In short
Google AI: Release Notes - Episode Summary
Episode Title
Gemini in Workspace: New Ways to Create Faster Host: Logan Kilpatrick Guest: Chandu, Google DeepMind Team
Episode Overview In this episode, Logan Kilpatrick and Chandu discuss the integration of the Gemini AI into Google Workspace, highlighting its impact on productivity tools such as Docs, Sheets, Slides, and Drive. They cover various features, rollout schedules, user privacy, and the future of AI in Google’s ecosystem.
---
Key Topics Discussed
- Gemini Integration in Google Workspace
- Introduction of Gemini:
- Gemini is being integrated into various Google Workspace applications to enhance productivity.
- The integration aims to provide a more AI-native experience across Docs, Sheets, Slides, and Drive.
- New Features and Functionalities
- Gemini in Docs:
- Introduction of a bottom bar for content generation directly in the document.
- Allows users to pull contextual information from their Drive and Gmail.
- AI Overviews in Drive:
- A feature that helps users find files using semantic queries rather than exact file names.
- Users can retrieve content based on abstract queries.
- Sheet Builder:
- Aimed at automating spreadsheet creation, such as generating P&L statements.
- AI-Powered Slide Generation:
- Users can input ideas, and Gemini can help structure and design slides visually.
- Rollout and Availability
- The rollout will start with A-Ultra users, followed by A-Pro and private commercial preview customers.
- Feedback from users is being actively monitored to improve the process.
- User Data and Privacy
- Emphasis on user privacy, clarifying that Gemini does not store or learn from individual user data.
- Data used is strictly for generating content based on user prompts without retaining any information for future use.
- Balancing Innovation and User Trust
- Discussion on the challenge of balancing cutting-edge AI capabilities with maintaining a reliable user experience.
- Google aims to respect users' preferences while integrating new AI features into familiar tools.
---
Key Takeaways
- AI as a Tool for Productivity: Gemini is positioned as a tool that assists users in compiling information and generating content, allowing them to focus more on directing outcomes rather than just gathering data.
- User Control and Transparency: Users will have control over how AI features are integrated into their workflow. They can choose to enable or disable these features, ensuring a balance between innovation and traditional usage.
- Future Developments:
- Continuous updates and improvements will follow as user feedback is integrated into the development process.
- Future plans include enhancing interactivity in tools like Slides and further integrating AI capabilities across the Workspace suite.
---
Conclusion The introduction of Gemini into Google Workspace marks a significant step towards enhancing user productivity through AI. Logan and Chandu's conversation reveals an ongoing commitment to innovation while maintaining user trust and privacy. The episode sets the stage for further advancements in AI technology within Google’s productivity ecosystem, promising exciting developments in the future.
--- This summary captures the essence of the podcast episode, presenting the main discussions and valuable insights related to Google AI and its integration into workspace tools.
Written by AI. May contain mistakes. Listen to the episode to check what was said.
Chapters
Tap a time to open that second in VOGemini Rollout Schedule
0:46 to 2:29
Discussion on the rollout schedule and user expectations for Gemini.
“Getting you out of meetings is hard because you are a very popular guy.”
New Workspace Features
2:30 to 3:17
Overview of new features in Google Docs and the integration of Gemini.
“I see that every time I do docs.new and I open up a new Google Doc to get started, I like see that experience.”
AI Overviews in Drive
3:18 to 4:16
Introduction of AI overviews in Google Drive and their usefulness.
“So some of these content generation is Flash.”
Customer Rollout and Feedback
4:17 to 6:06
Exploration of customer segments receiving the new features and their feedback.
“And the problem with some of the Drive corpus is, and I love my Drive and I have everything there, and it's often not organized, even though I would love to organize it a little bit more.”
Gemini's Impact on Productivity
6:07 to 8:05
Discussion on how Gemini impacts productivity and user experiences.
“No, I think on day one, you will see all of our A-Ultra users get it.”
The Future of Workspace Features
8:06 to 10:05
Insights into future developments and integrations of workspace features.
“Because I feel like we've made a huge amount of progress.”
AI in Document Creation
10:06 to 14:03
How AI is changing the document creation process and enhancing user output.
“or like you'll build it all integrated into the mainline version of the experience?”
Changing Content Creation
14:03 to 14:50
Explore how AI is transforming the way we create and proofread content.
“I think is going to change significantly and we are just getting started on that long journey.”
AI Integration in Workspace
14:51 to 18:40
Understand the thoughtful integration of AI in Gmail and other Workspace products.
“Sometimes I'm making, sometimes I'm not making sense.”
User Control and Transparency
18:41 to 21:26
Learn about the importance of user control and transparency in AI features.
