WEBVTT NOTE Sentence-level transcript of https://www.youtube.com/watch?v=buHC7bQE1X4 NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/buHC7bQE1X4.html. A cue ends where the next begins, or 2 s after its last word. s1 00:00:32.960 --> 00:00:33.280 All right. s2 00:00:33.280 --> 00:00:35.280 Hello everyone. s3 00:00:35.280 --> 00:00:37.680 Um, my name is Kevin. s4 00:00:37.680 --> 00:00:40.000 I'm going to be talking about anti-gravity. s5 00:00:40.000 --> 00:00:43.120 So, are there any World Cup fans out there? s6 00:00:43.120 --> 00:00:44.640 Woo! s7 00:00:44.640 --> 00:00:50.800 Imagine you are coaching Argentina and you're in the 89th minute and you have Messi on your team. s8 00:00:50.800 --> 00:00:53.280 What play are you running? s9 00:00:53.280 --> 00:00:57.840 It's called Give Messi the ball and get the heck out of the way. s10 00:00:57.840 --> 00:01:00.399 LLMs aren't just role players anymore. s11 00:01:00.399 --> 00:01:04.879 They can be your star player if you build the right product around them. s12 00:01:04.879 --> 00:01:10.400 And to let your star player cook, you have to get out of the model's way. s13 00:01:12.000 --> 00:01:13.280 We might want to get the slide. s14 00:01:13.280 --> 00:01:15.200 Are the slides up? s15 00:01:15.200 --> 00:01:15.920 Oh, they are. s16 00:01:15.920 --> 00:01:16.640 Great. s17 00:01:16.640 --> 00:01:23.040 Um, so Anti-gravity is Google's agent coding product for technical and non-technical users. s18 00:01:23.040 --> 00:01:30.880 Uh we launched back in November of 2025 and have been accelerating devs both within Google and externally ever since. s19 00:01:30.880 --> 00:01:36.000 My name is Kevin How and I lead part of the engineering team on anti-gravity. s20 00:01:36.079 --> 00:01:39.360 So let's talk a little bit more about what anti-gravity is. s21 00:01:39.360 --> 00:01:43.840 We have and always will be unapologetically agent first. s22 00:01:43.840 --> 00:01:54.079 So, we debuted the anti-gravity IDE last year with a brand new agent manager concept and it was a platform to manage and orchestrate many agents. s23 00:01:54.079 --> 00:01:59.840 Since then, we've actually extracted our agent and launched our own anti-gravity CLI. s24 00:01:59.840 --> 00:02:10.640 And last month at Google IO, we had the pleasure of launching anti-gravity 2.0. In the theme of getting the model out of the way, we actually decoupled the IDE from the agent manager. s25 00:02:10.640 --> 00:02:12.800 So, now you have two separate applications. s26 00:02:12.800 --> 00:02:18.000 Um, and now you can use the agent manager in a standalone app. s27 00:02:18.480 --> 00:02:23.200 And since pictures are worth a thousand words, here's a screenshot of anti-gravity 2.0 in action. s28 00:02:23.200 --> 00:02:33.599 As you can see, not only is it your own dedicated mission control for your agents and projects, you have sub agents, you have all the new models, you have work trees, scheduled tasks, s29 00:02:33.599 --> 00:02:34.480 voice mode. s30 00:02:34.480 --> 00:02:36.879 There are so many things to unpack with the product. s31 00:02:36.879 --> 00:02:39.440 But I don't want to spend today telling you about the product. s32 00:02:39.440 --> 00:02:47.760 I want to tell you a little bit more about the behind the scenes, some of the principles that went into it and notably some of the things that led to its roadmap. s33 00:02:48.239 --> 00:02:55.280 So, as some of you for the longtime AIG fans, uh this is actually my fifth time speaking at AIG. s34 00:02:55.280 --> 00:02:59.200 Um and I've been building developers tools since 2022. s35 00:02:59.200 --> 00:03:06.159 And the one thing that has stood above all other lessons that I've tal talked about is the idea of scaling with intelligence. s36 00:03:06.159 --> 00:03:09.920 This means that as the model gets better, so should your product. s37 00:03:09.920 --> 00:03:16.080 And the frontier edge of whatever model you are serving should be apparent inside of your users's product experience. s38 00:03:16.080 --> 00:03:20.239 So let's get into more concrete examples of what this means. s39 00:03:20.239 --> 00:03:28.720 So for those of you that follow me on X or hear me just yap generally for the last four years, you'll know that I've been working on a number of these