what, that tool seems cool. I'm just going to pay somebody else to run my agent for me.

SPEAKER_02

I totally agree. As someone who has vibe coded my fair share of tools, A, yes, the personal maintenance, but B, I vibe coded tools that we have re-released into production. And let me tell you, it's not as simple as saying, fix this bug. And I do think that that's something that is really missed in the SaaS apocalypse discourse. I got to say though, if one of the first things you did was an email agent, I'm super curious how you're doing your email right now, because I feel like things just got to a point where you can just do your email without doing your email. And I'm so excited about it. Yeah. I can tell a little bit more about the story. So the problem that started two years ago as I was using chatbots at work, because at that point that was the primary interface. Agent usage was not really a thing yet. And so my personal life, I have kids in three schools and if there are any parents out there listening, you know what it's like to get the PTO emails and the, what is the, what's the theme for today? And you know, the feeling of a missed, this is the worst parent feeling in the world. But if you miss the spirit day because your kid didn't do crazy hair day, you feel like you have failed at life. I will tell you that. And having done it more than once. And so I was, I can't miss another one. And I was, you know what I have,

SPEAKER_02

my personal life. I have kids in three schools and if there are any parents out there listening, you know what it's like to get the PTO emails. What is the theme for today? And the feeling of a missed—this is like the worst parent feeling in the world. But if you miss the spirit day because your kid didn't do crazy hair day, you feel like you have failed at life. I will tell you that, and having done it more than once. And so I was like, I can't miss another one. And I was like, you know what I have? And it was because I had to track 15 emails a day. You think we produce a lot of email in corporate America, wait till you get to the PTO emails at school. And so I was like, I can't read all these, but who can? Agents. And I was like, why can't I just do this? And one of the major agent platforms out there. And I was like, the missing part here is I just want to hook it up to email. So the first version of it was just literally grab an email inbox, look for the top email and literally paste it to one of the LLMs and dump the response back. And I know you like to talk a lot about prompts. My favorite prompts in those days was basically forward the email, and it would just be extract the facts. And it was always shocking to me that I would send a multi-page email and get three bullet points back.

SPEAKER_02

That is. Yeah, I remember those days. The wiring up and the copy and pasting. And I feel like that's so far away, but it's only a year or two ago.

SPEAKER_02

Yeah. Well, to your point about agents. So then I've added a memory system, right? Because to your point of not having fully automated and not being sure that I trust it to reply on my behalf, but having the memory system was a total unlock. And now how I've evolved it is I have a daily. This is an interesting thing. Open claw, I feel like hit on this, but the proactive part is I think the thing that really set it on fire. And my version of that was I would have my agent take a summary of all that stuff and send me an email every day at a certain time. And the unlock for me was instead of having to go to a tool and ask for the thing, it would just show up now. Not that it was particularly smart. It would just do at the same time every day. But I think where agents are going, it's much more proactive. And then thinking about, do I need to reach out and contact my owner and let them know what's going on?

SPEAKER_02

[SPEAKER_01] Yeah. So if that was where you were a couple of years ago, what are the workflow things that you have now that you rely on that you're excited about? It seems like there's some sort of brief functionality you're using in open claw, but what are a couple of things that you've been tinkering with that you like?

SPEAKER_02

[SPEAKER_01] I think one of the things I'm trying to figure out in my work life is around summarization. I think part of the job is understanding an immense amount of information and then figuring out how do I filter that information? And how do I imbue my agents with that skill of this is the thing that matters and this doesn't. And it's actually a really hard problem because there's a lot of stuff that you read and on the first pass, you're like, oh, that doesn't matter. And then it will matter three days later. And it's like, how do you describe which things matter and which things don't?

SPEAKER_02

It also feels like the agents are a little bit like one of the things that I will have it do is go through all my meetings. All the meetings in the company we recorded are in Notion. So I can have Codex just go through all the meetings and be like, here's all the stuff that you might be interested in, which is really cool because I can be in all these meetings that I'm not in. But then it feels like if it gives me stuff that is not quite right and I tell it it's not quite right, it overcorrects. And it gives me all the things that I said I wanted, but way too much and way too literally. And it's just like, you're never quite right somehow in this weird way.

SPEAKER_02

Yeah. I was curious to see where you're at on that because I feel like this is one of the unsolved problems at this point. I think we're all grasping for, as you said, with the email and your inbox, have you fully automated it? Does it reply on your behalf or do you approve every reply? Or what does that look like?

SPEAKER_02

I have to approve every reply, but basically what I have is in Codex I have a little app that I open in the Codex in-app browser that runs locally. And basically every day it sweeps through all my emails and gives me a list. Every email is on that page. And there's a draft that it has said, here's what I'm going to probably try to reply to. And because it has access to my computer and everything, if it's an email from my lawyers or whatever, it can go and search and be like, here's basically what I think I should say. And then I just scroll through and talk to it. We have this tool called monologue. So I just monologue into it and just say, no, go fix this. Or yes, you can send that draft. Or no or whatever. And I've been at inbox zero for four weeks straight. And that is huge. I've never—it's never happened before. My assistant is like, what the fuck is going on here?

