Transcript

Voiceover: The general population don't necessarily know how to interpret pressure. In fact, I didn't either. And so for each of those data blocks, you can actually click through and if we have an hourly forecast, we'll show it there as well. But then we have some text that helps explain it to you.

Rachid (00:06.966) Kristi, wondering, ever since starting work on a dedicated Weather App for Pixel, did your relationship with the actual weather change?

Kristi Bradford (00:17.123) I grew up in the Midwest. So I'm used to ice storms in the winter and craziness in the summer with thunderstorms and tornadoes.

I would say that working on weather, since I moved to the West Coast where, we get maybe five days of rain in San Francisco and everyone freaks out and then we go back to like perfect days. It was more like almost restarting my relationship with it. But there are definitely some points where I feel like it's kind of just muscle memory. You know, you check the weather in the morning and then you go out. Maybe you check it before you start to drive home or something. You don't really think too much about all the different numbers that you're looking at.

And once I knew where the data was coming from and then had access to meteorological experts inside of Google, it was pretty fascinating to assess, yeah, you're right. I actually didn't connect the humidity level and the wind speed with the temperature to understand how it will actually feel like. And in retrospect, I was like, that's probably why I end up on these hikes and I'm like really cold.

So I think there is reconnection with it just from experiencing it as a kid and all of those feelings. And then there's also just learning new things and that was really fun.

Rachid (01:40.768) So avid listeners of the Made by Google podcast will know you were on before, talking about many of the other Pixel Essentials apps, which include Clock, Calculator, Recorder. Before that conversation, you told us, well, we're working on a weather app. We're just not ready to talk about it. We invited you back. It's here now. It's a dedicated app for the first time for Weather on Pixel. What prompted the change?

Kristi Bradford (01:56.293) I did whisper it, yeah.

Kristi Bradford (02:05.55) So we did some studies. We knew that in the market for premium devices, we were the only one out there that didn't have a native pre-installed weather application.

But before we wanted to make an absolute yes, let's go for it, we did some studies and we found out that 82 % of users check the weather at least once per day. So that's a huge percentage. And then it ranks third. So just behind some of those productivity tools, like Gmail or your email services, it ranks third as the most used and most important application on your device.

So you're like, we've probably waited long enough, and it's time to really give that out of the box premium experience to our users so that something that you check this many times, at least during the week and definitely during the day, is ready right when you set up your phone.

Rachid (02:59.508) I think anyone working on products at Google right now is finding very new, useful ways to leverage AI for our users to make everything better or easier to use. Is there AI in the weather app? And how did that come about? What's your sort of insight where it would be useful to use AI?

Kristi Bradford (03:16.214) Yeah, I would say actually the AI, besides the beautiful interface that we've defined as uniquely Pixel, the AI is really our differentiators. And if I take a second to brag about the UI, AI is included there too. And so it's a little bit subtle, but there's 300 plus major cities across the globe where we actually use generated imagery. So landmarks, et cetera. If you look up Seattle, you've got the Space Needle. And all of that was generated AI imagery which helped us keep our costs down but also helped us tile it in the way that we wanted to.

But on top of that, the features in the weather app that include AI, probably the biggest one, and it's at the very top because we really feel like it's that moment, like I talked about, where I wasn't connecting different data points to an overall how to prep for the day or what you're going to experience. And that's what we tried to do with the AI weather report, which is really taking all of these points of data and trying to condense for you what's the key takeaways,

How do I prepare for it? All that I need to know is really glanceable when I'm in a rush in the morning. And that one just takes, again, all those data points and it summarizes it for you.

Rachid (00:02.83) And something else that's new is the weather map. I'm betting there is some AI in there as well.

Kristi Bradford (00:12.835) Yeah, you bet correctly. The weather map, it looks like a traditional radar map, which I think is great. It's something that people know and they know how to read it. But the model and the forecast six hours out for precipitation. That's actually an AI model. So it's built on historical data provided from some of our partners. And we actually just shipped this in August, we shipped it to the United States and to Western Europe countries. And then we just recently updated it to support Japan, which is perfect because it's the tail end of the typhoon season. precipitation map like watching these precipitation layers come in is actually pretty important. But essentially what it does is that it supplements the infrastructure on the ground. So any radar system that you have on the ground and ability to catch precipitation layers, plus this model that's trained on the historical data, is giving us hyper accurate and the ability to forecast out further than we would normally be able to.