“And in fact, even when you select your sources or whatnot, are all never going to cross the boundary of your training because training the models is in a completely separate process.”
Show all 18 chapters
The Power of AI in Workspace
21:27 to 24:26
Discover how AI can enhance productivity and collaboration in Workspace tools.
“So how do you give that control back to the users is the second one, right?”
Journey to Current AI Capabilities
24:27 to 28:03
Gain insights into the advancements in AI and the challenges faced in implementing them.
“vids and videos, if we can bring these abilities to the humanity in both their productivity and collaboration, I think that's a fantastic opportunity.”
Collaboration Behind Gemini Development
28:03 to 28:31
Learn about the collaboration between teams to enhance Google Workspace with AI.
“So there's a lot of really close collaboration between us and the DeepMind team to make these things work.”
Vibe Coding and Knowledge Work Expansion
28:32 to 29:55
Explore the concept of vibe coding and its impact on knowledge work in Google Workspace.
“Yeah, no, I think it's a great question.”
Transforming Content Across Formats
29:56 to 30:56
Discover how Gemini allows for dynamic transformation of content across different formats.
“and then mix these two and then create a slide deck, we think there is probably a way, and we have something called Apps Script that natively supported across all of our products.”
Evolving Document Writing Processes
30:57 to 32:58
Understand the changing nature of document writing and how AI assists in the creative process.
“Do you think that's something that, like, generally will make sense for a direction for slides?”
Enhancing Ideas with Contextual Information
32:59 to 35:24
Learn how to enhance raw ideas with context and information from personal data.
“Because actually we talked about this last night.”
Building Private Indexes for Enhanced Retrieval
35:25 to 38:54
Delve into the engineering challenges of creating private indexes for user data in Workspace.
“because now you have a lot more information, real information from real projects and real CLs and real proposals.”
Transcript
Automatic transcript. May contain errors.0:00We ship 14 products, 3 billion plus users use our products. and if you look at obviously Gmail, Calendar and Drive and Meet and Docsheets and Slides all of them are getting more and more Gemini first experiences Nice, and what does the day one rollout schedule look like? Because you and me are getting pinged by people where we announce them and it's like I know we're rolling out sort of progressively throughout the day People are always They're like where is it? I want to try, I want to try I need Gemini and Calendar I'm like Gemini, my query for this is Gemini to get me out of meetings. I'm like, whatever I can do to get out.
0:37I was like, that's my daily profit. How do I get out of meetings? Cancel them, move them around, send them the doc that they're really looking for, whatever it is, like, get me out of these meetings. I would love that. Getting you out of meetings is hard because you are a very popular guy.
0:56Hey, everyone. How's it going? Welcome back to Release Notes. My name is Logan Kilpatrick from the Google DeepMind team. Today, I'm joined by Chandu. We're talking about the all-new workspace experience. And maybe, Chandu, we can actually just get started with seeing some of the cool launch videos that you all put together. Let's do that. I'm excited. Do you want to sort of narrate what we're looking at? Yeah, what you're looking at is our new experience that's launching with docs. And in this experience, I think you will start seeing two distinct things. One is at the bottom of the canvas, you'll start what we call the bottom bar, which is a new entry point into invoking Gemini when you are on Canvas.
1:35And in this form, you get to select different sources that you want to use for your content generation. In this example, I think you saw the prompt. The user really wants to put together a marketing proposal for a pop-up store, and they want to model it after a pop-up store plan that they had done prior year. And then they also want to bring their spring collection So there are lots of references to the semantic information that is often always in your drive or your Gmail or some of your chats. So how do you bring all of that context into a single prompt and generate the entire doc in single shot, right?
2:17So that's the experience that you just saw. And we are super excited about it because this is the first time I think we are breaking down the silo barriers in the way you bring context into Canvas. I love it. And I feel like it feels, it felt like AI native, just like even like seeing that. I see that every time I do docs.new and I open up a new Google Doc to get started, I like see that experience. And I'm like, just as it's my like reminder, I'm like docs is in the AI era. And it's right here, front and center in the experience, which is really, really cool. No, absolutely. I think, you know, the teams have done a fantastic job and every single thing that you're seeing By the way, all of this experience is also supported in the old native side panel that we have built.
3:03So you can seamlessly transition between on-canvas entry point or the side panel. Bringing all of this together and bringing in some of the Gemini 3 family models, that really transformed the experience. Is this all powered by Flash? So some of these content generation is Flash. Cool. But if you think about some of the heavy computational, like, you know, we'll talk about spreadsheets, for example, we have a feature called Sheets Builder that can agentically can build a full P &L for a business. Wow. And that requires a pro power. So it uses the pro model. Nice. Awesome. Maybe we can look at the next drive, AI Overviews.
