Complete Course: AI Product Design

1. 0:00 – 1:52

   Intro

   1. AGAakash Gupta0:00

      You helped launch Google Maps in India. India doesn't use street names. Normally, the way it works in India is you just ask somebody on the side of the street and they say, "You go three gullies this way, then you turn left, then you turn right at the temple." And you're like, "Okay, there we go. [laughs] That's how I'm gonna get there."

   2. ELElizabeth Laraki0:15

      Ultimately ended up building and launching direction systems that were much more landmark based.

   3. AGAakash Gupta0:19

      Elizabeth Laraki is one of the original designers on Google Search and Maps. She's going to break down AI product design from every angle, including how to design AI features, how to use AI design tools. For anyone building AI products, founders, PMs, designers, engineers, this episode is packed full of insights.

   4. ELElizabeth Laraki0:37

      Google was always kind of nerdy and kind of quirky, sort of, I mean, even from the logo that was all the primary colors. Looking at the old versus the new, the new, one at a glance, just looks like Apple Maps, and two, like just feels colder and less real.

   5. AGAakash Gupta0:51

      What is the high-level process for designing AI features?

   6. ELElizabeth Laraki0:54

      There are really kind of three key steps. One is defining the product that you're building, the second is designing it, and the third is building it.

   7. AGAakash Gupta1:02

      What are some AI products that you think are designed really well?

   8. ELElizabeth Laraki1:05

      I think that there are tons of highly useful AI products. I think ChatGPT has nailed things.

   9. AGAakash Gupta1:11

      I think you have a pretty crazy story about an AI image expander tool.

   10. ELElizabeth Laraki1:15

       It was one of those things, if I just kind of... It made me pause for a minute.

   11. AGAakash Gupta1:19

       [upbeat music] Really quickly, I think a crazy stat is that more than 50% of you listening are not subscribed. If you can subscribe on YouTube, follow on Apple or Spotify Podcasts, my commitment to you is that we'll continue to make this content better and better. And now on to today's episode. Product design for AI products is incredibly difficult. How do you break out of the conversational chat UX? How do you design for the next generation

2. 1:52 – 4:19

   Elizabeth's background at Google

   1. AGAakash Gupta1:52

      of AI products? Today we are meeting with Elizabeth Laraki. She is one of the original designers on Google Maps and Google Search. She is one of the most knowledgeable people in product design in the entire world, and today she is going to break down how to design AI features, how to use AI tools for design, and everything else you need to know about product design for AI features. We are not going to gatekeep any of the knowledge. We are going to actually demo and show you guys the sauce. So strap in and let's get started. Elizabeth, welcome to the podcast.

   2. ELElizabeth Laraki2:33

      Thank you, Aakash.

   3. AGAakash Gupta2:35

      In 2006, you were one of four designers on Google Search, and what's amazing about Google Search is that those designs you guys created in 2006 and 2007, that's basically what we were seeing all the way until 2023. How did you guys nail those designs back then?

   4. ELElizabeth Laraki2:54

      Sure. Um, so I think kind of at that time, one of the first things that we were focused on -- So if you look, if you look at Google Search results, um, they are text-only list of kind of the top matching results based on keyword. Um, and at the time, we were really trying to diversify what we could show in the results. So we were looking at how do you begin to mix in images, how do you begin to mix in videos, map results, and kind of this growing corpus of, of information. And I mean, if you look at where Google Results is today, it does look quite similar. This is maybe not the best example, um, here, but it's kind of a, you have kind of dedicated information here. Um, so I mean, today what, what kind of... So one, designers have been evolving Google Search this entire time. So there are a lot of like little nuances and details that kind of have made it cleaner, have made it simpler, have kind of really allowed it to continue to be dynamic and fold in more and more information. Obviously, today you're starting to see kind of AI summaries show up as well. Um, and if you were to search for someplace like Chicago, being able to see maps and images and all sorts of hotels, all sorts of information, um, about, about that particular search.

   5. AGAakash Gupta4:08

      So I think there's a couple key takeaways there, right? You don't need to necessarily redesign the whole thing to be doing really good design work. It's, a lot of it is in the details. A lot of it is in taking

3. 4:19 – 6:19

   Google's AI search integration

   1. AGAakash Gupta4:19

      in the corpus of new information or what's being added new, whether it's video, whether it's maps. And as we both know, the latest [laughs] new thing, of course, is AI search. So you mentioned this as well. What is your take on how Google is leaning into AI in its search results? A lot of people actually seem to hate it. A lot of people seem to love it. It sometimes hallucinates and forces an answer that is not even true. How would you approach designing for this?

   2. ELElizabeth Laraki4:48

      I mean, one, I think it's great. Um, I'll put, I'll put, put myself squarely in the great camp. Um, that I think especially when trying to compete with something like ChatGPT, right? So whenever, I think, you know, for the last decade, 15 years, when people had... 20 years, when people had a question, they went to Google, right? Like Google became the answer of all things. And I think what we're seeing is with ChatGPT, more and more people are starting to go to ChatGPT for answers versus go to Google for answers. So rather than Google staying static and having ChatGPT be very different, right, I think beginning to fold AI naturally into the places that people already are makes a huge amount of sense. Um, yes, sure, it hallucinates. So does every other [laughs] LLM and AI tool on the planet, right? So I think there are safeguards with which, as users, we need to think about when kind of, um, assuming any answer. I think it's very much like the early internet, right? Where it was kind of people would be like, "Oh, no, no, no, it's true. I read it on the internet," right? Which is absolutely not like, the internet is not kind of a, um, definitive, credible source in and of itself, and neither is AI, right? So I think it, yes, it still takes some, some user discretion, but I think it's actually very smart of Google to be trying to fold AI thoughtfully into where people already are.

   3. AGAakash Gupta6:09

      Yeah. When designing products, I think it's like how do we figure out what are the core jobs to be done with new technology? How do we place it into that current flow that they're doing? So

4. 6:19 – 9:44

   Designing image & video for AI

   1. AGAakash Gupta6:19

      actually, where they're putting it is good, but this example we have here, for instance, where shake hands with an old bear. They're literally talking about shaking hands with an old bear. That's, part of that is a problem of working with the research teams, working with your evals teams, minimizing hallucinations, maybe not even showing these AI search results when we aren't as confident in them.

   2. ELElizabeth Laraki6:40

      Sure.

   3. AGAakash Gupta6:40

      So I think there's this key element around AI products that people need to acknowledge also when they're building so that this is a non-deterministic product, and we need to design accordingly.

   4. ELElizabeth Laraki6:51

      Yeah. I think, yes, that all makes sense. Yeah.

   5. AGAakash Gupta6:55

      You've talked about how Google could use image and video in Search. What would be a revolutionary new way to design for that?