SPEAKER_02

I will fully admit I am part of the religion of inbox zero. I've been running it for many years and I believe in it, but it sure does take a lot of work. I'd be curious about the monologue thing. Do you actually talk to it or do you type to it? Yeah. Does it know video or is it only the audio? It's only audio right now. Yeah. That is one of the hidden tips for most people. The audio unlock is huge. And one of the things that I've learned about it is it's kind of weird to talk to your computer. And so my trick for that is like—

SPEAKER_02

I will fully admit I am part of the religion of inbox zero. So I've been running it for many years and I believe in it, but it sure does take a lot of work. I'd be curious about the monologue thing. Do you actually talk to it or do you type to it? Yeah. Does it know video or is it only just the audio? It's only audio right now. Yeah. That is one of, again, hidden tips for most people. The audio unlock is huge. And one of the things that I've learned about it is it's weird to talk to your computer. So my trick for that is I actually use Loom a lot because it feels less weird to pretend like I'm screen sharing to somebody. And it allows me to actually talk through the problem. That's so funny. Huh? Like in the office? Yeah. I feel, well, I guess I do it from home mostly. So people don't hear me talking to myself, but even in the office, I feel like people aren't like, they'll just think you're on a Zoom or something. Yeah. Yeah. There's always, there was this barrier at some point now, everyone in the office, I just assume they're not talking to me. I assume they're talking to their computer. So it's weird when they're talking to me. What a world we live in. What a world. But I think it became this social thing where you can see, are they looking at their computer? Are they looking up or there's also the whisper. You know, they're just like getting close to their computer. And they're like, I want you to do this little thing. And you're like, okay, I know what's going on here. But it's twice as fast. I forget what it is. I don't know. It's roughly twice or three times as fast to talk. And I've got carpal tunnel and such. So it doesn't aggravate my hands as much. It's much more ergonomic. So yeah, huge unlock, use voice. I do want to get back though to the original thing that we were talking about, which is, let's just say, I think we're on the same page. SaaSpocalypse, not a thing. Actually making a piece of SaaS software that works all the time is a gigantic effort that only some people want to do and other people just want to pay for. But then let's dive more distinctly down into Figma land. So there are questions about in a design world, what kind of experience do I want? Do I want to just chat with my landing page and move things around that way? Or do I want to be on the infinite canvas? And I know internally, pretty much all of our designers, they're super AI-pulled, super early adopters. And they're all like, yeah, typing is good for a first pass, but to actually get the details right, I need to be able to move stuff around. So in the design world, how does that change? How do you think that changes the product strategy when the possibilities for how you might design something have changed so radically? Yeah. I mean, I think it's a lot to unpack and I think we're in the early innings here and we're figuring it out. I think we're still in this hangover of the text box rules. I think so much of us are defaulting to chat as the experience for generative UI.

SPEAKER_02

[SPEAKER_01] And I feel like we're starting to enter the second chapter of that, of what does it mean? And that's part of the reason I'm so excited about our agents launch. We've had it internally for a while. And for those who haven't seen it yet, it's the ability to use an agent directly on the infinite canvas. And I think going back, it's funny about what's old is new again. There's a lot of this happening throughout LLM and ML land of we've reinvented evals. It's like, okay, well, we had unit tests before. And it's like, we've reinvented prompting and we'd like, we had user input before. And it's okay. And design in the new AI era is still like the principles still matter. Right. So one of the core principles of design for me is the diamond. There's this idea of divergent thinking and then convergent thinking. And most design problems are like this. And this is the idea behind brainstorms. When people tell you don't ever shoot somebody's idea down. Brainstorming is all about generating ideas. And one of the things that I don't think we fully unlocked yet from these new capabilities is the ability to supercharge generative thinking. I think oftentimes we get stuck in our own life stories and lived experience. And we approach a problem from a certain angle. And this is what's so valuable about having teammates, right? You go to talk to your teammate and they have a totally different starting point. And the answer to this problem is different. And the creativity comes between the conversation between the two of you of, Oh, I hadn't thought about it from that angle. That's interesting. Let me take it and then build on top of that. Right. So going back to what does this mean in this new AI world? If we get outside of the text boxes, because I think text boxes are super limiting, and it's very much a linear, well, this and then that, and then this, if we get to the canvas, and you have the ability to have some of those same concepts, but the agents allow you to do divergent thinking that it's like, Hey, I have this frame. Let me start here. It's like, Hey, I think it should be grayscale. And then you have another frame. You're like, well, let me try SEPIA. And then it's like, the SEPIA thing is cool. But the type is wrong. And it's like, Oh, let me try that. Now it's like the accessibility is off. Let me duplicate the frame and try again. And so I think, and even that is still I think early innings, that's very much the human driving the input, right? And you kind of have talked about proactive flow, right? And it's like, I think we're starting to figure out, what if we had an agent that's like, here's a bunch of frames that I are on the canvas, your job is to push them, try different directions, or don't just double down. But then I think there's a separate set of agents that it's like, Okay, we have 25 frames

SPEAKER_02

[SPEAKER_01] is off. Let me duplicate the frame and try again. And so I think, and even that is still [SPEAKER_01] early innings, that's very much like the human driving the input, right? And like, [SPEAKER_01] you've talked about proactive flow, right? And I think we're starting to figure out, what if we had an agent that's like, here's a bunch of frames that I have on the canvas, like your job is to push them, try different directions, or don't just double down. But then I think there's a separate set of agents that it's like, okay, we have 25 frames on this canvas of concepts for a new marketing page, right? And then it's like, how do I

SPEAKER_02

channel them down, right? So then there's a convergent agent that's like, okay, these three are kind of like this, and these are clustered around this. And I think, and you can ask it for its opinion, you're like, pretend you're a customer clicking through this, which one makes the most sense, right? And so I don't think we've really tapped all of that stuff yet. And so I think even the best agents, the command line agents, don't have the ability [SPEAKER_01] to do those workflows. And so that's kind of where I see the future of design and product thinking. [SPEAKER_01] I think that makes total sense. And yeah, it seems like, from what I can tell so far, the agents are

SPEAKER_02

[SPEAKER_01] really good for, I already have a design system, I need a new landing page, get me the landing [SPEAKER_01] page in the design system I already have kind of thing, which, to be honest, [SPEAKER_01] a lot of designers don't really want to have to spend time doing the nth landing page or the nth [SPEAKER_01] graphic for this post or whatever, which is more convergent, and a little bit less [SPEAKER_01] divergent. What about the future of, you know, maybe Figma design tools, or just [SPEAKER_01] generally software allowing external agents in versus building your own agent, or having [SPEAKER_01] both, which you all do have? How do you think that works?