3:47Yeah, let's take a look. I feel like AI overviews is like a concept across information-dense, like search. Products, exactly. It's actually like an incredible concept. I love it. It's such a good idea. And it translates well, right? Like because when you are looking at web, you are looking at the web corpus, and you kind of need that single snapshot view of what you are looking at. And when you are in Gmail, we launched AI overviews in Gmail in January, and it's really getting lots of traction and lots of love from our users. Now we are introducing the same concept in Drive. And the problem with some of the Drive corpus is, and I love my Drive and I have everything there, and it's often not organized, even though I would love to organize it a little bit more.
4:34And because it's not organized, I tend to forget the exact file name. And so most of the queries are very abstract, like what was the deck that I presented to the audit committee or what was the document that has this LRP with certain revenue numbers. So you could do really these semantic queries even if you don't know the file name, even if you don't know the folder, AIO would be super helpful in surfacing the right file at the top of the Rav experience. I had an example this morning where I was trying to find a proposal doc that I had made about changing the AI Studio name two years ago. And so I actually did the like keyword search type of vibe this morning and I'm going to retry it again and see if it gets picked up.
5:19Give it a try, please. Yes. Inside of Drive. That's going to be great. So for both of these experiences, who is the set of customers who this is rolling out to? Is this like pro like paying Workspace subscribers are getting access to this to start? That's right. So this is our beta launch. We have a set of customers that are on Workspace commercial side, which is, you know, which we call the preview customers. So we are rolling out this to preview at the same time on the consumer side Which is our G1 AI Pro and a ultra customers. They also get this experience as well Nice and what is the day one rollout schedule look like because you and me are getting pinged by people where we announce And then it's like I know I know we're rolling out sort of progressively throughout the day People are always like where is it?
6:03They're like where is it? I want to try I want to try No, we are I think you know We heard the feedback and I know a number of times you get tagged on X and I get tagged on X. I know, you keep sending them to me. No, I think on day one, you will see all of our A-Ultra users get it. And then during the first week, we'll ramp to all of our A-Pro. Nice. And also our private preview commercial customers. Awesome. I'm super excited. And then there's one last new feature. There are multiple, actually. There are multiple. So the way we are thinking about this launch log and think about all of the things that we have been dreaming up about, you know, bringing into Gemini era, right?
6:45Like, you know, obviously the models have gotten incredible capabilities, especially with Gemini 3. I think, you know, you have both coding capabilities and the reasoning I think has gotten to a new standard. So we have taken that inspiration and then kind of reimagined every single canvas. And when I say every single canvas, I'm talking about Docs, Sheets, Slides, and now Drive. And then you have seen Gmail. So this is kind of our second milestone in our journey of bringing Gemini to all of our beloved workspace products. We are launching all three editors, Docs, Sheets, and Slides. And also we are bringing Gemini to Drive, which we just saw on AIO.
7:25And we are also bringing the Ask Gemini feature where you can go deep into a set of files or a set of folders. It's kind of like a more for your own personal drive. So you could actually have multi-turn conversation with your own files, and then you can export it to a doc or whatnot. I love that. I feel like we need that inside of Google because so much of like the organizational knowledge base is like living in drive, but it's completely unstructured most of the time. But for better or for worse. One of the, I've got many, many questions. Actually, so you mentioned the like second, And this is like the second milestone.
8:03I'm curious. I don't know if we have like, what's the third mile? Like what happens next, actually? Because I feel like we've made a huge amount of progress. Is it like, you know, you and I are having some of these side conversations about, you know, AI and Google Sites and Forms and all that stuff. Is that third milestone or what happens after this? No, I think this is a continuous journey, right? We ship 14 products, 3 billion plus users use our products. And if you look at, obviously, Gmail, Calendar, and Drive, and Meet, and Docsheets, and Slides, all of them are getting more and more Gemini-first experiences.
8:36And as we look at the entire roadmap, we are constantly talking to our customers, constantly talking to our users, and you and I are out there on X, and we get lots of feedback. So we are taking it all in and then thinking about our roadmap and bringing that out to our users and customers in a very thoughtful, meaningful way. I need Gemini and Calendar. I'm like, Gemini, my query for this is Gemini to get me out of meetings. I'm like, whatever I can do to get out, like that's my daily prompt is how do I get out of meetings, cancel them, move them around, send them the doc that they're really looking for, whatever it is, like get me out of these meetings.