   6. ELElizabeth Laraki7:03

      Yeah. So this example was actually not specifically Google. It was for ChatGPT. Um, and, um, I think one of the failings, I think, with a chat interface and a chat UI is that the entire conversation is linear, right? And, like, um, ChatGPT had shared a video about wanting to lower my bi- or wanting, somebody wanting to lower their bike seat and using ChatGPT to figure out how, how they could do it. Um, and I think when thinking about AI or thinking about kind of any, any product, I think what you're trying to figure out is, okay, like, what is the best possible kind of version of answers? Like, in this case, to the left are the experience that actually existed in ChatGPT. It felt like I walked into a bike shop and got, like, the least helpful bike mechanic I could possibly find, right? It's sort of like I ask a s- a specific question, they give a specific answer, right? So it's kind of a what you really want to have unfold is a much more dynamic dialogue, right? If I were to walk into a bike shop with somebody who, like, was really in service of me and my needs, it would be, "Okay, well, let's take a look at your bike seat. Let's sort of see, does it have a quick release lever? Do you need an Allen key? It looks like maybe it's this kind of size Allen key. Here, grab this one. Okay, um, now let's check out the height. Don't forget to tighten it. And, you know, go take a spin, see how that feels. If not, like, let's readjust it." And I think that's really, really hard to have in a linear conversation. Um, on the other hand, like, kind of what you'd want in sort of a remote interaction with a great bike mechanic would sort of be FaceTime, right? So I can kind of pop open a video and have a dialogue going on at the same time and this back and forth or even have the image stay central and just have a video... Or sorry, have an audio conversation kind of happening around the fringes, but the video stays central versus, or the photo stays central and video stays central versus it being this thing that kind of disappears, um, off the, off the screen while we're, you know, talking back and forth.

   7. AGAakash Gupta9:21

      Yeah. And one could imagine this could be a place even where potentially product design feeds input into research to say, "Okay, the ideal design experience here-

   8. ELElizabeth Laraki9:31

      Mm-hmm

   9. AGAakash Gupta9:31

      ... is that we're actually, we're highlighting where [chuckles] that bolt lever is," because just h- reading those words, people may not know that.

   10. ELElizabeth Laraki9:40

       Right.

   11. AGAakash Gupta9:40

       We're highlighting, "Oh, this might be the place to check. Do you need an

5. 9:44 – 16:05

   AI image expander disaster

   1. AGAakash Gupta9:44

      Allen wrench? So turn that and see." And so then you can go back to the research team. The research team can help you build that, and then you can almost implement that into your product.

   2. ELElizabeth Laraki9:53

      Yeah. Yeah.

   3. AGAakash Gupta9:56

      Awesome. So at this point, we've given people a taste and a preview, like AI design doesn't need to just sit in the conversational chat UX of ChatGPT or in the AI search overview answers of Google. You should be thinking about new ways to do that. And one of the ways, of course, that people have been doing a lot of that is within pictures, things like image expander tools. I think you have a pretty crazy story about an AI image expander tool.

   4. ELElizabeth Laraki10:24

      Yeah.

   5. AGAakash Gupta10:24

      It really highlights to me the pitfalls of AI design. Can you talk us through this?

   6. ELElizabeth Laraki10:30

      Yeah. Um, so, um, I was attending a conference, and they asked for a headshot. I gave... It was an AI conference. I, um, provided a headshot, which was the image, um, I'm not sure which direction we're facing. The image that's in color.

   7. AGAakash Gupta10:44

      [chuckles]

   8. ELElizabeth Laraki10:44

      And, um, uh, and then, yeah, I, I'm like what happened behind the scenes, basic- Well, I'll start with the beginning. I sent them the image, which is in color, and later I saw, a couple weeks later, I saw a promotional kind of image for the, for the, the conference that had the black and white image. And it was one of those things if I just kind of... It made me pause for a minute, and I was like, "Wait, some- something's different." And it took me, like to be, to be fair, it took me a minute to figure out what it was, and then I was like [gasps] , "Oh, my God. My bra is showing in this image. Like, has it been showing in my profile picture and I have never noticed it?" Um, and that's when I then, like, looked back at my profile picture. No, it wasn't there. And it was like, "Oh, my God, are they trying to, like, make me sexier for the conference?" Or like what, what is, like what is going on? Um, and it turned out it was, like, completely innocuous, um, and unintentional, but that basically I had sent my profile picture, who had sent it to the person who was doing the website, who had cropped all of the images to be square. And then somebody who was doing social media for the conference took that square image and used an image expander tool to kind of make it a, um, make it a, a portrait kind of size picture. And the image expander tool created this. Um, completely unintentional, completely reasonable workflow, right? But with very unintended consequences. And I think the, there, like there are lots of, [chuckles] lots of points to think about here. But I think one in particular is that, like, when you are interacting with, like-... real people or the sort of hybrid AI people mix, which can happen in many different ways beyond just image expansion, that I think there is needs to be additional scrutiny in place. Now, also, I tried to replicate the same flow using a bunch of different tools, and, like, this was really good [laughs] in comparison to what a lot of other, what a lot of other tools ended up with. Um, obviously it points to interesting cultural biases and all sorts of things that we can, we could, we could dive into, but I think the key is really that AI can have very unintended consequences, and, um, as people using these tools, we need to have, like, a heightened level of scrutiny and discretion.

   9. AGAakash Gupta13:12

      And how do we solve the design challenge? Like, we don't want to be adding in... Even I even felt weird just putting this image on the screen. Like, [laughs] we don't wanna be creating these types of images. How do we correctly build that as designers?

   10. ELElizabeth Laraki13:29

       Um, do you mean as far as the tools or kind of what-

   11. AGAakash Gupta13:32

       How do we build the right safeguards into the product, or how do we work with our research teams to make sure it's not happening? What is the right approach here?

   12. ELElizabeth Laraki13:38

       Yeah. Well, I mean, you can kind of go full cycle, right? So one, there's kind of what data are these models actually being trained on. I think, one, at least early models had a huge amount of porn content, so I hear. I haven't looked at kind of the ex- But it's like these are the images that are accessible. Um, and so I think there's that. There's, like, a whole training piece of, like, what, what is going, what are the, what are the models being trained on? What's, what's going in, into this? Like, I mean, I don't normally like to talk about my, my breasts in public. It is not something that, you know, I spend a lot of time thinking [laughs] about, nor this, but it's like some of these... Like, actually, Photoshop's auto-expander, you know, I mean, gave me enormous breasts, right? Where it's just kind of like, oh, well, this is an interesting, this is an interesting kind of model or an interesting sort of, I don't know, pattern that clearly the model has assumed that, like, when sort of shaped like this, then, you know, we expand in this way or that way. So one is obviously kind of training data. Um, and then I think the other piece is, like, one, I, I, I do feel like... I mean, I'm sure there will be tools that automate a lot of this, and so that becomes another, another question as well. But certainly, like, with human involvement, I think even in this case, like, it's when expanding an image, you go from original image to expanded image, and you don't necessarily see, like, a very clear differentiation of what was expanded versus what was original, right? You just sort of have original as an input, and then you have, like, the options for the outputs. But, um, I mean, some tools cover up just based on where, how the, um, how the screen is organized and where the pixels are. Like, some have tools that overlay the new part of the image, right? Which obviously isn't helpful t- for discerning, like, what, what, what was added looks like. Um, and nine times out of 10, it doesn't matter because you're adding, you know, a little bit of extra, uh, space on the sides to a landscape image or all of these different things. Um, so I think it is hard to necessarily put these safeguards in, and I think it really has to be something more that lives in people's heads. Um, but there are sort of ways with the UI to kind of make the actual versus AI-generated portions of these sort of mixture hybrid images much clearer as to what was original content versus what was AI-filled content.