Rachid (05:07.316) Right, and that's very useful, a place where I live near Amsterdam where it's raining all the time. So that's definitely useful there. Just coming back to sort of the AI summary, as you mentioned, the AI weather report.

So that's like a couple of sentences at the top of the app, that summarizes everything that's in the graphs and all that. How does that get generated?

Kristi Bradford (05:13.869) Yeah.

Kristi Bradford (05:33.302) It's interesting. It's the core function of LLMs right now. Which is taking this input that can be a short form or long form and in this case, it's about medium form. So all the data points that you can see across your app is what it's taking into account too, but it's including all of those hourly forecasts and daily forecasts as well. And it's basically weighing all of those and trying to talk like a meteorologist would.

So it's trying to give you a summary to say, it's going to be a cold one out there but the rain will stop around 5 pm and then when you expand it it actually has the hard data bullet points so that you see exactly where the source of the data for that summary came from.

Rachid (06:22.163) And am I right that this is generated by Gemini Nano on -device?

Kristi Bradford (06:26.958) Yeah, we wanted to keep it on the device. We wanted to give that flexibility for the users and reduce the latency of it. And then because we know 82 % of people are taking it once a day. And then we've got some power users who check it multiple times, especially in areas where the weather gets a little bit crazy or it's constantly changing. It meant that we were going to be running this multiple times a day.

Rachid (06:55.166) Now, Kristi, I know there are certain jobs that from the outside might look simple, but they are not. And I imagine people might think, well, what's so hard about a weather app? You probably have some sort of weather service that will tell you the weather you put up, like the current temperature, maybe the temperature for the next day, and you're done. So what's your experience of actually building a weather app? And was it perhaps less or more challenging than you expected upfront?

Kristi Bradford (07:21.248) Yeah, And I agree with them. There's a lot of examples out there. There's a lot of third party sources that you can also download. But more data when we embarked on this journey was that over half people prefer the installed version on their phones. So they don't want to have to go out and try to find a third party source. And I think what was easy and what we anticipated was easy is we're powered by Google weather data.

So the same data that powers some of those surfaces that you know and love, including the frog, the data was accessible and we had to kind of plug and play. But then there were other places where with the weather map, that's actually brand new. And this became quite a bit of a conversation of how do we articulate or should we articulate that part of it is modeled forecasting data? And then how do we allow people to kind of experience something that they know and love because if you have examples that you're going off of like radar maps, you don't necessarily want to break the mold with what people understand and how they visualize it.

But I would say probably the hardest part was figuring out how to say no. So if you're basically designing a toothbrush, there's core functionality in it. And yeah, you could say, when I brush my teeth like the toothpaste gets everywhere. So I want to include this like huge visor and then you can keep building on top of that.

For us, those examples were, trying to figure out where we kind of curate the features to make sure that functionally this is delivering exactly what you expect. And it's like centered on glanceable data. And we're not going too far. We're not like reducing too much data to make the UI super simplistic and beautiful. We're giving you just exactly what you need. And I think that's probably the piece that I agree with from the surface level. It's pretty obvious. But once you start getting into the details and you have people giving opinions about what they'd like to see, you really have to come back to a principle of what does this app need to do functionally for everybody? And I think it was that curation process that really was difficult.

Rachid (09:50.88) Well, let's talk about design then for a minute. I think the Pixel Weather app is strikingly beautiful, which, as you mentioned, there are so many examples, yet it is very much a Pixel app, right? How did the design process start? Where do you start designing a weather app where, as you mentioned, there are so many examples and still you have probably all sorts of core ideas in there? How do you get from sketches into an app?

Kristi Bradford (10:16.526) You know what's interesting?

I was reflecting on this the other day that we had the benefit of we are zero to one. So once you have an app out there, you're kind of iterating on lots of loud voices and lots of data, but you have to maintain what you've established. And so I think the benefit to the design was actually learning from other applications that are already out there and iterating, like taking small incremental iterations and learning from those key pain points.

And so that's why we knew that precipitation was one of the most important data pieces after temperature. We wanted to keep that top of the line, like very top of the page. And we also knew that people wanted some sense of customization. So we didn't want to put users in a place where they could either hand their phone off to like their niece or nephew and get into a tricky spot with the glanceable data. But we wanted to make it feel like yours.