9:14I would love that. Getting you out of meetings is hard because you are a very popular guy, but we have a very exciting roadmap where we can assist more and more users with their planning. And we also see a lot of students using Calendar for their planning. I think we are seeing incredible amounts of use cases. Obviously, the basic thing like, can you schedule 30 minutes with this person, that person? Those are the kind of things, obviously, we want to bring to Calendar. But getting you out of meetings, I don't know. Yeah, yeah, that's the dream. I'm curious, actually, So you mentioned sort of the compatibility with the side panel.
9:54And as we look up here, I see the side panel open inside of Drive. What's the story for the future? Is the future of the side panel like it'll just continue to be there because there's all these use cases where that makes sense or like you'll build it all integrated into the mainline version of the experience? I think there is a place for side panels. So think of this example, right? So you have a document that you opened, and you can go to the main entry point at the bottom and then say, hey, write this doc for me, and that's the demo that you have seen. But there are scenarios where you just want a summary of the doc or you want to ask a question on the doc.
10:34And so when you say summarize this doc for me, obviously the response does not belong to the canvas. So where do you put it, right? So you need slightly high fidelity real estate to have a longer set of non-Canvas binding content. That's where we think that the side panel will continue to exist. And you can also have multi-turn conversation on a PDF because there isn't much you can do on Canvas. And so that will be side panel. So we'll continue to support these two because I think there is a very seamless way where we can bring this experience and not make it look like it is a hard switch for the users.
11:11and so we are very thoughtful in the way we are baking all of this into a unified experience. I love that. We were talking last night at dinner and you were telling me about those sort of use cases for like how you're actually using this day to day and you were pushing me because I was saying that I'm still writing. I write all my docs by hand right now. And part of it I think historically been because like I was too lazy to go get all the contacts and bring it into the contacts window. I feel like now with all the new stuff that's rolling out like I think that gets a lot easier, but I'm curious, where are you finding yourself using the new AI experience and workspace most?
11:44No, I think we have been obviously internally dogfooding this for a while, and I had access from day one. And we talk about this internally of new way to work, right? And this new way to work is going to be a lot less about compiling information and a lot more about directing outcomes. And what it means by that is if I'm writing an update or I send, you know, these periodic emails to my team and now I need to go compile all the information from my past emails and my meetings and my notes and so on and so forth, just compiling the information on a Friday night takes a lot of time. Now, with some of the source selection stuff that you have seen and some of the continuous context that is currently available in this launch, I am able to synthesize an initial draft of a doc in less than a minute.
12:36And of course, I'll put my finishing touches on it. So what happens is, instead of compiling all the information yourself, you are more and more putting these finishing touches for the initial draft that is created, which is like literally directing the outcomes. And this 90-10 flip, like 90 % of the work that you delegate to the AI and 10 % you are going to take control of the final product. That's the mode that I've been operating and we launched Help Me Write in Gmail. We see tremendous amount of usage on Help Me Write. And with the docs that we are launching, I think you'll probably start seeing that and we see that in my behavior, it has changed the way I write documents.
13:17Same thing for slide generation. So we are also launching slides. Slide generation, I think, is something that a lot of people have a blank side slide, and they know how to write all the text. But organizing it in a beautiful way is always hard. I can't do this. This is why I try not to make slides, because I'm always stuck on this. No, I think that makes two of us. Now I can just go write my entire thought stream in a big paragraph on a slide, and I say, enhance this slide. and it comes up with this beautiful design because Gemini has this ability to take this text information and come up with these visualizations using the SVG format that slide support and then puts it back in a beautiful slide that I can just edit and then ship it, right?
14:02The way we create content, I think is going to change significantly and we are just getting started on that long journey. Yeah, I'm excited for it. My one feature request is when I use Gmail today, I'm writing like more or less good, well articulated English, but Gemini is always like all, there's a lot of like those purple squiggly lines under everything that I'm writing. And I'm like, sometimes I'm like, okay, you just let me write my, my sort of proofreading your content, my somewhat broken, not, not even broken English context. It's like, just, yeah, not, not perfect grammar. Um, so I don't know what I need to tell Gemini to, cut me a little bit of slack so that it's not always purple squiggling all my stuff.
14:44No, we do. Yeah. So I think, you know, that's a good point. Like, you know, the proofread is one of our most widely used feature and we get a ton of good feedback. It catches some good stuff. Sometimes I'm making, sometimes I'm not making sense. I can totally see how sometimes even you're trying to be intentionally different. Yeah. You don't want that to be highlighted as like, hey, may I suggest, yeah, totally, totally fair feedback. Yeah. Back to this thread of like sort of people's personal experience using the Workspace products. When we landed the Gmail phase one of this AI and Gmail story, I think there was lots of comments of like people who, A, sort of just want the traditional Gmail experience.