SPEAKER_02

[SPEAKER_01] I mean, I think we embrace both, right? And I think design workflows are [SPEAKER_01] different than engineering workflows, but the lines are blurring. And so I think in the [SPEAKER_01] future, we're going to be all builders, right? And that comes down to which perspective are you [SPEAKER_01] coming at the problem from? And so we definitely very much support third party agents today. And [SPEAKER_01] our answer for that is our MCP server, right? And so I think one of the nice things about MCP is [SPEAKER_01] it allows a standardized interface across all these different kinds of tools, right? And so we

SPEAKER_02

[SPEAKER_01] think about the problem in two directions. We think about it as code to design. So it's like, [SPEAKER_01] okay, in that scenario you just said, that's a pretty common thing. You're like, [SPEAKER_01] hey, I have a signup page, but it doesn't support GDPR, right? Like, most people are not [SPEAKER_01] going to be like, you know what I should do, I'm going to start with a greenfield page and re-imagine [SPEAKER_01] what our signup flow should be to add GDPR. Most people are trying to get their job done. It's like,

SPEAKER_01

[SPEAKER_02] you're on Monday morning, you're like, okay, I just got to get this thing done, right? And [SPEAKER_02] let me add the checkbox here, right? And so for that workflow, it's like, if you are comfortable [SPEAKER_02] in Codex or Cloud or Windsurf or Cursor, you pull up your code base, you fire up the MCP server, [SPEAKER_02] and you ask it like, hey, can you go to this page, fire up the dev server, [SPEAKER_02] go to this page and copy it into Figma Canvas, and it will actually do it. Like, that's one of the [SPEAKER_02] releases we had earlier this year, which is mind blowing that agents can do this

SPEAKER_01

[SPEAKER_02] faithfully, but it turns out they can. And now you have just removed all that drudgery, right? [SPEAKER_02] And you've got it into a medium where you can actually interact with it. It's like, okay, let me move things around very precisely with the direct manipulation tools that most people [SPEAKER_02] are comfortable with. And then the other workflow that we think about is, okay, [SPEAKER_02] now that we've got the design, take that design and bring it back to code, right? And so we've got [SPEAKER_02] a tool called Git Design Context, which takes a Figma design, wraps up all the different properties and

SPEAKER_01

[SPEAKER_02] components that you're using, as well as any other kind of guidelines you've provided in your

SPEAKER_02

design library and provides it to the agent. And again, it's magic. It's like the agent will be like, okay, cool. Let me look at your current code base. I'll make a branch, create a PR, make the changes. And then you can ask the agent, be like, okay, take me a screenshot and put on the PR. And then your job is, kind of what you're talking about [SPEAKER_01] in your workflow earlier with email. It's like, you don't merge it, but you're like, okay, [SPEAKER_01] I've got a good starting spot, right? And then you can come in there and riff. [SPEAKER_01] What have you learned about what makes for a good internal agent experience internal to a

SPEAKER_02

product that you may not have known before Figma agent? I don't know. That's an interesting question. What would make it a good product? I think specifically, this is very specific to Figma, but context and personalization matters. In other products like I've worked on in the past for AI companies, personalization is often kind of the last thing. You get it just working for everybody first. But I think the thing that really [SPEAKER_01] differentiates an okay agent to one that people really love is the personalization aspect. Like we talked [SPEAKER_01] about memory as a form of it in these third-party chat agents. I think for Figma's version of that, it's

SPEAKER_02

[SPEAKER_01] the design system. If you have an assistant, but without the concept of how we [SPEAKER_01] structure our designs and how we put them together, the designs that it creates just [SPEAKER_01] aren't usable from that perspective. I don't know what your plans are for Figma and [SPEAKER_01] proactive agents. But I'm curious, to the extent you have those plans and those experiments, how that's working. Obviously, we've talked about that being hard to get right. Yeah. I mean, I think that's where the future is going. [SPEAKER_01] If you look at how agents have evolved, I will say we've got a lot of things

SPEAKER_02

[SPEAKER_01] cooking internally. I can't talk about specifics too much. I definitely think this [SPEAKER_01] is an area we want to head. Right. And so I think I can talk about the problems that we see. [SPEAKER_01] aren't usable from that perspective. I don't know what your plans are for Figma and the proactive being a proactive agent. But I'm curious to the extent you have those plans and those experiments, how that's working. Obviously, we've talked about that being hard to get right. Yeah. I mean, I think that's where the future is going.

SPEAKER_02

[SPEAKER_01] If you look at how agents have evolved, I will say we've got a lot of things cooking internally. I don't know. I can't talk about specifics too much. I definitely think this is an area we want to head. Right. And so I think I can talk about the problems that we see today. It's if the amount of software is really exploding in the world, one of the bigger challenges then becomes how do you make sure that it's consistent with your values? Right. And so I think, and we become the bottleneck then, right. It's we only have so many human eyes to review all of this work. And so how do we provide a

SPEAKER_02

[SPEAKER_00] solution that allows people to make sure that they can continue to innovate at the speed that these agents create, but also maintains their values. What has been the transition like internally in Figma in terms of your own workflows in the engineering org, in the product org, in the design org from a pre-AI world to now? So I joined in January and I would say even in that period of time, it's been night and day.

SPEAKER_02

[SPEAKER_01] I think folks in January were experimenting with these new ways of working and across all the functions, right? I think probably engineering was leading the way as they do in most of these cases, but I'll give you an example on the product org. So we had an offsite and I think you might have actually come by. Yeah, weirdly enough, small world.