sort of transformations s40 00:03:28.720 --> 00:03:30.799 year over year over year. s41 00:03:30.799 --> 00:03:34.799 In 2022, I was working on autocomplete and chat sidebars. s42 00:03:34.799 --> 00:03:46.560 This was based on embeddings, rules, files, a syntax tree parsing, basically everything inside of that app is deterministic because that's all that the model could really handle. s43 00:03:46.560 --> 00:03:52.959 And in 2024, when agents came onto the scene, it completely changed how developers were going to do work. s44 00:03:52.959 --> 00:03:59.040 With it came new primitives like MCPs, custom tools, and permission systems. s45 00:03:59.040 --> 00:04:09.680 And with 2025, we introduced anti-gravity's agent manager with many other products following suit in that similar form factor with users managing many agents at once in parallel. s46 00:04:09.680 --> 00:04:15.200 And this led to things like skills, hooks, artifacts, and a couple other primitives. s47 00:04:15.200 --> 00:04:18.079 Um, and that sort of defined the 2025 era. s48 00:04:18.079 --> 00:04:23.440 So, let's talk a little bit about 2026 and what those primitives might be. s49 00:04:23.440 --> 00:04:29.199 Before we answer this question, I want to take you back to some of these battle scars that are a little bit closer to home. s50 00:04:29.199 --> 00:04:31.759 Scaling with intelligence really is not easy. s51 00:04:31.759 --> 00:04:39.919 It's really hard to take away something that users love and are familiar with to lead them down potentially, and that's a big keyword, a better path. s52 00:04:39.919 --> 00:04:47.120 We aren't right 100% of the time, but there are two that jump to mind when I was putting together the slides for this talk. s53 00:04:47.120 --> 00:04:49.919 The first one is giving AI a terminal. s54 00:04:49.919 --> 00:05:01.440 We all remember fears about son of Anton deleting your entire codebase and doing catastrophic things to both, you know, your startup, your company, etc., etc. But as models got better and people invested in primitives s55 00:05:01.440 --> 00:05:08.400 such as permission systems, users ended up building faster, they ended up shipping more, and they did so safely. s56 00:05:08.400 --> 00:05:09.919 So, we were able to overcome this. s57 00:05:09.919 --> 00:05:16.639 And as models got smarter, they were able to make better decisions about what they should and should not run in your terminal. s58 00:05:17.120 --> 00:05:25.759 The second instance is um this tweet which is very representative of sort of the yelling that I got uh when we removed chat from windsurf. s59 00:05:25.759 --> 00:05:33.280 So a lot of users were yelling at our team because we took away something that was very dear to them the chat sidebar and replaced it with only an agent. s60 00:05:33.280 --> 00:05:36.960 Now at the time this is something that was familiar and rather difficult to swallow. s61 00:05:36.960 --> 00:05:40.080 But when we look back, models have advanced. s62 00:05:40.080 --> 00:05:44.800 Multistep research, agentic research and execution became the new paradigm. s63 00:05:44.800 --> 00:05:50.000 And here we are today using and loving all these agentic products. s64 00:05:50.000 --> 00:05:52.160 And so now I bring you to today's battle. s65 00:05:52.160 --> 00:05:53.840 What is going on today? s66 00:05:53.840 --> 00:05:56.960 So we decoupled the agent manager from the IDE. s67 00:05:56.960 --> 00:06:01.039 And with anti-gravity 2.0, we split them into separate applications. s68 00:06:01.039 --> 00:06:06.319 We believe that the IDE is to the agent manager what the debugger was to the IDE. s69 00:06:06.319 --> 00:06:14.240 You don't always need a debugger, but it definitely is helpful to have it if you need to go a layer beneath and go one step deeper into that abstraction stack. s70 00:06:14.240 --> 00:06:23.120 And our prediction is that this idea of agent orchestration, you can call it agent teams, you can call it swarms, you could call it software factories, is the future. s71 00:06:23.120 --> 00:06:26.479 And we're willing to bet on that future. s72 00:06:26.479 --> 00:06:31.360 So here are the primitives for what we're calling the agent teams 2026 era. s73 00:06:31.360 --> 00:06:36.080 These are things like sub aents, generative