15:24So like, hey, AI is great. Or maybe AI is not great. I don't want it. I'm just interested in using my email. Like, let me continue to do that. People will be able to do that. there's like a bunch of misconception about like what that even that first rollout meant. And I'm curious also for this next set of rollouts sort of, I think there's a lot of questions about like, you know, there's AI in this experience. What does that mean for my data, et cetera, et cetera. So I'm curious what the, yeah, like what the top level narrative is for like a data perspective. Yeah, it's an excellent question. Even in Gmail, I think, you know, there are, and you'll see this in this, in this launch as well.
15:59We are always thoughtful about a couple of things, right? When you introduce a new product like AI Inbox, we know folks love their inbox. We have spent probably decades customizing which label and then how to organize your incoming emails. And we don't want to touch that. If you want to give it a try where AI is organizing your emails, you can give it a try. So your main inbox is not touched, right? And we take the similar approach, a very thoughtful approach in Docs, Canvas experience, Sheets experience, Lights, and Drive, we are not trying to force AI to be the be-all and end-all. So I think it is more about how AI can be assistive in your core journey.
16:44But then while we are offering that assistance, we also want to give you the control of how you can be assistive. A good example is you have seen the source selection. Yeah. Then you're going to the prompt box and say, hey, write me this marketing report, you have the full control of what inputs you want to feed into the prompt, right? Because Gemini, by default, is not going to sit there and read all of your drives or docs or whatnot, right? So you have to kind of allow the access to certain sources and say, hey, use this and then help me draft. The best way to explain it, let's say you are using a simple calculator, right?
17:22Calculator has obviously some intelligence inside, like, you know, it can do some math and it can do arithmetic really well. And you input a bunch of numbers. And let's say you get 5 plus 5 and it comes back to 10. So it knows how to do the math. In the process, you are inputting some data. The calculator is not remembering what you input. This is your data, 5 plus 5. And it's not learning how to do math because it already knows how to do math. The same thing is happening behind the canvas, right? So when you say, read these docs and read these slides and read these chats and then help me a draft, you are synthesizing a bunch of information saying, hey, Gemini, take this, understand what they are saying, and then bring me back a draft that I am going to start working from.
18:08In the process, it is going to look at the data that you provided, understand all of it, and then organize it into a really compelling draft document, which Gemini knows how to do, and then use that intelligence to produce your first draft and give it back to you. In this entire process, there is never an instance where your data is crossing the boundary of using it to produce your document versus learning something new. It's not happening, not in Gmail, not in Docs, and not in Sheets, and not in Slides. And in fact, even when you select your sources or whatnot, are all never going to cross the boundary of your training because training the models is in a completely separate process.
18:52And all we are doing here is invoking the model to understand the context you are providing and producing the content that you can use as your first draft. Right? And I think, you know, that's basically what is happening on every single canvas across our products. I love this analogy of like in this context, Gemini is a tool and like the calculator analogy. I feel like it works really well. As I talk to people across Google who have, you have the responsibility of building a product for many, many billions of users, this like inherent product tension of like Google is a frontier AI company. We're sort of at the, you know, in the race, we're building new products, we're shipping all this stuff.
19:30We want to sort of make sure we're on the cutting edge. And at the same time, you know, there's many, many customers who are sort of, they're happy with Gmail, they're happy with the current docs experience. And I think that tension, we have this big responsibility to find the right balance in that tension. And I'm curious, like actually, if there's any, yeah, if there's any interesting, and maybe the AI inbox example that you gave is the best example. There's other examples of like how we're trying to find the balance in that tension of be frontier and also give customers this experience that they sort of know and love and want to continue to use.
20:02Yeah. We want to build products that are truly helpful, right? That's the end goal. At the end of all of this, we want to be helpful. And we also believe, and I know we talk about it all the time internally, is that AI as a technology has the potential to affect every single human in a positive way. And when you take these two, us trying to be helpful in everything that we do and then the potential of the AI technology, I do think we do internally believe through and through that we can positively help influence human productivity and collaboration by bringing these AI capabilities to workspace.
20:42So when we do that, we also have the tremendous amount of responsibility on making sure whatever we bring is done in a thoughtful way and that is truly helpful. That's number one. And all of these products are beloved in the sense that billions and billions of users use them in a certain way today because they have been part of their life for many decades. How do we respect those choices? And how do we thoughtfully bring these AI experiences that only net attitude to their experience and don't come in the way, I think is something that we always think about. That's one. The second thing is, how do we also provide the control over what is available and when, right?