And so we ended up applying customization further down the page with all of the data blocks so that for me, again, San Francisco never rains. And so I don't need that precipitation block. And I understand that it's placed at the top because most places in the world do experience that. But the UV index for me is probably the biggest one. Like the potential skin issues that I could face with the sun in California, I want to pull that up.

So that was really the benefit of we had examples out there, even Google examples, and we had core pain points that were provided from user feedback. And we could really iterate on it for our first showcase when we launched the V1.

Rachid (12:03.552) So you're saying just to be sure, I can sort of long press on a data block like the UV level and move it more upwards, or in my case, in the Netherlands, probably more downwards, right? And so everyone can see what they want to see and be their sort of own super user.

Kristi Bradford (12:19.47) Yeah, exactly. Even the, I guess for you, the weather map, you can pull that all the way up to above the 10 -day forecast and jump straight into it.

Rachid (12:28.266) I absolutely did, yeah. So there are many data blocks. It can get sort of complex, but also interesting. How do you sort of help people curate that and understand what they're looking at? Apart from designing it in a way that it looks really nice, how do you make sure people actually can understand what it means?

Kristi Bradford (12:50.582) Yeah, I think that was another thing for us that we kind of realized, because none of us came from that background, but there was this huge benefit that Google actually has some meteorologists in-house. And we were leaning a lot on their experience as well as user feedback. And people don't necessarily, and this isn't applicable to everyone, but the general population don't necessarily know how to interpret pressure. In fact, I didn't either. And so for each of those data blocks, you can actually click through and if we have an hourly forecast, we'll show it there as well. But then we have some text that helps explain it to you.

So that example of pressure, it ties it back to what are the symptoms that you would experience in high pressure or low pressure situations. And again, it comes back to curation where getting these crash courses in meteorology from our in -house experts, we wanted to include as much as we could there.

We basically wanted to include whole YouTube videos to explain weather but acknowledging that this is an everyday use case very glanceable we wanted to provide education and keep it small so that you could interpret it and then move on with your day.

Rachid (14:09.35) So far we've been talking about what you see in the weather app, but then there's also a part of how the weather app feels, quite literally. When I opened the app, when it was really, really raining quite hard over here, the haptics play a role in the phone. So it sort of feels like it's raining. It's hard to explain, but it feels like you feel the drops in your hand on your phone. How did that come about?

Kristi Bradford (14:16.12) Yeah.

Kristi Bradford (14:34.495) You know, I'd like to say that I came up with this idea, but obviously I didn't. Haptics for Pixel are one of the most powerful, immersive feelings that when you switch to a different phone, you can feel the missing of them or the gap in the experience there. And we had this new technology that could basically take your animations on screen and match the haptics, the vibrations that you feel to that. And we thought, okay, this is a great opportunity, not only for accessibility, so you can like kind of immerse yourself in it, but also for that delight of, again, I'm San Francisco and I really miss rainstorms, not to the degree of like the Netherlands, but whenever there's rain showers or rain events you will actually feel a haptic vibration that matches the density of the rain in the current conditions

Rachid (15:34.601) Right, so if it rains harder the haptics are... well, you feel them more, Wow.

Kristi Bradford (15:39.135) Exactly. Yep. And there's actually a fun little Easter egg in there too.

Rachid (15:45.439) Really?

Kristi Bradford (15:46.261) Yeah, so one of the other things that is beautiful about rain if you don't have to commute anywhere is the sound that it makes. And so if you find a place like right now as we're recording Taos, New Mexico is raining. If you find a place like that, you press on the icon for the current condition next to the temperature, you'll get this beautiful soundscape of the rain.

Rachid (16:12.085) Really? Wait, can we hear that? Let me try that.

Rachid (16:31.274) Amazing. Okay, that is a cool easter egg. And I guess you heard it here first at the Made by Google podcast. That is really nice. Coming back to design, and you already mentioned the frog, which by all accounts, I think is named Froggy internally.

Froggy did somehow not make the cut in the Pixel Weather app. Is he on vacation? Is he okay? And just for people who don't know, maybe people have seen the frog in our weather reports, right? When you search for the weather on Google search, or maybe you have a Nest Smart Display, you can also show the weather on the screen rather than pictures, for example. And I will feature Froggy, the weather frog, I suppose.

Rachid (17:28.053) But not in the Pixel Weather app. Why is that?