6. 16:05 – 17:50

   Ads

   1. AGAakash Gupta16:06

      Today's episode is brought to you by Vanta. As a founder, you're moving fast toward product market fit, your next round, or your first big enterprise deal. But with AI accelerating how quickly startups build and ship, security expectations are higher earlier than ever. Getting security and compliance right can unlock growth or stall it if you wait too long. With deep integrations and automated workflows built for fast-moving teams, Vanta gets you audit ready fast and keeps you secure with continuous monitoring as your models, infra, and customers evolve. Fast-growing startups like LangChain, Writer, and Cursor trusted Vanta to build a scalable foundation from the start. So go to vanta.com/aakash. That's V-A-N-T-A.com/A-A-K-A-S-H to save $1,000 and join over 10,000 ambitious companies already scaling with Vanta. Today's episode is brought to you by the experimentation platform Kameleoon. Nine out of 10 companies that see themselves as industry leaders and expect to grow this year say experimentation is critical to their business, but most companies still fail at it. Why? Because most experiments require too much developer involvement. Kameleoon handles experimentation differently. It enables product and growth teams to create and test prototypes in minutes with prompt-based experimentation. You describe what you want, Kameleoon builds a variation of your webpage, lets you target a cohort of users, choose KPIs, and runs the experiment for you. Prompt-based experimentation makes what used to take days of developer time turn into minutes. Try prompt-based experimentation on your own web apps. Visit kameleoon.com/prompt to join the waitlist. That's K-A-M-E-L-E-O-O-N.com/prompt.

7. 17:50 – 18:28

   AI safeguards & human-in-the-loop

   1. AGAakash Gupta17:50

      So a consistent takeaway I'm hearing about AI features versus regular features is that you're gonna have to go work [laughs] with the research team on the underlying AI, whether that is the training data that is going into it or some of the evals that are built on top of it. Perhaps they create an eval that when expanding human body around private parts, go through these checks. Are we showing enough diversity? Maybe that's a good time for us to show an A and a B option to a user. Small, large breasts. I don't know. We have to think through it correctly.

   2. ELElizabeth Laraki18:26

      Yeah.

   3. AGAakash Gupta18:26

      But there could be some more sensitive

8. 18:28 – 31:29

   3-step AI design process

   1. AGAakash Gupta18:28

      ways to handle it, and I really liked your point on then outside of working with the AI, actually working with the UI as well.

   2. ELElizabeth Laraki18:35

      Mm-hmm.

   3. AGAakash Gupta18:35

      So having that UI really clear to whoever it was, poor person on that social media team-

   4. ELElizabeth Laraki18:40

      I know

   5. AGAakash Gupta18:40

      ... for that conference.

   6. ELElizabeth Laraki18:41

      Yeah.

   7. AGAakash Gupta18:41

      This is what we expanded.So that they can just see it very clearly and they can do their own human-in-the-loop check.

   8. ELElizabeth Laraki18:48

      Mm-hmm.

   9. AGAakash Gupta18:49

      So improving the underlying AI, but also making these human-in-the-loop checks, I think is probably the best way for people to deal with AI's non-deterministic nature.

   10. ELElizabeth Laraki18:58

       Yeah.

   11. AGAakash Gupta18:59

       So we covered the basics of AI. What is the high-level process for designing AI features?

   12. ELElizabeth Laraki19:08

       So I think there are really kind of key three steps or three key steps. Um, one is defining the product that you're building, the second is designing it, and the third is building it.

   13. AGAakash Gupta19:20

       Can you say more? What are some AI products that you think are designed really well?

   14. ELElizabeth Laraki19:25

       Um, so I think that there are... I think, I think that there are tons of highly useful AI products. Um, some of the ones that I'm just pulling up right here are kind of... I mean, one, I think ChatGPT has, has nailed things, right? Where it has really become this, like, all-purpose tool for, like, anything and everything under the sun. And [chuckles] whether it's that I just have a question to something, whether it's I want to, you know, begin to think about planning a trip or what would be interesting places there, or, um, I wanna translate, um, uh, a WhatsApp message, right? It's kind of like it does... Or look at a recipe for how to cook something. It's just kind of, it's like, it's so good at so many different things that it really has become this kind of indispensable tool for me and for many, many others. Um, and I was thinking about this where, like, I also do use Claude and Gemini, but, like, more as, like, reference checkers than sort of my first line of knowledge. And I was thinking, um, kind of reminds me a little bit of, like, early Uber/Lyft usage, right? Where I would default Uber was my app, but then I would also just either if it was taking a while to find a driver or if, you know, the pricing seemed really ridiculous, I would go check Lyft as sort of this, like, back pocket app, but that it really... Uber was my dominant app. And I feel the same way with ChatGPT. It has just become so ingrained in, like, being the dominant go-to for so many things, um, that I, I think, yeah, they have done and continue to do an, an incredible job. Um, and then I think, um, so one, I would say, um, [chuckles] there are so many AI tools that cover such an incredibly diverse set of use cases that, like, I don't even know the half of them. Um, and it's impossible, I would also say, to stay up to speed on, on what's changing, what's happening day to day, unless it is, like, your full-time job, um, to just, like, try to keep track of, of what's changing and what's new and what, what's happening. Um, but I think I'll add some, like, new use cases that have opened up, uh, or that, that I think kind of AI tools have opened up for me, um, is that, one, um, sort of either Riverside or Descript, but, like, I don't really know video editing all that much, but I can really trim something down very effectively and very easily using Descript or Riverside. Um, which is just kind of, it, it begins to open up a new tool, um, a new kind of output for, for me and for others. Um, and then Midjourney is another one that I use quite a bit as well, and I don't... I mean, I tend to be fairly utilitarian in my usage of things. Um, and so I think what it has really replaced is, like, trying to find stock imagery or just even looking at is there a possible way to add in imagery that otherwise I would've just, just skipped and not used imagery for anything. And there are, you know, 1,000 different tools. I have a lot of commentary about Midjourney's UI because I feel like it could be a lot better in many different ways, and there are a whole bunch of other, of other obviously AI image generation tools, but, um, those are some that come to mind.

   15. AGAakash Gupta22:45

       So let's walk through these e- one by one.

   16. ELElizabeth Laraki22:47

       Yeah.

   17. AGAakash Gupta22:47

       ChatGPT. What are the design takeaways for somebody building their own product that they should be noticing in ChatGPT that they've done really well?

   18. ELElizabeth Laraki22:59

       So, I mean, I think the, like, common wisdom on the street, right? [chuckles] Is like find a target audience and a target use case and build for that and then expand. And obviously, like, Google or ChatGPT are, like, the total inverse of that, where they're everybody's tool for everything. Um, and I think most, I think many, um, kind of AI tools start with this kind of general purpose for everybody piece that I think is kind of harder to replicate. Um, as far as takeaways, I mean, one, there's just, like, not a lot of crap cluttering things. There's a downside to that too, though, um, which think we can dive into a little later about, like, how do you, for newer people, like, how do you start? How do you kind of overcome the blank, the blank screen problem? And obviously they have sort of, like, suggestions and tips and things. Um, you can use it under many different modes, right? So, like, I can converse back and forth with it, or, or, sorry, via text, or I can input via voice, or I can use it solely in voice mode. I can do all of these kind of hyper-honed, um, sort of wrapper GPTs within it, like for translation and kind of things like that. Um, and so I think maybe main lessons are, one, like, the core use case, which is to come and find information, is really dead simple and easy to do. And then it's, like, all these different, [chuckles] like, um, sort of level ups, right? Of like, "Oh, you can also use it for this, and you can use it for that, and you can use it for that, and you can go in and specialize it and make your own GPT and things," but that the most common core use case is just very simple and approachable.

   19. AGAakash Gupta24:51

       Exactly. And I think that they're really genius in re- getting out of the user's way, where your average product, they're always trying to stuff in those walkthroughs, the step-by-step everything. They're just like, "Here. Here's some suggested prompts. Let's go." And it's been able to handle so many different use cases, and they've been layering in so many cool things, like voice mode. It's kind of a master class in product application design on top of AI. The next category you mentioned was Descript and Riverside. So for people who haven't used those, what are the AI features within, let's take one of those, maybe Descript, that you think are really powerful, and how did they from a design perspective execute on those well?