21:24And then how do you get to turn them on or off, depending on your choice? So how do you give that control back to the users is the second one, right? Third one, when you want something and when you, let's say, I did turn on something, giving you the transparency of what is being used to produce what, I think is the third thing. Combining all of these things, building a feature that is delightful, but also honors this, you know, truly assistive and gives you the full control and gives you the full transparency is something that we think day in, day out. And that's really what drives our product building philosophy.
22:00And that's what we do coming into work every day. I love that. I've got a slightly tangential question to this, which is I see you post a bunch of like the Google Vids stuff all the time, which made me think about it. I'm not a like active Google Vids customer. I've tried it out a few times. But like what's your favorite, maybe it's Vids, but like what's your favorite workspace of the 12 or 14 experience that like you think folks don't use or like not that many people use, but as like a big unlock for you. You seem to love Vids, so I'm like, maybe that's too easy. Vids is my new favorite app. Vids kind of does let me do things that I never imagined I would be able to do, right?
22:44And that's the kind of superpower I think, you know, AI kind of gives you is it gives you capabilities that, you know, if you asked me last year, hey, can you create a video podcast every week on what happened in Workspace, I would have laughed at it and said Logan. Simply not possible because I am not the most creative person and I also don't know how to edit videos. So no, thank you. But now with WITS, I think it takes less than an hour for me to, you have seen some of my videos on X and I now make a video of all things that happened in Workspace as a WITS every Friday night. That's my like pastime.
23:22And I think that's the power of AI. And going back to your question, there isn't a day that I can imagine without using many of Workspace products. The productivity boost that I see and also the transformation that is happening right in front of our eyes with Gemini in Workspace. I think if you say, what is the biggest unlock? I think most people haven't really seen the power of Gemini in Workspace, but we are really excited about bringing what you heard in Gmail in January. We are bringing that to the editors and drive in March. And of course, we have products like Vids that are really taking off on usage.
24:04I would love for you to use it too at some point. The Vids team is great. They pay me all the time. The Vids team is on fire. So no, I think we have a great set of products. And it is truly a privilege to be able to work on these products. They do positively influence billions of humans. And if we can bring AI to enhance those journeys and give them these abilities, like me doing vids and videos, if we can bring these abilities to the humanity in both their productivity and collaboration, I think that's a fantastic opportunity. I agree with you. I promised you that I would ask you some of the hardball questions that I had as well.
24:47We were talking about this. One of them that was top of mind is, and it's not strictly a hardball question, but there's an interesting arc of the AI workspace story. And I feel like we've, obviously today with all the launches, the product experience is great. Can you talk a little bit about the journey to get here? And obviously the models have improved a ton, the infrastructure has improved a ton to make this moment happen. And obviously, like, I think I'm thinking of the tweets that you and I have both seen where it's like, you know, people asking Gemini and whatever it is, docs, like, can you edit the doc?
25:20And it's like, sorry, I can't do that or whatever it is. I think we've seen lots of that feedback. And I think hopefully address lots of that feedback in this current launch. But I also think, like, the most folks don't understand the sort of, yeah, the progress that it takes to get to this point. So I'm curious, like. No, I think it's a very humbling experience to see all the feedback. And internally, too, you have given me plenty of feedback. Yes, I think in today's launch, especially the deep integration of Gemini that we have brought into the canvas where these conversations on the side panel, you'll say, rewrite this paragraph for me on the canvas.
25:58And it will say, no, here is the thing, paragraph, why don't you copy it over, right? We have fixed it. So Google says it's fixed it. and we have gone further. You can actually control to a character level on Canvas using natural language. You can say things like, you know, turn every other word into bold, Google Sans 18, and it will do it. So the real journey really is, Logan, models have improved dramatically over the last 12 months or so, and we are witnessing breakthroughs. Korai and team and DeepMind, they are putting our breakthroughs almost on a monthly basis at this point, and it is too fast.
Read the full transcript
26:36It is too fast. We are having a hard time keeping up. But also I think there are a number of application level integrations that require this very deep understanding of how models perform. That's one. And second, how applications actually behave. And I'll give you a simple example on the docs itself because you brought that up. The docs that you see today, like we have really rich formatting that we support on canvas visually. Underneath that we have a significant amount of really complex representation. Models by definition, obviously they are really smart and if you instruct them to generate content in certain format, they can generate it.
27:20But then translation between the model and then the canvas is happening in a, let's say, a markdown format. It's a very lossy conversion, right? Doc Canvas can actually support much richer formatting. Models understand it. But then the way these two speak was happening through to Markdown. So how do we fix that? How do you make Gemini model to speak Doc's language? And Doc expresses back to the model in Gemini language. That's the kind of really hard engineering work that we have done. And that's true for Sheets. That's true for Slides. Slides has this SPG format that we go deep into verbalizing it properly.