Kristi Bradford (17:31.224) I honestly hope that he is on vacay. He's been working hard for many years. He deserves a vacation. It was a tough decision, because as a character you see basically half the population loves him and half the population has a different feeling of him. And we didn't want to only cater to one side. So we kind of took Froggy out of the equation for a second and we just thought, what is the Pixel design aesthetic? What are we aiming for? And it came back to those principles of we want it to be glanceable and we want it to be beautiful and immersive.

And so in that case, that's why we chose to make the cityscapes a little bit more subtle in the background, the conditions, this beautiful animation, and keep that data front and center. So there was truly a discussion about it and we ended up choosing to go down a path that evokes a little bit more immersion into the weather itself and this like feeling of beautiful and premium. So yeah, I hope Froggy's enjoying his vacation. You never know what the future holds.

Rachid (18:44.627) I'm sure he's doing fine, for sure, for sure. Let's touch upon the, and you mentioned it before as well, the precipitation map as well. I guess there are many data sources that have some sort of a precipitation map. Is there anything we do differently?

Kristi Bradford (18:59.7) Yes, so if you're in a place where I don't know if you've ever seen something called now casting, but it's this little insight and it gives you a forecast for when the rain will start and stop down to the minute. It's actually pretty incredible and extremely accurate. It's that same model that we use in our precipitation map. And so what that does is that it takes in this infrastructure data. So these radar towers and everything that we get from licensed data providers, and then we build on top of there a model to forecast out. the model is Google built and that's kind the magic of it is that again this is Google data and it connects to another experience that can tell you down to the minute start and stop of the rain in your area.

Kristi Bradford (20:36.215) But I think this ties back to learning from other applications and pain points. So the one thing that is a little unique to our team. And so we did build in some things that would have application towards us, but our team, we have team members all across the globe. So we've got team members like you in the Netherlands. We've got team members in London, in Israel, in Taiwan. And then all of us also have family members or friends.

And one of the big drawbacks that we were seeing just as a team is that some weather applications give you sort of like a cut of how many locations you can save and look through. And so in our list view, we really wanted to give you as much freedom as possible. And today there is not an upper limit. I think we roughly track towards 30 or 50, which should be hopefully way beyond how many saved locations people need. But in general, this gives me the ability to kind of like check in on some of my family in Tokyo, my family back in the Midwest, friends in Bangalore.

And do all of that without having to delete one location in order to save another one. And then you can see it all on the list view, and I can see exactly how different the temperature here in San Francisco is versus my friend Deepa in Bangalore.

Rachid (p2 05:01.898) Were there any other considerations given that everyone uses a weather app, no matter where you are in the world, no matter where you come from, no matter what your job is, have you been thinking about how different kinds of people will try and use the same app?

Kristi Bradford (P2 05:21.109) Yeah, we've definitely thought a lot about that. I think that the next level for weather is probably twofold. And this is the weather in general. This is my hypothesis is that even more accurate and global forecasting, again, using models instead of, or I guess in supplement to infrastructure. But on the top of that is personalization because today you and I see the exact same report. However, you might be a sailor or you might have certain conditions that make you more susceptible to low pressure and high humidity. Whereas for me, high humidity is the most amazing thing that I can go for. It just means that my skin feels nice and fresh and not super dry. So I think that's the thing that is really digging into now that the data is accurate and continues to get more accurate. What do each individual user, what do they need that's different from somebody who's right next to them?

Rachid (P2 00:17.597) What about the future of the Pixel Weather app? I know it's been out now for just a couple of months, but perhaps there are already new features in your mind that you'd love to work on.

Kristi Bradford (P2 00:26.774) Yeah, we already have that haptics, like feeling, evoking sort of an immersive feeling of the weather outside for when it's rainy and that really cute, audio Easter egg.

But we also wanted to include some more impactful data. And so we've actually launched a pollen data block. And for the first version, it's only available in a couple countries. in the UK, Italy, France and Germany. But this is right aligned with some of the seasonal effects of the weather climate and how people prepare for their day. So that data will also show in the AI weather report.

But the block itself is pretty cool. It separates into like three of the most common categories for pollen. And then it gives you a very simple and glanceable gauge of how bad the pollen index is for today. And on top of that, for a lot of those pieces of pollen, so there's grass, tree, and weed, when you click into it, you actually get like the four day forecast.