   20. ELElizabeth Laraki25:32

       Yeah. So I think what's interesting with some of these, right, is that, like, I don't necessarily actually know specifically what's happening with AI versus what is not. So, um, and it kind of doesn't matter that much either. Um, but I mean, I think the fact that I can, like... So there's maybe a couple things, like one, the being able to just edit from the transcript and have that edit the video versus having to listen to the vedi- video and go back and edit it. I don't actually know. I mean, obviously the auto translation is happening from AI, so AI is in there somewhere. But I think the easy tools of being able to edit in what feels like a very intuitive way, um, is quite different than any... And, and I'm also not vel- [laughs] not, not very well-versed in video production. But it's kind of it feels very approachable from an amateur, for an amateur. Um, and then I think things like being able to remove filler words. Like, we all talk with a lot of likes and ums and ahs and things, and the fact that, like, it can just go and identify those, it's not perfect, but, like, it pulls out a big chunk of them. Um, and then sort of loop them or kind of try to fill in or loop things pretty seamlessly is pretty incredible. And there are also features that, again, I don't even know if it's AI, but it's like being able to make it look like somebody is always looking at the camera versus looking at the person they're talking to on the screen, which I hope maybe you use for this, 'cause it's a lot harder to look at the camera than look at somebody that you're actually interacting with, um, on the screen. Um, and, um, and then, like, even then I think there... And I haven't used these sets of tools, but the ability to kind of then auto-generate a bunch of sort of clips and even thinking about things like titles, um, or sort of sections and titles for sections based on the content that was there. Like, it really provides so much. It, it kind of extracts so much data and information that it feels more like kind of almost putting together puzzle pieces versus starting from this massive, unapproachable, like, 60 or 90 minutes of video or whatever it is. Um, and so it does a lot of the, the sort of pre-work for you.

   21. AGAakash Gupta27:44

       What I'm taking away from this is that they're not sprinkling AI on top. They're baking it into the cake. What are the core jobs to be done of editing a podcast? Removing ums and ahs, having a transcript that you can edit off of, being able to create magic clips, being able to brainstorm titles for those magic clips, being able to add the word transcriptions, being able to think about what the description once you put it on YouTube is. So they've basically taken the entire life cycle of editing a video, and they've put AI into each of those steps, and they've allowed you to see that, okay, this is powered by AI, so I might need to double-check it. But otherwise, they're not shoving AI into your face. They're just putting it into the key jobs to be done.

   22. ELElizabeth Laraki28:24

       Yep.

   23. AGAakash Gupta28:26

       And then the final category of product we were talking about is these products like Midjourney-

   24. ELElizabeth Laraki28:33

       Mm-hmm

   25. AGAakash Gupta28:33

       ... which have very controversial [laughs] user interfaces. People who haven't used Midjourney, you might have heard or... You have to sign up for Discord. You join the Midjourney Discord server. Then you join a channel. Then you type in, like, /imagine or /create, and you give it a prompt. And then it comes up with four options. And then you click on that option. It's very convoluted sort of core user interface or user experience. So why has it succeeded despite it, and how might they improve it?

   26. ELElizabeth Laraki29:05

       So to be fair, like, I never used Journey. I never used Midjourney through Discord.

   27. AGAakash Gupta29:10

       Mm-hmm.

   28. ELElizabeth Laraki29:10

       Um, so, um, I'm trying to think of when. Maybe it was sometime last fall that I, maybe a year ago or so, that they made an entry in without having to sign up for Discord. Um, so that being said, I think I did actually sign up through Discord, but, like, I just use a web interface for Midjourney.

   29. AGAakash Gupta29:28

       Hmm.

   30. ELElizabeth Laraki29:28

       So I don't... I like, I... And, like, Discord could be a whole other conversation for another time.
9. 31:29 – 33:25

   Ads

   1. AGAakash Gupta31:29

      Today's episode is brought to you by the AI PM Certification on Maven, run by Miqdad Jaffer, who is a product leader at OpenAI. This is not your typical course. It's eight weeks of live cohort-based learning with a leader at one of the top companies in tech. As you know by now, the future of PM is AI, and this certificate will give you the learnings plus the hardware to show you are ready for an AI PM role. I myself took the course and recommend it. Put on by the amazing team at Product Faculty, including Mo Ali and Paul Mulhern, it's worth it. Former students come from companies like OpenAI, Shopify, Stripe, Google, and Meta. The best part, your company can probably cover the cost. So if you want to get $550 off, use my code AAKASH550C7. That's A-A-K-A-S-H-5-5-0-C-7, and head to maven.com/product-faculty. That's M-A-V-E-N.com/P-R-O-D-U-C-T-F-A-C-U-L-T-Y. AI evals are one of the most important skills for PMs. And do you know who large parts of the OpenAI and Anthropic teams learned AI evals from? Hamel Hussein and Shreya Shankar. Today's episode is brought to you by their AI Evals for Engineers and PMs course on Maven. Most teams are winging it with basic metrics and hoping for the best. Meanwhile, the teams that actually ship reliable AI, they've cracked the code on systematic evaluation. Hamel and Shreya's live Maven course will teach you their battle-tested frameworks behind 25-plus production AI implementations. It's four weeks with live instruction and insane guest speakers, plus a free book. Enroll at maven.com with my code AG-EVALS for over $1,330 off. That's AG-EVALS. Start shipping AI that actually works. Okay, so

10. 33:25 – 38:25

    Designing AI voice interfaces

    1. AGAakash Gupta33:25

       it sounds like key takeaway is big unlock was moving it out of the Discord platform into the web browser, making it easier for people to onboard. Similar way, ChatGPT, open it up, you can use it. That's really important for an AI product, whether that's a chat product or an image generation product. The other really important category of AI products is voice products. How do you design for AI voice?

    2. ELElizabeth Laraki33:50

       Um, so I think voice UI is interesting. One, I don't actually have a background in voice UI. I haven't done much. I know that there has been a lot of research, and especially from, um, uh, sort of engineering, accessibility, um, sort of these different, these different lenses of like how do you kind of replicate interfaces through, through voice. Um, I think a couple of examples that I will share is that one, like with ChatGPT's voice UI, so my husband, um, actually, uh, set the kind of shortcut button on, um, on his iPhone to just be ChatGPT voice. And so we were driving with the kids in the car the other day. Someone was like, ask... I have three kids. They always have all sorts of just kind of random questions they're asking, right? And, um, so he's like, "Well, just ask ChatGPT." And then it was kind of like, "What else do you wanna know?" And it was fascinating for me to have the kids just kind of be piping up with question after question, and it became sort of like educational, um, entertainment in a way. Um, but it was just sort of always on on the background or in the, in the background. Um, and I think what's interesting about designing a UI for this is that like the visual UI doesn't matter, right? It's not even like visible on the screen. In this case, we were driving, like nobody was even looking at the screen. But it was like, it-- I think there what felt so magical was that it truly felt just like a conversation.

    3. AGAakash Gupta35:27

       Yeah.

    4. ELElizabeth Laraki35:28

       Like we had another person in the car hanging out with us. Um, and so I think it's very important to think about the con- the context and different contexts people are in when thinking about designing, um, uh, designing kind of the, the interface for it. Um, I think Liminal's in, or sorry, Lim- um, Limitl- is it Limitless?

    5. AGAakash Gupta35:51

       Yeah.