SPEAKER_02

[SPEAKER_01] I think one of my favorite memories of that offsite is that we have a product operations team and they had put together what we call PMOS. And to take a step back for a minute, one of the unlocks to me about AI is you kind of realize every problem becomes a context problem. And it's all then the work becomes about framing the problem with the right set of information. And so our product operations team is like, a lot of the work that we do is in the structured data as a PM. And why don't we aggregate that all together? And okay, let's get a copy of the org chart. We'll throw that in the SQLite table and put it in the file system. And then okay, why don't we create a connector to Asana and then we'll connect Slack and then we'll connect GitHub and a few other things. Right. And then the real insight was skills had really taken off at this point. And okay, the magic sauce here is one of the skills that I really like is this onboarding file creation. So when you add a new team member to your team, as a manager, you got to create a customized, okay, here's the channels you should know. Here's the people you should know. And it takes a lot of that knowledge that I would have previously said was entirely in my head. But then you start to look at how, if you shape the context, right, that data was actually already there, right? I already had the org chart and it can walk the org chart and figure out who's the team and who's the trifecta on the product engineering design side for this new team. You just have to identify here's the new person, here's the team they're going to join. And okay. And then it can do a bunch of research. And then it goes into Slack and it figures out the channels. It's oh, this team is probably these three channels.

SPEAKER_02

And then it reads the last 30 days of content. And then it's here, and then it goes and checks the Asana board and finds all the projects. And so it comes back and you're like, oh, this is uncannily good. It's yeah, this is a pretty good starting spot.

SPEAKER_02

That's one of the things that I think is the thing that made Cloud Code so good and is the thing that makes Codex so good right now is everyone tried to start with agents that live in the cloud are always on, but then you have to manually connect them to everything. And Cloud Code is just no, it's an agent on your computer has access to everything that you have access to. And then that just totally changes all the stuff it can do because it can get all the context it needs. And same thing with Codex. I can just ask a random question to Codex. We just published an article today and I was like, who should I send this to? And it just went through my emails and my texts. And I didn't even know it had access to any of that. It just found five people that I probably would have forgotten about, but that I should have sent it to. And I did. And that's the sort of magical thing that's starting to happen now. That was very, I think the technology, AI itself, if you gave it all the context would have been able to do this for a while, but it's only now that it's in the right harness and form factor, and it can do it a little bit more independently than it was able to before. Yeah. I'm going to put a plea out there. I remember WWDC 24. I can't remember when it was, Apple Intelligence, whenever the headline was WWDC. Oh my God. I think that was more recent. I think it was 25.

SPEAKER_01

[SPEAKER_02] It was 25. Gosh, AI time, who knows what year it is. I upgraded my phone. I was all in, I upgraded my iPad. I was I'm going to get, because I had just upgraded the last cycle and then they're like, well, you need a new processor. I was like, okay, I upgraded all my stuff. Solid WWDC. I was like, this is great. Because they had this concept of our phones have all of this personal data. And I was like, oh my gosh, this is going to be it. And then I will say I'm really hoping WWDC this year is actually it. Because this past year whatever has not been it. Right. Because to your point, the technology has been there. The hard part that's missing is tying it all together. And the mobile phone ecosystem has all that content. And so I'm waiting for the thing where it's the always-on Siri that runs in the background and is actually smart rather than the

SPEAKER_01

[SPEAKER_02] Our phones have all of this personal data. And I was like, oh my gosh, this is going to be it. And then I will say, I'm really hoping WWDC this year is actually it. Because this past year and whatever has not been it. Right. Because to your point, the technology has been there. The heart that's missing is the tie it all together. And the mobile phone ecosystem has all that content. And so I'm waiting for the thing where it's the always on Siri that runs in the background and is actually smart rather than the one that's like, what was that? I didn't understand you, Matt. One day. Do you think they're going to be able to get that right? And if they don't, do you think it matters?

SPEAKER_01

I think it still matters because I think even them being late to the game, they are still the king of context. Right. And I think that's what's been interesting to watch about Google IO too this year is seemingly Google has also woken up to that, that they don't maybe have as much data as Apple, but they have a lot. And it seems like they're now starting to marry their AI products. If you can keep all the names straight, I saw this great tweet that was like, is it Bard or Gemini Pro or Spark or whatever the thing is, but there's some, I think it's Spark is the product that is supposedly going to be the always on agent that is auto connected to all of your Google content. And so I'm waiting for the day that it runs my inbox for me and I get to inbox zero. Yeah. I think I just have this feeling about Apple that when OpenAI took off, everyone started buying Mac minis and you're just like, that's such a good business. Like they didn't have to even be in the AI race because they win by default because it's the thing that everyone runs the AI on. And even if they're behind on Apple Intelligence, which they are, and historically their software products have been lagging behind their hardware, because their hardware is so good, it doesn't matter. They have a lot of time to catch up.

SPEAKER_01

Yeah. Well, and I think their strategy is very smart, right? It's the privacy play, right? And I think it is scary to upload all of your information to the cloud. And so I do think they're in the game and I really am hoping that they've got something interesting this year. So if we look back over the last year, there's been this big sea change in how we build stuff, how good the tools are. And then maybe correspondingly, how software works and how we build software and all that. What do you expect or what are you planning for over the next year as the capabilities increase, both in how you make stuff and what you make?

SPEAKER_02

I think the big thing this year will be about how do we review better? I think that's where the bottleneck is now, right? I think we have agents that are capable of producing all this stuff. They're available enough. They're cheap enough. And now we're just being inundated with all of this new content. And I think people are getting overloaded of what do I do with all of this? It's not just summaries of stuff that's been out for a while, but now it's okay, this is net new content that it's like, do you want me to go or not? And I think we have to solve the problem of how do we scale our value system, how do we evaluate that this new thing that we're creating is actually good and feel confident and have enough trust in it that it can go to some degree in auto mode, if you will.

SPEAKER_02

Do you have any inkling about how that will work inside of Figma or what the interesting design considerations are for that kind of flow?

SPEAKER_02

And I think that's one of the problems that we're really focused on is talking to customers, understanding. I think a lot of customers and us are figuring out at the same time, right? I think the industry in general is trying to understand what is the new way? Is it a video walkthrough that's recorded, right? Or is it screenshots? Or is it another agent that's with a different prompt that reviews the work? And then you trust this agent so much that you approve its decisions. I don't know. It's hard to predict the future, especially at this time.