So you can see if let's say your camping trip, there's going to be a high level of pollen. And if somebody in your family or you yourself suffers, it's great to kind of know that ahead of time. And it also gives you those cross reactions or the types of plants that produce this type of pollen. So that was one level where personally, I don't suffer from pollen, but I've got quite a few friends who do, especially in the Seattle area around the time that the trees are producing a lot of pollen. So it's been very hard to hold it in and not tell them that, this is coming and it's coming for certain countries. And then after that, we hope to expand very quickly.

Rachid (23:50.294) Kristi, can you explain to us what it's like to sort of test the Pixel Weather app? Like, are you actually testing if the information that is shown is accurate, given that it's usually a third-party source? Are you testing with users to see if they like the customization? And what about testing maybe the AI Weather Report? Was it always accurate? Or how do you test that?

Kristi Bradford (24:09.057) It's a good point. So we do test the accuracy of the data. A big portion of that is through the data provider team inside of Google. And then there's another portion with dogfooding. So if we've got people in Berlin and we're reporting a different number or different condition than it actually is, that's great feedback. And that's the type of user-generated feedback that we'd love to open broadly.

And then for the UI, we love to get feedback on where people wanted to place blocks or what additional types of data they want and all of this feeds into how we prioritize further down the road. And then the AI weather report, because it is AI generated, and so with that you accept some level of, let's say fun or adventure, we had the opportunity to essentially, let's call it like a tune the model to have a person.

Kristi Bradford (25:14.276) So we wanted the AI weather report to be informative first and foremost, but we also wanted it to be engaging. Otherwise, kind of what's the point? And so let's pretend like we have this knob and at first we put the knob pretty cold, like very little personality. And we quickly realized from feedback and from our own gut that it wasn't engaging enough. And so we turned the knob all of the way up and kind of saw within there some very interesting scenarios where the personality might have gone a little too far.

So maybe think about like you're the most fun friend that you have who might be a little bit loud and sometimes says interesting things.

Rachid (27:32.916) Right, and I presume with every update there's maybe a change to the tuning, everything you learn, all the feedback you see to make, let's say, the personality a little bit more consistent with what you had in mind.

Kristi Bradford (27:44.255) Absolutely, and if you even look at your AQI, the strings in different regions.

Rachid (27:50.143) And that's the air quality index, right?

Kristi Bradford (27:52.533) Yeah, see, I'm pretty far down the rabbit hole. But the string in different regions is based on some regional definitions. So the strings can say low health risk. And if you feed that into an LLM, sometimes it'll start to interpret that as low health risk overall or health associated with cold and flu season. So we kind of have to pull it back, basically tell the LLM, no, sorry, that was just specific to AQI. So that's some of the tuning that we've been able to do and like quickly iterate on it.

Rachid (28:32.278) Kristi, what I'm curious about, apart from working on the Weather app, you and your team also work on the recorder, you work on clock, calculator, and I'm sure we can talk about all those for many, many hours, and people who interested should definitely listen to the previous episode when we had you on talking about all these apps. I'm just wondering, what does it feel like to sort of be the author of these essential and core apps used by millions of people multiple times a day?

Kristi Bradford (28:59.593) You know, I was about to say that it's a lot of pressure, but I feel like the feeling that's above that is actually quite a bit of pride.

Just a factor of considering how far technology has gotten. I don't have a physical alarm clock next to my bed anymore. In fact, I think the one analog clock that we have in the apartment is just for design aesthetics. I'm not sure that the batteries are working. And then similarly, when I go into some vintage shops, I'll pick up tape recorders. But it's just kind of a testament to how many things that you used to have to have, like how many pieces of like physical devices or experiences that you would separate between in order to achieve like everyday tasks of setting an alarm, checking the weather, recording sounds or jam sessions that all fit now on this device that you just have with you all of the time.

So I'd say it's really, really fun. And there's always something new coming and always an opportunity to evolve the technology with Google. And then they raise a lot of responsibility. Like you really can't allow a scenario where the quality would degrade and you would not be able to show someone their weather because they're depending on that. They don't have anything else because it's now known that it's inside of the phone that's in your pocket.

Rachid (30:27.84) Kristi, thank you so much for joining the Made by Google podcast. If there's ever another update from the Essentials team, please come back.

Kristi Bradford (30:34.908) Yeah, yeah, there's a whole fun backlog of things I'd love to build. So I hope that I will be back soon.

Rachid (30:43.615) Amazing, thank you.