28:02Same thing with Sheets verbalization to Gemini. So there's a lot of really close collaboration between us and the DeepMind team to make these things work. Everything that you are seeing today is really built ground up in the last six months, right? And it is tremendous amount of work. Kudos to the team on both sides to get us to this point. But it is really hard work. Yeah, it's crazy to see the level of progress that's happened. And I'm excited for folks to get their hands on the experience. Something else that you and I have been talking a lot about is how can we get a, or sort of maybe more generally, like thoughts on everything that's happening with vibe coding and sort of like the expansion of the code domain to like knowledge workers.
28:47and obviously like a ton of knowledge work that happens in humanity is taking place on workspace and where you think sort of the idea of like, like code as an artifact, as like a new primitive potentially inside of the suite and workspace could live. Yeah, no, I think it's a great question. I know you are driving - Make it happen. Yeah, you are with EA Studio, I think, you know, really is a trailblazer in setting the standard for vibe coding. and I love seeing all of the things. Some of our own product leaders in our group actually start on AI Studio to prototype things quickly to bring their ideas to life, which I love seeing that.
29:27But I think the coding, as you know, coding as a domain is evolving. Most of our workspace products, right? Like you take Sheets, for example, modeling some of the functionality as a coding problem also has yielded multiple breakthroughs. So models, again, are really good at doing really complex things. But then if you take some of the workflows that people do, you know, take something in docs and then take something in sheets and then mix these two and then create a slide deck, we think there is probably a way, and we have something called Apps Script that natively supported across all of our products.
30:08And there are some ideas that, you know, we are brainstorming with your team on how to think about revamping, you know, wipe coding on top of Apps Script and then bring, you know, more ubiquity across all of our surfaces with simple instructions. And I think there is tremendous amount of potential. Hopefully we can share something there soon. I love it. Over the holidays, one of the experiments that I did was like, again, through this lens of like code becoming a fundamental primitive code as slides was like, I actually, like, obviously, Nano Banano works really well for this. But also, like, I feel like you can do lots of, like, really interesting, like, interactive things was, like, the coolest use cases.
30:52Like, I could, you know, it had buttons and it had all this other stuff. And I could, like, actually in real time pull information and render it. Do you think that's something that, like, generally will make sense for a direction for slides? No, I think, you know, it's a good question. we are seeing a trend where, you know, information is information. How that information is transformed into an artifact, I think, is a post-fact, right? For example, if you have a piece of content, there are multiple ways to present it. You can present it as a document, or you can present it as a slide deck, or you can present it as a spreadsheet in some cases, because a table format is probably the most appropriate.
31:34And models are really making this transformation really dynamic and easy. Gemini can do all these transformations with a piece of content. So we are seeing more and more of our users telling us, like, hey, they want to maintain all of this context in certain place. And then on demand, how do I take this doc and turn it into a slide deck or take this slide deck and turn it into a sheet? So I think the transformations are becoming less and less about specific formats, more and more about how do you keep the entire context in one place and then morph it. And I think that's what you are suggesting.
32:11I think, you know, we talked about this over the holidays. Then how do you then take that concept and then push it further and make it not just these static canvases that are producing just a content display, but make it more and more interactive. And I think that is going to happen. And we are seeing the first glimpses of that with some of the things that you see, even with this launch. There is a lot more interactive pieces, especially on sheets. You'll start seeing that. I'm excited. Chandu, something actually back to this thread of how we use the tools. there's this interesting thread around the tension between like obviously the models can and especially with the latest update and docs like make a great doc it has a bunch of the contacts etc on the other hand it's like people actually and this is how I feel about my own doc writing process like I'm usually actually writing the doc because I need the doc as like the the mechanism by which to formalize my own thoughts and I'm curious like how you see that evolving over time and how this happens for you?
33:16Because actually we talked about this last night. Yeah, we did talk about it. It's a great question. So the way I think about this is when you are starting with your original ideas, I think that's where human creativity trumps everything. Let's say I'm coming up with a proposal for something, a really new technique in the way we either post-train the model or build a new feature or something that is novel in terms of how we want to approach an existing problem, that is really where I think human creativity. And I also start with a blank canvas and then start jotting down my thoughts. Where I think models are really helpful, and I talked about the fragmented context and the silos, et cetera, and where you can save a bunch of time is once you have this raw collection of thoughts, how do you enhance them either with supporting arguments or arguments that go against it with your own context?