    6. ELElizabeth Laraki35:52

       Um, is another interesting example that I've had. I actually don't, um, use it, but I have several friends who've had, and especially the, um, the... I mean, I think there are several utilitarian use cases for it of like, oh, you know, give me the summary of this conversation or remind me what, you know, this person said about this thing. But I think one of the most interesting aspects of it is the sort of coaching or feedback piece that comes from it. It's like the thing that everybody who has it talks about first of like, "Oh, you know, it gave me feedback on like, oh, I really shouldn't hog the conversation so much, um, or I should, you know, make sure to kind of really like let my children finish what they're saying about something before I interrupt and jump in." And I think what's so interesting there is it's like this omnipresent sort of, um... The context is it is omnipresent, and then there's a question of, okay, what, what can you do kind of with, with this data? Um, and then how do people kind of inter-interf-inter- interact and interface with, with the data that it's, that it's, um, sort of creating. Um, one other interesting one I think to think about is, um-Uh, I gave my husband a pair of, um, the, um, Meta AI glasses

    7. AGAakash Gupta37:12

       Mm

    8. ELElizabeth Laraki37:12

       ... the Ray-Bans. Um, which one of the interesting things, I have a lot to say about them that I think is awesome, but, um, one of the things, like, we were in Spain and, um, the menu was all in Spanish. My husband was like, "Oh, I'm kind of curious how they'll translate, uh, the menu for me." And he asked it to translate the menu, which it did, or which the glasses did, but, like, they read it through the menu like a screen reader. What it- Like, it just started at the top of the menu and literally read every single description for every item, and it's like, whoa. And okay, no, that's not how you actually read a menu, right? Like, it's kind of more this, this dialogue that you would expect to unfold. Like, "Okay, here are the appetizers, here are the main courses. Is there anything in particular you're in the mood for, or do you want me to go through, like, the full section of what's here for you?" Um, where there was a lot of opportunity there to make it much more human. Um, so anyhow, those are some thoughts there.

    9. AGAakash Gupta38:14

       So it's all about understanding people's correct context, being approachable, being able to... Like, the ChatGPT voice mode, if you had asked it to analyze a menu, I don't think it would've necessarily

11. 38:25 – 41:52

    Designing beyond chat

    1. AGAakash Gupta38:25

       responded that way, and it would've actually given you, like, you know, "There's 12 appetizers [laughs] and 16 mains that are across fish, beef, and chicken. Do you have any preferences, or do you want me to read all of them out to you?" Something like that-

    2. ELElizabeth Laraki38:37

       Yeah

    3. AGAakash Gupta38:37

       ... it probably would've done. And so all- also about bringing in, like we've been saying from the beginning, the right AI [laughs] and, like, really testing your product a lot around these different actual contexts that users are gonna be using, maybe even collecting the data around, okay, these are the top ways people are con- interacting with it, going out in the field, testing out how it's working, and then potentially sculpting that response as a result. How do you design for AI without chat? Because everybody is just relying on this tired, forgotten paradigm, and the future is supposedly something else. What is that future?

    4. ELElizabeth Laraki39:11

       I think we talked about this a little bit earlier, um, right, which is that a chat gives you... I, I think one of the biggest constraints with chat is that it gives you a linear output, um, and it is difficult to go back and reference sort of in, uh, relevant information that happened before. Um, and it does, I think by nature of being linear, limits the use cases that it can do well. So we talked a bit earlier about kind of instead of having a linear conversation, being able to have an image sort of be the central theme and having a conversation around that image. I think another example, um, is, um, I had tried to use ChatGPT to create an itinerary for my son and I for a weekend in Madrid, and I was actually very surprised in some ways by how good it was at pulling out kind of salient information, but how bad it was at helping me get to what I wanted, which was effectively, like, a Word document, right? That was, "Okay, here's the, here are the things we're gonna do. Here are the timetables. Um, here are options if we decide not to do this or that." And I think the challenge was really the UI, that it was, it, it would kind of create this information in line, and then I would say, "Oh, you know, well, I don't wanna do this thing," or, "This place actually looks to be closed," and it would say, "Okay, no problem. Regenerate the itinerary." But it would also hallucinate a little bit each time, and I think just due to the sort of more dynamic and non-deterministic nature of it makes it really hard to do more deterministic tasks. So in this case, like, I want control over being able to have a fixed body of content, and I want ChatGPT to help me co-create it. And so I think the more that we can think about these experiences where, you know, AI is helping co-create something, I think the more we can sort of position chat as a tool versus chat as the interface. Um, one company that I have seen, um, kind of begin to break away from this is Cove. Um, and this was started by the same duo that did Google Street View and Uber Eats, um, [smacks lips] Anish, Andy, and Steven. But, like, I think what is interesting is that it basically gives you a canvas to work from and kind of pull bits and pieces of different information from. So you could think of it as almost sort of having different

12. 41:52 – 44:49

    AI design tools for designers

    1. ELElizabeth Laraki41:52

       chats or documents that you're working kind of on in parallel, um, versus having one linear, one linear chat box.

    2. AGAakash Gupta42:01

       Very cool. So this is what the future of AI products is gonna look like. We're gonna continue to be shaping it. I think the future isn't totally wr- And whoever's watching this video and building that next generation of AI products, comment below. Let us know what you think it is. Share your tool. We'll take a look at it as well and see. But it seems like it's heading in this direction, like a Cove direction. So we've gone pretty deep on designing AI features. I wanna kind of flip gears a little bit and talk a little bit about AI design tools, because there's all those AI design tools out there. We talked a little bit about AI image expanders. What AI design tools should people have on their radar, and how should they be using them?

    3. ELElizabeth Laraki42:44

       So I think this is a, this is an interesting question. So I am not actually using that many AI design tools. Um, and I probably should be using a lot more than I am. But I think in the same way that I can now do, as an amateur, a little bit of video editing, a little bit of image creation, a little bit of coding, like-I think what I find frustrating with design tools is that my expectations are pretty high, and the output does not match what I want from them. And so I keep not using them and just going back to Figma, not using any of the kind of suite of AI tools around it. Um, so I don't, um... That being said, I have several colleagues and friends who are using them more throughout the design process, both kind of using ChatGPT as far as figuring out product spec, um, using Figma and Figma Copilot to kind of really, um, sort of get a first pass of something and then edit it, um, and then plug it into kind of making sort of prototypes, prototypes and demos. Um, but I, like, keep getting stuck as I try to use these, so.

    4. AGAakash Gupta44:07

       So what would be your recommendation for the average person and-

    5. ELElizabeth Laraki44:14

       No

    6. AGAakash Gupta44:14

       ... the product designer? It sounds like for the average person, you're saying, you know, maybe AI design tools can help, just like they're helping you when, who is below average on video production get up to average. So if you wanna get an average result, yeah, use Figma Copilot, use, you know, your Veo 3 Midjourney Kling workflow to create your design or use the AI prototyping tool design that comes out of V0, Lovable, Bolt, Replit. But if you're a product designer, probably just apply your own taste and where possible, use it for productivity like alignment and those types of things, but don't expect it to suddenly

13. 44:49 – 57:04

    Live design: LinkedIn for AI

    1. AGAakash Gupta44:49

       create the design for you.

    2. ELElizabeth Laraki44:51

       So for designers, like, I think some of these tools can be helpful around the edges with sort of different aspects, but I think that fundamentally we all still have to rely on our own taste and our own kind of processes to really get the output that is at the bar that we expect.

    3. AGAakash Gupta45:11

       Mm-hmm. Mm-hmm. So some potential, but probably the tools aren't there yet, guys, for you to just solve your design. Vibe designing is not quite here yet, but we'll do an updated video if it comes out in the future. I wanna talk a little bit about the design process. I wanna do a little bit of a live design session so people can get inside the mind of a master designer like yourself. Sam Altman recently said he's working on a LinkedIn at OpenAI. So can you break down LinkedIn for us and help live design what a LinkedIn for AI would look like?