SPEAKER_02

Yeah. One last question for you. So I know there's been a lot of back and forth over the last year or two about, is there a future for PMs? Is there a future for designers? And if you want to be a PM, how do you break into the industry now? Because maybe there are fewer PM seats or engineers don't need PMs. How do you think about the career progression for a PM and how someone who's not senior becomes and gets to where you are?

SPEAKER_02

Yeah, that's an interesting one. I think the fundamentals still matter, right? I think the best analogy I've seen is there was math class in school, but you still had a calculator, right? But at the same time, we all learned long division, right? And how to do that. And I remember integrating and taking derivatives and the rest of that. And do I do that on a daily basis now? Absolutely not. But I think it's incredibly important in order to drive these systems that you have an understanding of what these concepts are, be able to do it by hand. And so I think the foundations still matter. And I would be really curious actually at this point to see what CS classes look like, what is a one-on-one CS class, right? Because I think there are two parallel worlds. There's one where it's oh my gosh, you can just dump your answer or question into ChatGPT. And it's literally here's the bubble sort for you, right? Like, which of the 42 ways do you want it? And then there's a version of that where it's I'm a really curious person. I wrote the bubble sort in C, but take it, put it in assembly and then explain it line by line to me and explain what a register is and what is L1 cache and L2 cache and the rest of it.

SPEAKER_02

What CS classes look like, what is it like a one-on-one CS class, right? Because I think there's two parallel worlds. There's one where it's, oh my gosh, you can just dump your answer or question into ChatGPT. And it's literally, here's the bubble sort for you, right? Like, which of the 42 ways do you want it? And then there's a version of that where it's, I'm a really curious person. I wrote the bubble sort in C, but take it, put it in assembly and then explain it line by line to me and explain what a register is and what is L1 cache and L2 cache and the rest of it. And I think being a curious person, I guess, maybe this is the answer. I think the most important thing is to be a curious person, right? Because I think with these new tools, the people who cannot leverage them are the ones that just accept the output. And the people that are able to invent the next set of tools and drive the tools to their maximum are the ones that are pushing the boundaries and understand how it's put together. And in order to be able to do that, you have to be the curious person. You can't have been the one who's answer this problem for me. Right. And just give me the answer. You have to be that curious person to be, how is this thing put together? And how does this actually work? And help you understand the next level. Right. I agree. And it's so much more fun to live that way. Totally. I mean, it's catnip for me. I remember because it's, I don't know if you're a Hitchhiker's Guide to the Galaxy person, but I feel like LLMs are the book. It is literally the manifestation of it. And you can even have them on your, I have this when I go on airplanes. I don't run a lot of local LLMs, but I'll download an 8B model and run it on an airplane. It's literally that. It's the Hitchhiker's Guide to the Galaxy. You can ask it a question, why is the sky blue? And it breaks it down into the refraction and all the rest of it. And you're, well, what is a squirrel? And it will give you an answer of that. And so it's, and you know, they're not perfect. And some of them are a little weird, especially at the 8B size, but I don't know. It's a magical time to be alive for curious people. I'd say. I totally agree. Matt, it was a pleasure. Thanks. Oh my gosh, folks, you absolutely positively have to smash that like button and subscribe to AI and I, why? Because this show is the epitome of awesomeness. It's finding a treasure chest in your backyard, but instead of gold, it's filled with pure unadulterated knowledge bombs about ChatGPT. Every episode is a roller coaster of emotions, insights, and laughter that will leave you on the edge of your seat craving for more. It's not just a show. It's a journey into the future with Dan Shipper as the captain of the spaceship. So do yourself a favor, hit like, smash subscribe, and strap in for the ride of your life. And now without any further ado, let me just say, Dan, I'm absolutely hopelessly in love with you.

SPEAKER_02

okay, let's just, let's just say, I think we're, we're on the same page. SaaSpocalypse, not a, not a thing like actually making a piece of SaaS software that works all the time is a gigantic effort that only some, some people want to do and other people just want to pay for.

SPEAKER_02

But then let's like dive more, more distinctly down into Figma land. So, you know, there, there are questions about in a, in a design world, what kind of a, do I want to just chat with my landing page and move things around that way? Or do I want to be on the infinite canvas? And I know internally, like pretty much all of our designers, they're super AI pulled, super early adopters. And they're all like, yeah, like the, you know, typing is good for a first pass, but like to actually get the details, right. I need to be able to like move stuff around. So in, in, in the design world, like how does that change? How do you think that changes the product strategy when the,

SPEAKER_02

the, the possibilities for how you might design something have changed so radically? Yeah. I mean, I think it's a lot to unpack and I think, you know, we're in the early innings here and we're kind of figuring it out. I think we're still in kind of this hangover of like the text box rules. Like I think so much of us are defaulting to like chat as the experience for generative UI.

SPEAKER_01

And I feel like we're starting to enter the second chapter of that, of like, what does it mean? And like, that's part of the reason I'm so excited about our agents launched. Like we've had it internally for a while. And for those who haven't seen it yet, it's the ability to use an agent directly on the infinite canvas. And I think going back, it's funny about like, what's old is new again. Like there's a lot of like this happening throughout LOM and ML land of like, we've reinvented evals. It's like, okay, well, we had unit tests before. And it's like, well, we've reinvented like

SPEAKER_02

prompting and we'd like, we had user input before. And it's like, okay. And design in the new, like AI era is still like the principles still matter. Right. And so like one of the principle, core principles of design for me is like the diamond. And so like, there's this idea of like divergent thinking and then convergent thinking. And like most design problems are like, and like, this is the idea behind brainstorms. And when people tell you like, don't ever, you know, shoot somebody's idea down. Like brainstorming is all about just like generating ideas. And one of the things that I don't think we fully unlocked yet from these new capabilities

SPEAKER_02

is like the ability to like supercharge generative thinking. Like, I think oftentimes, like we get stuck in our own life stories and like lived experience. And like, we approach a problem from a certain angle. And like, this is what's so valuable about having teammates, right? It's like,