34:20Now, I'm not even talking about web context, right? For any given idea, I can guarantee you within my drive, within my Gmail, within my chat, there is a ton of context that is trapped in these three artifacts that either supports my hypothesis or refutes it, right? Just doing one quick gut check by selecting my bullet points and then say refine this into some detail using this context of my email and chat will instantly give me an amount of validation that I could never get before. Once I have that validation, like what do I do with it and how do I then take my idea and morph it into either a different direction or continue on the path, it's obviously entirely up to me.
35:04So the way I have been using these new docs experience as a tool to enhance my own thinking is that you start with your blank canvas, you jot down your rough ideas, enhance them with your own context, not web context, with your own context, because it kind of pulls all the context that is in your corpus. Then you start mutating it because now you have a lot more information, real information from real projects and real CLs and real proposals. then you start thinking like, okay, so there was this other project that tried to do it like 12 months ago. Okay, why didn't it go well? And what are the things that we learned?
35:41What can we avoid this time? This entire exploration process, I think would have taken probably hours before, but now take minutes, right? I think that's the power of the experience that I see every time I use it. But then for creative writing, would I invoke it as my first choice? probably not because I do enjoy creative writing and it is not about compiling information right that's the difference is like if you are compiling information if you are spending a lot of time gathering pieces of data how can Gemini can jump in and then help you with this in seconds I think that's where the power of Gemini comes through on our canvas I love that and and you mentioned like context a bunch of times I think a part of this story is like obviously it's not retrieving over all that.
36:23That's like the whole, that's the whole corpus. But I feel like it's a really difficult context engineering problem in order to like surface the right things. And I'm curious if there's just like any detail about like how we're doing that. I think, you know, if you look at the amount of data that our users store in their drive and in their Gmail and so on and so forth, obviously it's their personal data, really private data, really sensitive data. So So anything that you are doing with that index, which is the token-based retrieval indexes that we have built in the previous years, where you can put in a keyword and search your Gmail, and you can put in a keyword and search your drive, obviously worked.
37:05But they are not good enough in this world of, you can express longer and longer queries, expressing all of the semantic meanings of things, not just your file names and email subjects. That's where I think we have to rethink the way we are building the index for your email and for your drive. And we use these Gemini embedding models that bring a tremendous amount of dimensionality to the data, right? So typically a human can think of like, you know, in less than 10 dimensions at a given point of time. But these models can synthesize data in 768 dimensions in a single shot and then get the right nugget of information with no effort at all.
37:51And so that's the kind of architecture that we are now bringing to all of Workspace. Obviously, it is technically a really challenging problem on corpus like web. You build web index once and reuse it for everybody because it's the same web page that you get to see and I get to see. But your inbox is different from my inbox. So your index is very different from my index. So we have to kind of, for every human, you have to go build this index. in their own silo and then only use it to themselves and then keep it real time because you get probably an email every other second and so you need to keep it refreshed.
38:29Exactly. And same thing for chat messages. So from a pure engineering perspective, while the models and the embedding models, et cetera, have gone really come along, the scaling aspect of keeping it one fresh but also maintaining this private index for every single human that uses workspace as their core product is an enormous engineering challenge. And there are extraordinary amount of engineering effort that our teams put in to scale the index, serve it at a latency that is really, really low in terms of how fast these results are retrieved. And then of course the quality, like when you ask for these really abstract questions, whether it's Drive or Gmail, how do we bring out the right information to your fingertips through AI overviews?
39:21That's where I think most of the work has gone in the last six months. Yeah, that's incredible to see. Well, kudos to you and the workspace team for pulling this off. I'm excited for chapter two. I'm excited for hopefully chapter three. I'll see you at Cloud Next. Hopefully we'll see a bunch of awesome workspace stuff there. And congrats on the launch again. Thank you very much. And I think, as I said, teams have been working really hard. You and I talked about it, right? So in the age of AI, velocity is part of core strategy. And I think, you know, our teams have fully embraced it. And so, yes, really excited about what we are going to show it next.
39:59And I'll see you next. I love it. Thank you, Chandu. And thank you, everyone, for watching. We will see you in the next episode.
From the publisher
Chapters:
1:15 - Gemini in Docs
3:17 - Which models power Workspace
3:45 - AI Overviews in Drive
5:22 - Rollout and availability
6:33 - Reimagining every Workspace canvas
8:58 - Gemini in Calendar
9:50 - The future of the side panel
11:16 - A new way to work
13:18 - AI-powered slide generation
15:08 - User data and privacy
19:07 - Balancing AI innovation and user trust
22:04 - The power of Google Vids
24:42 - The journey to deep canvas integration
28:32 - Vibe coding meets Apps Script
32:42 - AI as a tool for thinking