    4. ELElizabeth Laraki45:46

       Sure. Um, um, the first that I'd actually sort of heard about this really was through, through your comment about, about kind of, uh, Sam Altman saying they were designing LinkedIn for AI. And then I went and looked up a little bit about it. Um, actually know Fiji quite well from having worked with her, um, at Meta. And, um, I think I would say, like, my top level thinking was, okay, is this actually more, um, uh, I don't wanna say like a marketing piece. So I think this could go a couple different ways, right? But one, I think, is it, like, what's actually the real objective of this? Is it to make it less scary that AI is, you know, eating and has the ability to eat a whole kind of class of, of different jobs across many different sectors? Um, or is it really about the AI kind of education and certification and things like that? Or is it more like some of the, um, uh, dating apps that are really kind of trying to look deeper than like, um, uh, sort of bits and, and pieces of people's character or attributes and matching them? So is it trying to use like more sophisticated sort of matching technology? And, um, so I think, one, there are some questions, right, about kind of what really, what really is the objective of, of doing this kind of LinkedIn for AI app. Um, then I also think about, okay, so there's LinkedIn, and what are people actually really using LinkedIn for? So, um, I've never actually personally used LinkedIn to look for a job myself. Um, I have used it for helping try to either find people that I'm looking to hire or helping other people try to hire for, for different jobs. But, like, there is an entire set of LinkedIn that I have never used and really not interacted with at all, which is the job hunting and job seeking kind of chunk of things. It is sort of equivalent to sort of the dating piece that I mentioned earlier, um, sort of matchmaking piece. Um, I think there is a whole nother piece of LinkedIn, and this would be very biased by my own usage of LinkedIn versus, like, how LinkedIn is actually used. Um, so I will, I will caveat that as a bias. But I think, um, I am seeing and have been seeing LinkedIn be used more and more as just kind of another sort of social feed in many ways, um, where it is about sharing content, it's about interacting with content, it's about kind of connecting with people and sort of building out this kind of network, um, that is very different to me than just the job searching, right? That feels more equivalent to a Twitter or Instagram or Facebook or whatever, but kind of around the context of work. Does that make sense?

    5. AGAakash Gupta48:52

       Yep.

    6. ELElizabeth Laraki48:53

       Okay.

    7. AGAakash Gupta48:53

       Tracking.

    8. ELElizabeth Laraki48:54

       So I think, like, where we really sit with this is at kind of product positioning. So I will also say, like, I didn't even, like, open LinkedIn for a very long time until I was at, um, some kind of happy hour that we were hosting in the city, um, I don't know, a year ago, a year and a half ago. And-Somebody was like, "Oh, we should connect on LinkedIn." And I was like, "Is this person trolling me?" Like, I don't... Because I, like, look old and I don't belong here? Like, I don't... What is going on? But okay, sure, we can connect on LinkedIn. And then somebody else said it too, and I was like, man, either everybody here are really jerks or, like, maybe LinkedIn is starting to, like, grow again. And, um, and then, like, it kind of felt like this wave ... Maybe my timeframe is off here a bit, but I, um, like, and there was this wave of, like, another colleague being like, "Oh yeah, no, my kids who are graduating high school and in college, like, are totally using LinkedIn, and they have, like, their kind of like resumes all set up, and, um, you know, they're trying to connect and network through LinkedIn." I was like, "Oh, this is fascinating, that it really does feel like LinkedIn is, like, beginning to make a, a comeback." Um, so I think all of this leads us at, like, obviously I'm not like ... I am not moving pixels around as we're talking, but I think this is where I had said earlier as far as like, you know, the first thing is really defining the product. Like, who is this for, and what are the tasks or kind of use cases that, that it supports? So I think the first thing I would ask would be, okay, we've got, we've got this idea. Like, which, which direction do we wanna take it? And I think there's some guesswork here because obviously I don't know what Sam and Fiji and everybody is actually thinking about kind of behind this. Um, so I think what I would do, um, and I don't know, I'm like, I can pull out pen and paper if we want. Do you wanna do that? You're not? I don't know how to make-

    9. AGAakash Gupta50:45

       Yeah. I would love to understand-

    10. ELElizabeth Laraki50:46

        Okay

    11. AGAakash Gupta50:46

        ... how you break down a hazy problem-

    12. ELElizabeth Laraki50:48

        Yeah

    13. AGAakash Gupta50:48

        ... like this.

    14. ELElizabeth Laraki50:48

        Okay.

    15. AGAakash Gupta50:48

        'Cause that's the type of problem people are given. [laughs] Yay, oh, LinkedIn Open AI. And you're like, okay, well, this is how an expert will break it down.

    16. ELElizabeth Laraki50:57

        Yeah. Okay. But I think I would just start with, all right, so we've got this idea. Can you see this or no?

    17. AGAakash Gupta51:07

        Yep.

    18. ELElizabeth Laraki51:08

        Okay. We've got this idea of, like, LinkedIn for AI. Um, so we've got a couple options, right? One is about, like, matchmaking. Sorry, I do this in chicken scratch. But one is kind of about matchmaking. One sort of direction is really about, like, certification and training. And a third would be, um, sort of more about, well, I guess you could say, like, networking, and another would be more about, like, content-

    19. AGAakash Gupta51:44

        Yeah. So many different angles

    20. ELElizabeth Laraki51:45

        ... sharing and distribution.

    21. AGAakash Gupta51:46

        So let's say we just we're inside the mind of Fiji and Sam, and they said-

    22. ELElizabeth Laraki51:50

        Yeah

    23. AGAakash Gupta51:51

        ... "You know, all we really care about is matchmaking. Because a lot of people are going to ChatGPT, and they're looking for jobs, and they're trying to apply for jobs via ChatGPT. And so we really wanna break that part of it and disrupt LinkedIn there."

    24. ELElizabeth Laraki52:03

        Yep. So we'll double down on matchmaking. And I actually feel like this could be quite a smart direction. Um, yeah. I'll just leave it at that for now. Okay. So let's say we're gonna go into matchmaking. I think first I would start with the question of, like, what makes a good match, right? So you have obviously ... I'll just put, like, job seekers, and then you have, um, uh, we could put employers. Um, and then I think what you, what you're sort of looking for here is the magic that is between those. To get to there, you need a set of attributes across each to kind of understand, like, what, what matches well with the other. Um, just thinking about sort of UI here as well that you'd mentioned, there's obviously gonna need to be some sort of, like, onboarding or kind of input flow for these, for these pieces. Um, but then I think, you know, this feels still pretty unsophisticated, right? So I think the question is, like, what is, like ... What would be the things that the patterns and things that AI could see that humans or just, like, a very rudimentary, like, match A with A, um, kind of means? Like, what, what goes beyond that? Like, what additional class of things could, could we begin to capture and, and input, right?

    25. AGAakash Gupta53:38

        Mm-hmm.