UI, and sidecars. s74 00:06:36.080 --> 00:06:42.160 And we'll talk more concretely about what those things are and some examples of how they manifest inside of the product. s75 00:06:42.160 --> 00:06:45.919 But it's really important to first understand the why. s76 00:06:45.919 --> 00:06:52.639 What brought about these changes and what model changes, what model properties actually led to the development of these new things? s77 00:06:52.639 --> 00:07:01.840 And as a product team, do you force the new era of primitives or is it something that comes to you by using the model and experiencing the model? s78 00:07:01.840 --> 00:07:04.560 The answer is kind of both, right? s79 00:07:04.560 --> 00:07:10.960 And the privilege of being inside of Google DeepMind is that we do have that relationship between the product and the model. s80 00:07:10.960 --> 00:07:18.160 So you remember the crux of anti-gravity 1.0 is to manage agents in parallel, to put the human in the driver's seat. s81 00:07:18.160 --> 00:07:23.759 And if you remember my last talk, I talked a lot more about this research product flywheel. s82 00:07:23.759 --> 00:07:32.160 And now, as promised, because of the anti-gravity product, Gemini has now learned a thing or two about how to manage a team of agents. s83 00:07:32.160 --> 00:07:40.400 There's still a lot of headroom to make multi- aent systems better, more collaborative, better at deconstructing tasks into smaller tasks. s84 00:07:40.400 --> 00:07:43.680 But we've got a really good head start with Gemini. s85 00:07:43.680 --> 00:07:55.520 and all the basics have been imbued to the model so that we can build a product like anti-gravity 2.0. Gemini 3.5 FL yeah Gemini 3.5 flash was launched back in April s86 00:07:55.520 --> 00:08:03.680 and this brought to market a lot of those capabilities that we had been working on in the background with anti-gravity and flash now isn't just good at executing tasks. s87 00:08:03.680 --> 00:08:06.240 It's actually really good at leading teams. s88 00:08:06.240 --> 00:08:14.639 It's faster and cheaper, pushing the paro curve of what is intelligent versus the speed and the cost at which you run those things. s89 00:08:14.639 --> 00:08:20.960 And putting this all together, we were really excited to announce agent teams in public preview inside of anti-gravity. s90 00:08:20.960 --> 00:08:29.520 All you have to do is simply type the slash command/teamwork and you'll see a new mode where you can enter and unleash a swarm of agents onto the task at hand. s91 00:08:29.520 --> 00:08:32.560 So we'll talk a little bit about how this works. s92 00:08:32.560 --> 00:08:35.440 You as a user will specify your task. s93 00:08:35.440 --> 00:08:37.039 The more specific you are, the better. s94 00:08:37.039 --> 00:08:45.519 Though the nature of these agentic communication styles is that if it needs something more, it can actually ask you for more until everything is basically clear. s95 00:08:45.519 --> 00:08:51.440 You'll work with that lead agent and it will manage a team of arbitrary size to get that work done. s96 00:08:51.440 --> 00:08:53.680 And what I like to say, it's kind of like the Avengers, right? s97 00:08:53.680 --> 00:08:56.560 It'll take a bunch of specialized roles. s98 00:08:56.560 --> 00:09:03.440 It may front-end engineers, backend engineers, infrastructure specialists, QA, design, the list goes on and on and on. s99 00:09:03.440 --> 00:09:07.360 And there are infinite possibilities for what each of those sub aents could take on. s100 00:09:07.360 --> 00:09:11.680 Each sub aent is dynamically generated and can operate independently. s101 00:09:11.680 --> 00:09:16.160 Um, and it can even actually select a different model from what the main agent is using. s102 00:09:16.160 --> 00:09:18.080 And this is done so by that main agent. s103 00:09:18.080 --> 00:09:21.279 Again, we are scaling with intelligence. s104 00:09:21.839 --> 00:09:25.200 And one of the coolest aspects of this is that it can use generative UI. s105 00:09:25.200 --> 00:09:30.080 With a model that is as fast as Flash, things can happen nearly instantaneously. s106 00:09:30.080 --> 00:09:32.640 If you ask, hey, what is the status of my task? s107 00:09:32.640 --> 00:09:34.399 show me a canban of what's going