SPEAKER_01

you go to talk to your teammate and they have a totally different starting point. And it's like, the answer to this problem is different. And like, the creativity comes between the conversation between the two of you of like, Oh, like, I hadn't thought about it from that angle. That's interesting. Let me take it and then build on top of that. Right. And so like, going back to the, like, what does this mean in this new AI world? If we get outside of the text boxes, because I think text boxes are super limiting, and it's very much like a linear, like, well, this and then that, and then this, if we get to the canvas, and you have the ability to like have some of those same

SPEAKER_01

kind of concepts, but the agents allow you to do divergent thinking that it's like, Hey, like, I have this frame. Let me start here. It's like, Hey, I think it should be grayscale. And then you have another frame. You're like, well, let me try SEPA. And then it's like, Oh, the SEPA thing is cool. But like, the type is wrong. And it's like, Oh, let me try that. Now it's like the accessibility is off. Let me like duplicate the frame and try again. And so like, I think, and even that is still like, I think early innings, that's like very much like the human driving the input, right? And like,

SPEAKER_01

you kind of have talked about like proactive flow, right? And it's like, I think we're starting to figure out,

SPEAKER_02

like, what if we had an agent that's like, here's a bunch of frames that I are on the canvas, like your job is to like, push them, like try different directions, or don't just like double down. But then I think there's a separate set of agents that it's like, Okay, we have 25 frames on this canvas of like concepts for like a new marketing page, right? And then it's like, how do I how do I channel them down, right? So like, then there's like a convergent agent that's like, Okay, like, these three are kind of like this, and these are clustered around this. And like, I think, like, and you can

SPEAKER_02

ask it for its opinion, you're like, pretend you're like a customer clicking through this, like, which one makes the most sense, right? And so I don't think we've really tapped all of that stuff yet. And so like, I think, like, even the best agents, like the command line agents, like don't have the ability

SPEAKER_01

to like do those workflows. And so that's kind of where I see the future of design and product thinking. I think that makes total sense. And yeah, it seems like, from what I can tell so far, the agents are really good for, I already have a design system, I need a new landing page, like, get me the land, get me a landing page in the design system, I already have kind of thing, which, to be honest, a lot of designers don't really want to have to spend time doing the nth landing page or the nth, like graphic for this, you know, post or whatever, which is more convergent, and a little bit less

SPEAKER_01

divergent. What about, what about like, the future of, you know, maybe Figma design tools, or just generally, generally software with allowing external agents in versus building your own agent, or having both, which you which you all do have? How does how do you think that works? I mean, I think we embrace both, right? Like, and I think this is like, I think design workflows are different than engineering workflows, but like, the lines are blurring. And so like, I think, in the future, we're going to be all builders, right? And that kind of comes like, which perspective are you

SPEAKER_01

coming at the problem from? And so we definitely very much support third party agents today. And like, our answer for that is our MCP server, right? And so I think one of the nice things about MCP is like, it allows like a standardized interface across all these different kinds of tools, right? And so we kind of think about the problem in two directions, we think about it as like, code to design. So it's like, okay, in that scenario, you just said, like, that's like a pretty common thing. You're like, hey, I have a signup page, but like, it doesn't support GDPR, right? Like, most people are not

SPEAKER_01

going to be like, you know what I should do, I'm going to start with a greenfield page and re-imagine what our signup flow should be to add GDPR. Most people are trying to get their job. It's like,

SPEAKER_02

you like it on Monday morning, you're like, okay, I just got to get this thing done, right? And like, let me add the checkbox here, right? And so for that workflow, it's like, if you are comfortable in Codex or Cloud or Windsurf or Cursor, you pull up your code base, you fire up the MCP server, and you ask it like, hey, can you copy, like, go to this page, like fire up the dev server, go to this page and copy it into Figma Canvas, and it will actually do it. Like, that's one of the releases we had earlier this year, which is like a little bit mind blowing that agents can do this faithfully, but it turns out they can. And now you have just removed all that drudgery, right?

SPEAKER_02

And you've got it into like a medium where you can actually interact with it. It's like, okay,

SPEAKER_01

let me go like move things around very precisely with the direct manipulation tools that most people

SPEAKER_02

are comfortable with. And then the other kind of workflow that we think about is like, okay, now that we've got the design, take that design and bring it back to code, right? And so we've got a tool called Git Design Context, which takes a Figma design, wraps up all the different properties and like components that you're using, as well as like any other kind of guidelines you've provided in your design library and provides it to the agent. And again, it's kind of like magic. It's like the agent will be like, okay, cool. Let me like look at your current code base. I'll make a branch,

SPEAKER_02

create a PR, make the changes. And then like, you can ask the agent, be like, okay, take me a screenshot and put on the PR. And then like your job is like, kind of like what you're talking about

SPEAKER_01

your workflow earlier with email. It's like, you don't merge it, but you're like, okay, well, I've got a good starting spot, right? And then it's like, you can come in there and then riff. What do you, what have you learned about what makes for a good internal agent experience internal to a

SPEAKER_02

product that you may not have known before, before Figma agent? I don't know. Like, that's like a interesting question. What would make it a good product? I think specifically, this is like very specific to Figma, but like the context and personalization matters, like in other products, like I've worked on in the past for like AI companies is like personalization is often like kind of the last thing, like you get it just working for everybody

SPEAKER_01

first. But I think the thing that really differentiates like an okay agent to one that people really love is the personalization aspect. Like we talked about memory, like as a form of it in these like third-party chat agents. I think for Figma's version of that, it's like the design system that it's like, if you have assistant, but without like the concept of like, well, this is like how we structure our designs and like how we put them together, like the designs that it create just aren't usable from that perspective. I don't know like what your plans are for Figma and, and you know, the like proactive being like being a proactive agent. But I'm curious to the extent

SPEAKER_02

you have those plans and those experiments, how that's working. Obviously, like we've talked about that being kind of hard to get right. Yeah. I mean, I think that's where the future is going.