    26. ELElizabeth Laraki53:38

        Is it like, um, I don't know. Like, just thinking, I, um, had a conversation with someone a couple weeks ago whose job is actually administering personality tests to, um, CE- potential CEOs for hire, um, uh, so that they can actually, the, the company has a much better sense of kind of how will fit into the organization. I'm like, well, that's actually pretty interesting. Like, I'm sure there are ways you could sort of automate that and make it much more accessible beyond it being right now prohi- prohibitively expensive and just doing it for CEOs. You could do that for anybody and everybody, right? Like, and through a series of questions like, you know, um, I don't even ... Things of, like, how do you start your morning, right? Is it more social and talking to your colleagues, or is it that, like, you like to get in and just kind of start doing your work right away? Um, obviously aspects of, like, introversion, extroversion. Do you get more energy being in the office with people, or you kind of prefer to be at home? Um, I think there's a whole set of kind of skills and potential that could begin to be identified as well, and maybe even sort of predicting poten- potential and fit, um, within a certain group, even within a, within a company. Um, and so I think what's interesting when I think about the UI for this is that, you know, you have kind of two separate UIs. You have, and it is ... Yeah, you have sort of, it's a marketplace, right, right? But you have UIs for job seekers, and you have UIs for employers. And then all the sort of magic in the AI is kind of happening in, in the middle.

    27. AGAakash Gupta55:12

        Yeah.

    28. ELElizabeth Laraki55:13

        Um, I think there are opportunities to look at, okay, where does AI fit potentially on either, on either end of this? Um, but I think-You know, I don't know. Then I'm like, there even could be like a Tinder, a Tinder swipe or something along those lines of like, yes fit, no fit. Um, but I would also sort of look at other kind of matchmaking apps. Like I wouldn't, I think Link- I think the only parallels with LinkedIn really would be that LinkedIn is about jobs and work and that you can post and look for jobs through LinkedIn, right? But I would then diversify and look a lot further, um, beyond, beyond, sorry, is it this? Beyond, um, kind of LinkedIn to kind of look more at dating apps and all sorts. I don't know, even think about like college admissions, right? Like all of these things of kind of where are you really trying to sort of find good fit. So I don't know. Those are my two cents.

    29. AGAakash Gupta56:12

        So what I took away from this is you're not just gonna dive into pixels and Figma [chuckles] when you're solving a hairy design problem. The very first thing is actually create alignment on, well, what is the business goal here? Then it sounded like you were going very deep on, well, what is the new technical innovation that we are bringing to market? Then you go deep on, okay, well, what are the different user groups that we need to design user interfaces for? And that's where we kind of ended here, right? And so there's this stepwise process in design. This is a classic design problem that somebody probably created 60, 70 years ago that still applies for AI features. And you can't just all of the sudden... I think what people wanna do instead is they just wanna go into Lovable and say, "Design me [chuckles] LinkedIn for AI." And

14. 57:04 – 1:04:14

    Google Maps redesign story

    1. AGAakash Gupta57:04

       they would've expected us to just open up Figma and then go iterate on that design. But there's this actual classic design process is super, super important first. Um, is there anything else people should know about as they're tackling and building these hairy sort of vague design challenges?

    2. ELElizabeth Laraki57:23

       Um, I mean, I think like the goal I would say is really to emerge from ambiguity with a clear sense of what you're building and for whom, and what you want it to do in the world.

    3. AGAakash Gupta57:38

       Love it. So that is most of our masterclass on AI. You worked not just on Google Search, but actually on Google Maps. In 2007, you were one of two designers on Google Maps, but it was becoming a cluttered mess. How did you guys redesign Google Maps into one of the most loved apps in the world?

    4. ELElizabeth Laraki58:00

       Um, so I think one of the things that I love about this story is it didn't seem, um, I don't know, quite frankly like that interesting or that revolutionary at the time. Um, but similar to your question about like how does, um, why does Google Search still look so much today like it did 20 years ago, I think the same is actually true with Google Maps. So, um, when I joined Google Maps, um, the, uh, the UI looked very much like you're showing here, um, which is that there was at the top a tab effectively for maps, for local search, and for directions. And, um, Maps allowed you to look for one place. Local search allowed you to look for a category or restaurant name, whatever, in a place, so two input boxes. And directions had a from and a to with also two input boxes. And, um, you know, we had a lot of other things that we wanted to fold into the mix of things that you could search for on Maps. Um, we were looking to integrate with transit, so it wasn't just driving directions, but you also had to be able to then figure out like different modes of, of transport for how you wanted to get places. And, um, and also we were bringing in user-generated content, um, as sort of a whole field of content. And basically you very quickly start to run out of space at the top for these tabs, and it just starts to get more cognitively hard to use it because you're showing up and you're having to make all these different choices. Um, so what we ended up... And, and actually I think like it's funny because I was talking with a colleague who worked on Maps, um, after my time, um, and I think Google Maps has gone through this cycle over and over again of be- trying to kind of add more and more features, and then it starting to feel too crowded and too cluttered, and having to go back to core principles and kind of design a UI that works well for core principles. Um, so what we were looking at is really what are the use cases for Google Maps? Well, there is like trying to find places. There is, um, trying to, um, get directions. And then we were kind of, and I feel like every project starts to stretch a little bit, where it's like, "Oh, we should really be a place for discovery. Oh, and also wouldn't people wanna create?" But, and one of the things that we did was sort of look at really what's the, what's the model for this. Um, and so this is an example of diving in to get directions, but, you know, you have kind of the four main categories. And then within directions, um, you know, you have sort of the starting point. You have the destination, how you're gonna get there, any specific kind of elements of the route. Um, and then how does that play in, right? So then you start with directions. You get the route. You have it in a list and a map. How do these things work? What are the actions that sit over all of these different use cases versus what are the actions that are specific to sort of a specific use case or one use case, and how do they play together? And so this is where I do talk quite a bit about like design architecture, but it's kind of a you have the user experience for the product. You have the UI people are looking at.How are all the, all the parts and pieces organized? And then how do people move through them, and where does it make sense to allow people to go from one path to another, for example. Um, and, um, I don't know if you have the kind of, um, next UI in this, but basically, okay, the TLDR is that we went from effectively three tabs with a total of five different search boxes to one single search box that... I mean, this, like, seems very like a no-brainer at this point, but at the time, it was highly controversial because Directions was the bread and butter for Google Maps. Um, obviously, we were inspired by Google Search that had a single search box that worked for everything and felt like it really could work for Maps, but that also we were probably gonna have to teach people what they could search for and how they could use Maps in the same way with only a single search box instead of, for example, two with directions.

    5. AGAakash Gupta1:02:20

       Mm. Very interesting. It strikes me that it's one of, like, the most used probably AI backend products out there now, and they've been ex- iterating with the design a lot. One of the most recent designs, you actually said you didn't love. Why?

    6. ELElizabeth Laraki1:02:42

       Um, I don't love it because I don't feel like they focused on what matters, right? So I think there are sort of visceral just responses that, like, Google was always kind of nerdy and kind of quirky and sort of, I mean, even from the logo that was all primary colors, right, that looking at the old versus the new, um, the new, I mean, one, at a glance just looks like Apple Maps, and two, like, just feels colder and less more, less real, right? It looks much more digitized and, um, [lip smack] uh, than kind of representing just, I don't know, more human colors. So that was a visceral response that I had in just kind of looking at one versus the other. Um, but then it's like, okay, now when you start looking, oh my God, there's so much crap all over the screen, and what are all these things? And I've just ignored them for so long, but now that I'm actually looking, like, why are all these things here? And if you're gonna do a cleanup, why don't you actually, like, really clean things up?

    7. AGAakash Gupta1:03:43

       Mm-hmm.