on. s108 00:09:34.399 --> 00:09:39.120 Or maybe, you know, you prefer something a little bit more like uh the the Chrome debugger tool. s109 00:09:39.120 --> 00:09:40.560 It can show you a timeline like that. s110 00:09:40.560 --> 00:09:46.720 And all these things are generated on the fly because it's able to generate UI on demand. s111 00:09:46.720 --> 00:09:52.399 So, some of the projects that the system has implemented, um we've built a photo editor. s112 00:09:52.399 --> 00:09:55.760 You can actually edit raw photos directly inside of your browser. s113 00:09:55.760 --> 00:10:00.480 Um we've also built a messaging app that might look a little bit familiar to those in the room. s114 00:10:00.480 --> 00:10:07.360 Um, and each of these took hundreds of sub aents uh, and took almost half a day to run. s115 00:10:07.360 --> 00:10:13.519 But to really put it through its paces, one of the hero runs that we did was actually building an entire OS kernel. s116 00:10:13.519 --> 00:10:20.480 This is something that, uh, we got to show off at Google IO, but we built a complete OS kernel from scratch and actually played Doom on it. s117 00:10:20.480 --> 00:10:23.760 And my colleague Verun was able to demo this at Google IO. s118 00:10:23.760 --> 00:10:40.560 We were super proud of this particular milestone because it really demonstrated that if you throw more intelligence, you throw more sub aents um at this sort of problem, a model like Gemini 3.5 Flash could do this in a way that was not only very very powerful, but also scalable and you know, mildly affordable. s119 00:10:40.560 --> 00:10:46.480 Obviously, we're not going to spend thousands and thousands of dollars to build an OS kernel every day, though it is possible. s120 00:10:46.480 --> 00:10:57.040 And some of the stats out of this, it took 93 sub aents over the course of 12 hours, made 15,000 requests, two billion tokens, and it was under $1,000, s121 00:10:57.040 --> 00:10:59.760 which was one of the really cool aspects of this project. s122 00:10:59.760 --> 00:11:08.560 And so, as you can see with this particular example, sub agent primitives are one of the defining parts about building a 2026 era of agent teams. s123 00:11:08.560 --> 00:11:10.480 So, agent teams are just that first example. s124 00:11:10.480 --> 00:11:16.959 And I want to show you another example that our team uses internally that sort of demonstrates some of these new primitives. s125 00:11:16.959 --> 00:11:20.000 Um the second one is about automating research tasks. s126 00:11:20.000 --> 00:11:21.760 So we work inside of Gemini. s127 00:11:21.760 --> 00:11:26.480 We help sort of make Gemini better at coding related tasks, agentic related tasks. s128 00:11:26.480 --> 00:11:29.680 And this is where the real magic starts happening with the product. s129 00:11:29.680 --> 00:11:34.720 We have an internal version of anti-gravity that researchers, engineers, nontechnical folks can use. s130 00:11:34.720 --> 00:11:42.320 And when they understand the primitives that anti-gravity offers, it becomes a very very powerful way to automate your own workflows. s131 00:11:42.320 --> 00:11:45.760 So we'll take the example of sideby-side eval. s132 00:11:45.760 --> 00:11:50.480 So this is a very common workflow not only at DeepMind but just generally in the industry. s133 00:11:50.480 --> 00:11:56.240 You essentially will take multiple rollouts 1 2 3 4 etc. Um and you want to compare them. s134 00:11:56.240 --> 00:12:02.160 So you'll take a set of tasks, you'll do some rollouts, you'll get some results and they'll essentially be in two different tables. s135 00:12:02.160 --> 00:12:14.480 Now you'll look at the control, you'll look at the experiment, and then you'll have to figure out not only what the difference was, but perhaps what are the reasons for those differences and how can we actually iterate from there and make a better version of for the next experiment. s136 00:12:14.480 --> 00:12:18.959 Now, traditionally, this was a lot of Jupyter notebook elbow grease essentially. s137 00:12:18.959 --> 00:12:25.920 But when you start working with the new primitives in 2026, you end up with a lot cleaner of a workflow. s138 00:12:25.920 --> 00:12:33.440 So researchers were able to automate 90% of this workflow by simply asking