SPEAKER_01

Like if you look at like how like agents have kind of evolved, I will say we've got a lot of things cooking internally. I don't know. I can't talk about specifics too much. I definitely think this is an area we want to head. Right. And so like, I think I can talk about the problems that we see today. It's like, if the amount of software is really exploding in the world, one of the bigger challenges then becomes like, okay, well, like how do you make sure that it's like consistent with your values? Right. And so like, I think, and like we become the bottleneck then, right. It's like,

SPEAKER_00

we only have so many human eyes to review all of this work. And so it's like, how do we provide a

SPEAKER_02

solution that allows people to make sure that they can continue to innovate at the speed that these agents create, but also maintains their values. What has been the transition like internally in Figma in terms of your own workflows in the engineering org, in the product org, in the design org from a pre AI world to now? So I joined in January and I would say even in that period of time, it's been like kind of night and day.

SPEAKER_01

Like, I think folks in January were kind of experimenting with these new ways of working and like across all the functions, right? Like I think probably engineering was leading the way as they do. And I think most of these cases, but I'll give you an example on the product org. So like we had an offsite and I think you might've actually come by. Yeah, weirdly enough, like small world. I think one of my favorite memories of that offsite is that we have a product operations team and they had put together what we call PMOS. And to like take a step back for a minute, like one of the, one of

SPEAKER_01

the like unlocks to me about AI is like, you kind of realize every problem becomes a context problem. And it's all like, then the work becomes about like framing the problem with the right set of like information. And so our product operations team is like, huh, like a lot of the work that we do is like in the structured data as a PM. And it's like, why don't we like aggregate that all together? And it's like, okay, let's get a copy of the org chart. We'll throw that in the SQLite table and put it in the file system. And then it's like, okay, why don't we create like a connector to Asana

SPEAKER_01

and then we'll connect a Slack and then we'll connect, you know, you can GitHub and a few other things. Right. And then the real insight was like skills had really taken off at this point. And it's like, okay, the magic sauce here is like one of the skills that I really like is like this onboarding file creation. So like when you add a new team member to your team, like as a manager, you got to like create a customized, like, okay, here's the channels you should know. Here's the people you should know. And like, it takes a lot of that knowledge that I would have previously said

SPEAKER_01

was like entirely in my head. But then you start to look at like how, like, if you shape the context, right, that data was actually already there, right? Like I already had the org chart and it can walk the org chart and figure out like, who's the team and like, who's the trifecta on the product engineering design side for this new team. You just have to identify like, here's the new person, here's the team they're going to join. And it's like, okay. And then it can do a bunch of research. And then it goes into Slack and it figures out the channels. It's like, oh, this team is probably these three channels.

SPEAKER_02

And then it's reads the last 30 days of content. And then it's like, here, and then it goes and checks the Asana board and finds all the projects. And so like, it comes back and you're like, oh, this is like uncannily good. It's like, yeah, this is a pretty good starting spot. That's one of the things that I think it's the weirdly, the it's weirdly, I think the thing that made cloud code so good and is, is the thing that makes codecs so good right now is everyone tried to start with agents that like live in the cloud are always on, but then you have to like manually connect them

SPEAKER_02

to everything. And cloud code is just, yeah, no, it's just an agent on your computer has access to everything that you have access to. And then that just totally changes all the stuff it can do because it can get all the context it needs. And same thing with codecs. Like I can just ask a random question to codecs. Like we just published an article today and I was like, who, who should I send this to? And it just went through my emails and my texts. And like, I didn't even know it had access to any of that. It just found like, like five people that I probably would have forgotten

SPEAKER_02

about, but that I should have sent it to. And I did. And that that's like the sort of magical thing that's starting to happen now. That was very, it was, I think the technology, like AI itself, if you gave it all the context would have been able to do this for a while, but it's only now that it, it can, it's in the right harness and form factor, and it can do it a little bit more independently than it was able to before. Yeah. I'm going to, I'm going to put a plea out there. Like I remember WWDC, it was like 24. I can't remember when it was like Apple intelligence, whenever the headline WWDC. Oh my God. I think that was more recent. I think it was 25.

SPEAKER_02

It was a 25. Gosh, again, AI time, who knows what year it is. I upgraded my phone. I was like all in, I upgraded my iPad. I was like, I'm going to get, cause like I had just upgraded the last cycle and then they're like, well, you need a new processor. I was like, okay, up to all my stuff. Solid WWDC. I was like, this is great. Cause like they had this concept of like, our phones have all of this personal data. And I was like, oh my gosh, I was like, this is going to be it. And then like, I will say, like, I'm really hoping WWDC this year is actually it. Cause like, you know, this past year and whatever has not been it. Right. Cause like, to your point,

SPEAKER_02

like the technology has been there. The heart that's missing is the like tie it all together. And like the mobile phone ecosystem, like has all that content. And so like, I'm waiting for the thing where it's like the always on Siri that runs in the background and is actually smart rather than the one that's like, what was that? I didn't understand you, Matt. One day. Do you think they're going to, do you think they're going to be able to get that right? And if they don't, do you think it matters? I think it still matters because I think even them being late to the game, they are still the king of

SPEAKER_02

context. Right. And I think that's, what's been interesting to watch about like Google IO too, this year is like seemingly Google has also like kind of woken up to that, that like, they don't maybe have as much data as Apple, but like they have a lot. And it seems like they're now starting to marry their AI products. If you can keep all the names straight, like, like I saw this great tweet that was like, is it Bard or Gemini pro or spark or like, like whatever the thing is, but there's some, I think it's spark is the product that is supposedly going to be the always on agent that is

SPEAKER_02

auto connected to all of your Google content. And so I'm waiting for the day that it like runs my inbox for me and I get to inbox zero. Yeah. I think I just have this feeling about Apple that, you know, like when open claw took off, everyone started buying Mac minis and you're just like, that's such a good business. Like they didn't have, they don't have to even be in the AI race because they win by default because it's the thing that everyone runs the AI on. And, and, and so even if they're behind on Apple intelligence, which, which they are, and historically like their software products have been kind of lagging behind their, their hardware,