    8. ELElizabeth Laraki1:03:43

       And as I had mentioned earlier, that I think Google has, um... One of the things Google Maps has done extraordinarily well is it keeps purging features, um, and keeping things clean, and I feel like they're very, they're very overdue, um, for a round of that at this point, um, that, you know, I don't know the usage of any of these things, but you have, like, work, you have pills for, like, work, restaurant, gas, parks at the top, plus the weather. And, um, then you have Explore and Go

15. 1:04:14 – 1:10:09

    Google Maps India landmarks

    1. ELElizabeth Laraki1:04:14

       and Saved and Contribute and Updates, and then also, like, another sheet that's, like, between that and the map for, like, latest happening in this area. And predominantly, I use Google Maps in the area that I know, which is where I live, and I'm trying to use it to, like, beat traffic, um, or avoid traffic going between a set of common routes that I often do. Um, or I'm leaving the area and going somewhere, but that feels like a less common use case. Again, I'm a user of one, but things feel very crowded and cluttered to me on Google Maps today.

    2. AGAakash Gupta1:04:50

       Yeah, and that was the key takeaway. It's really full circle to where our discussion started at the beginning of being able to simplify some of these things. Get-- Look at the magic of ChatGPT of making things relatively uncluttered. One of the most phenomenal things you did in your tenure on Google Maps was you helped launch Google Maps in India. And if people haven't been there, I've been there. I go there every year. India doesn't use street names. Normally, the way it works in India is you just ask somebody on the side of the street, and they say, "You go three gullies this way. Then you turn left. Then you turn right at the temple." And you're like, "Okay, there we go. [laughs] That's how I'm gonna get there." How did you guys manage this innovation? I think it's gotta be one of the world's most important.

    3. ELElizabeth Laraki1:05:34

       So, um, I think this was 2007, 2008, kind of somewhere in there that Google did launch. So Google was looking to expand internationally. Obviously, India was a very attractive market as far as, um, as far as number of people. And, um, and Google Maps did launch in India. And I think it was, like, I don't think any designers were involved at all at that point. It was like you turn a few switches, and you just expand the, the geo space that, that Maps is available in. Um, and, um, but it wasn't, it didn't have a lot of, a lot of usage. And one of the researchers, one of the user researchers on our team started to look into why. And she was like, "Oh, it's because the directions look like this." I mean, here she is sitting in Mountain View just trying to get directions from A to B and is looking at the UI, and it's like, oh, yeah, turn left at NH 17 in 11 kilometers, turn left in point seven, turn left in point two. And this is also, like, before people had, you know, geolocation or geo, um, uh, geo-enabled, um, smartphones. And so unless you had, like, a compass and a pedometer, right, that, like, these are totally useless directions. Um, and so from that, she-- So, yes, also, she was originally from Russia. She knew and, and I think kind of the team was aware that, you know, obviously, different parts of the world navigate in different ways, that we do it very much in the US by turn, through turn-by-turn directions, but other places in the world do it much more kind of, um, landmark based, where, um... My husband's from Morocco, and we laugh about this too, but it's kind of like you have, like, a rough approximation of where you're going, and then you have human GPS when you get closer, right?

    4. AGAakash Gupta1:07:28

       Mm-hmm.

    5. ELElizabeth Laraki1:07:28

       Of kind of you ask the person on the street to go here and then the person there, and eventually you get, you get to where you're going. And, um-Uh, and so this kind of unleashed this question of, okay, well, actually, like, how do people navigate in India? And if people are navigating through landmarks, like, one, what landmarks are important? And two, how do they actually use them in, in the navigation? And, um, so Olga, who's the researcher who worked on this, um, grabbed the designer who was working on directions at the time, Janet, hopped on a plane, they went to India, and they started to just do a bunch of research to begin to understand these parts and pieces. Um, and obviously was working with the engineering team as well to understand, okay, what was possible, what landmarks did we have, what landmarks could we possibly grab and pull in. Um, and ultimately, they ended up launching directions that were building and launching direction systems that were much more landmark based. Um, as you see here, like, take the first right towards Arabic College Main Road, um, pass by UNA Cycle Traders on the right. And so also, I think, like, some of the interesting things they found as far as, like, what landmarks were useful and interesting was, like, something like the Empire State Building in New York is not a good example, um, because you're walking by on the ground, and it doesn't stand out from everything else, right? So it's really, it really was these things that were prominent and noticeable from the streets that were good landmarks, and that included things like temples, petrol stations, at the time, Big Bazaar, and kind of things that were, that were sort of highly noticeable.

    6. AGAakash Gupta1:09:05

       Yeah.

    7. ELElizabeth Laraki1:09:05

       And so they ended up rewriting maps to integrate directions, and it wasn't just sort of turn by turn, but it also was, you know, oh, also landmarks are used for verifying that you're still on the right path, right? Of like, "Oh, you'll pass this on your left," or like, "If you hit the roundabout, you've gone too far." Um, and I think, yeah, did, um, a really wonderful job kind of integrating landmarks back in, into directions. Um, and one thing I learned recently is actually that landmarks are available everywhere. Like, this wasn't a feature that, that Google built just for India, um, but it's only turned on in locations where it makes sense to be using landmarks to navigate, so...

    8. AGAakash Gupta1:09:52

       I think it's a fascinating story, and wanted to end on that because it is so timeless. It goes back to the key lessons that we've emphasized throughout this episode for you guys, which is the core user research, problem discovery, solution discovery process has not changed

16. 1:10:09 – 1:12:00

    Where to find Elizabeth

    1. AGAakash Gupta1:10:09

       for AI features. We did go through a bunch of stuff that did change, whether that's working with your AI researchers, understanding edge cases and pitfalls, building the UX to accommodate that. But fundamentally, these core problems are what you're gonna be doing. Most of you guys are gonna be building, you know, an application on top of a model company. So deeply understanding, just like we left you with, with that Indian user, understanding how they're doing it today, bringing it in not as sprinkles on the cake, but baking it into the cake is the key lesson from today's episode that you guys should walk away with. Before we go, one final question. What, what are you focused on today? What is that business? And if people wanna learn more, how can they help you?

    2. ELElizabeth Laraki1:10:52

       Sure. So I'm a design partner with Electric Capital. Um, we have about a billion dollars, um, of investments, or invested. Um, and I also am kind of really trying to tell the story of design and how, how we all can design for humans and for people, um, and really design products that resonate and work well for people.

    3. AGAakash Gupta1:11:16

       And where can people find you? Is Twitter the best place?

    4. ELElizabeth Laraki1:11:19

       Um, so a couple different places. Substack is probably where I enjoy writing for, um, the most, um, although I tend to try to use both Twitter and, um, and LinkedIn as sort of pointers to Substack articles.

    5. AGAakash Gupta1:11:35

       Amazing. And what is that Substack called?

    6. ELElizabeth Laraki1:11:38

       Um, so I am just Eliz Laraki, E-L-I-Z L-A-R-A-K-I, on all three places.

    7. AGAakash Gupta1:11:45

       Awesome. Find her there. She has the most amazing storytelling posts out there. It's a really worthwhile follow. Elizabeth, thank you so much for being on the podcast.

    8. ELElizabeth Laraki1:11:57

       Thank you, Aakash.

    9. AGAakash Gupta1:11:59

       Bye, everyone.

17. 1:12:00 – 1:12:37

    Outro

    1. AGAakash Gupta1:12:00

       So if you wanna learn more about how to shift to this way of working, check out our full conversation on Apple or Spotify Podcasts. And if you want the actual documents that we showed, the tools and frameworks and public links, be sure to check out my newsletter post with all of the details. Finally, thank you so much for watching. It would really mean a lot if you could make sure you are subscribed on YouTube, following on Apple or Spotify Podcasts, and leave us a review on those platforms. That really helps grow the podcast and support our work so that we can do bigger and better productions. I'll see you in the next one.

Episode duration: 1:12:47