SPEAKER_02

because their hardware is so good. It doesn't matter. They have a lot of time to catch up. Yeah. Well, and I think their strategy is very smart, right? It's like at the privacy play, right? Like, and I think like, it is scary to upload all of your information to like the cloud. And so like, I do think they're in the game and like, I really am hoping that they've got something interesting this year. So if we, if we sort of look back over the last year, there's been this like big sea change in how we build stuff, how good the tools are. And then maybe correspondingly, like how software works and

SPEAKER_02

how we build software and all that kind of stuff. What do you expect or what, what are you planning for over the next year as the capabilities increase, both in how you make stuff and what you make? I think the big thing this year will be about the, like, how do we like review better? Like, I think that's where the bottleneck is now, right? That it's like, I think we have agents that are capable of producing all this stuff. They're available enough. They're cheap enough. And now like, we're just being inundated with like all of this new content. And like, I think people are getting kind of overloaded of like, well, what do I do with all of

SPEAKER_02

this? It's not just like summaries of like stuff like that's been out for a while, but now it's like, okay, this is like net new content that it's like, do you want me to go or not? And like, I think we have to solve the problem of like, how do we scale, like I said, kind of our value system of like, how do we evaluate like that, that this new thing that we're creating is actually like, and then feel confident and have enough trust in it that it's like, it can go to some degree in like auto mode, if you will. Do you have any inkling about how that will work inside of Figma or what the, what the, what the interesting design considerations are for that kind of flow?

SPEAKER_02

And I think that's like one of the problems that we're really focused on is like talking to customers, understanding, like, I think a lot of customers and us are figuring out at the same time, right? That it's like, I think the industry is in general trying to understand like, what is the new, like, is it a video walkthrough that's recorded, right? Or is it like screenshots? Or is it like another agent that's like, with a different prompt that reviews the work? And then like, you trust this agent so much that you approve its decisions. I don't know. It's hard to predict the future, especially at this time. Yeah. One last question for you. So I know, I feel like there's

SPEAKER_02

been a lot of back and forth over the last year or two about, is there a future for PMs? Is there a future for designers? And, and if you're, if you want to be a PM, like, how do you break into the industry now? Because, you know, it's maybe there's fewer, there's fewer PM seats or engineers don't need PMs. How do you think about the career progression for a PM and how someone who's not senior become like gets to where you are? Yeah, that's an interesting one. I think the fundamentals still matter, right? Like, I think, I think the best analogy I've seen is like, I mean, there were, there was math class in school,

SPEAKER_02

but like, you still had a calculator, right? But like, at the same time, we all learned long division, right? And like, how to do that. And like, I remember like integrating and like taking derivatives and the rest of that. And like, do I do that on a daily basis now? Absolutely not. But like, I think it's incredibly important in order to drive these systems that you like have an understanding of like what these concepts are, be able to do it by hand. And so like, I think that the, like the foundations still matter. And like, I would be really curious actually at this point to see like,

SPEAKER_02

what CS classes look like, what is it like a one-on-one CS class, right? Because like, I think there's like two parallel worlds. There's one where it's like, oh my gosh, you can just dump your answer or question into chat GPT. And it's like, literally like, here's the bubble sort for you, right? Like, which of the 42 ways do you want it? And then there's a version of that where it's like, I'm a

SPEAKER_01

really curious person. I wrote the bubble sort in C, but like, take it, put it in an assembly and then explain it line by line to me and like explain like what a register is and like, what is L1 cache and L2 cache and like the rest of it. And like, I think being a curious person, I guess, yeah, maybe this is the answer. I think the most important thing is to be a curious person, right? Because like, I think with these new tools, the people who like cannot leverage them are the ones that just kind of like accept the output. And the people that like are able to like invent the next set of tools and like really get drive the tools to

SPEAKER_01

their maximum are the ones that are like pushing the boundaries and understand how it's put together. And in order to be able to do that, you have to be the curious person. Like you can't have been the one who's like answer this problem for me. Right. And just give me the answer. You're like, have to be that curious person to be like, how is this thing put together? And like, how does this actually work? And like, help you understand the next level. Right. I agree. And it's, it's so much more fun to

SPEAKER_02

live that way. Totally. I mean, it's like catnip for me. It's like, I remember it's like, cause it's like kind of like, I don't know if you're a hitchhiker's guide to the galaxy person, but like, I feel like LLMs are the book, like, like it is literally the manifestation of it. It's like, and you can even have them on your, like, I have this when I like go on airplanes, I don't run a lot of local LLMs, but like I'll download an 8B model and run it on an airplane. It's like literally that it's like the hitchhiker's guide to the galaxy. Or it's like, you can ask it a question,

SPEAKER_02

like, why is the sky blue? And it's like, breaks it down into like the refraction and all the rest of it. And you're like, well, what is a squirrel? And it will give you an answer of that. And so it's like, and you know, they're not perfect. And some of them are a little weird, especially at the 8B size, but like, I don't know. It's like, it's a magical time to be alive for like curious people. I'd say. I totally agree. Matt, it was a pleasure. Thanks.

SPEAKER_02

Oh my gosh, folks, you absolutely positively have to smash that like button and subscribe to AI and I, why? Because this show is the epitome of awesomeness. It's like finding a treasure chest in your backyard, but instead of gold, it's filled with pure unadulterated knowledge bombs about chat GPT. Every episode is a roller coaster of emotions, insights, and laughter that will leave you on the edge of your seat craving for more. It's not just a show. It's a journey into the future with Dan Shipper as the captain of the spaceship. So do yourself a favor, hit like, smash subscribe, and strap in for the ride of your life. And now without any further ado, let me just say, Dan,

SPEAKER_02

I'm absolutely hopelessly in love with you.