the agent about the eval in question using natural language. s139 00:12:33.440 --> 00:12:41.760 Then the agent that is now primed with skills and an understanding of Google's massive monor repo codebase is able to crunch the numbers and get back to you with a delta. s140 00:12:41.760 --> 00:12:47.440 Now what's really cool here is instead of just taking that delta then handing it back to the user, it went the extra step. s141 00:12:47.440 --> 00:12:54.800 It spun up for a research agent specialist that proposes a hundred different hypotheses over why those deltas might occur. s142 00:12:54.800 --> 00:13:08.720 And then it uses sub aents to then spit up one sub aent for each hypothesis and basically drills into that particular case in parallel mapping back to a single response and then telling the researcher, hey here are some areas that I found. s143 00:13:08.720 --> 00:13:11.600 Now uh here's a report that you can review. s144 00:13:11.600 --> 00:13:13.920 And what's really cool is that it doesn't stop at just the report. s145 00:13:13.920 --> 00:13:25.360 It actually puts together a generative UI for you to look through, interact, select dropdowns, filter, segment, slice, and actually interact richly with that data. s146 00:13:26.399 --> 00:13:29.600 And internally, we care a lot about this sort of workflow. s147 00:13:29.600 --> 00:13:36.560 Improving the model, improving the product, and understanding the ways that users find success and failure internally at Google. s148 00:13:36.560 --> 00:13:40.480 So, what used to be a very manual process now takes minutes. s149 00:13:40.480 --> 00:13:47.760 So, what used to be handineering, you'd have to build your own async pool of agents, you'd have to set up your judges, uh you'd have to tape together data pipelines, s150 00:13:47.760 --> 00:13:53.279 all of this now starts becoming grounded in these new primitives that we've established earlier in the slideshow. s151 00:13:53.279 --> 00:13:56.480 You have a sub aent graph that is completely dynamic. s152 00:13:56.480 --> 00:14:03.760 The generative UI comes in at the end to richly convey the findings in a way that the user best understands or maybe caters to their learning style. s153 00:14:03.760 --> 00:14:07.279 And all of these things can be regenerated and redone on the fly. s154 00:14:07.279 --> 00:14:12.480 All the user had to do was load up a skills file and ask away. s155 00:14:12.560 --> 00:14:19.120 So with teamwork and this eval example, we start arriving at these 2026 primitives that I keep talking about. s156 00:14:19.120 --> 00:14:24.560 And these model characteristics really change the way that we have to think about the product and how we have to develop the product. s157 00:14:24.560 --> 00:14:29.279 So the three examples that we've talked about, first we have the dynamic sub agent. s158 00:14:29.279 --> 00:14:34.079 And to provide a little bit more color here, basically no two sub aents are the same. s159 00:14:34.079 --> 00:14:37.519 The main agent is the one that is orchestrating this entirely on its own. s160 00:14:37.519 --> 00:14:41.440 It's configuring and prompting and seeding these sub aents on the fly. s161 00:14:41.440 --> 00:14:42.880 They can operate in parallel. s162 00:14:42.880 --> 00:14:53.519 They can operate in different types of secure environments be it a sandbox be it a remote execution system and they can all take on infinitely an infinite number of specialized roles. s163 00:14:53.519 --> 00:15:03.440 So the scaling story here is quite obvious and from the last two examples you can probably tell as the model gets smarter your team will become more specialized it will become more collaborative s164 00:15:03.440 --> 00:15:08.639 and ultimately that means it'll be capable of getting more complex work done for you. s165 00:15:08.639 --> 00:15:14.079 And now the second is this new concept and we've alluded to it slightly in the past, but it's called sidecars. s166 00:15:14.079 --> 00:15:17.279 This is a new plug-in protocol that we're bringing to anti-gravity. s167 00:15:17.279 --> 00:15:21.120 A sidecar process is essentially it is a sidecar process. s168 00:15:21.120 --> 00:15:27.440 The naming sort of reflects what's going on under the hood, but it's a longived utility and it's responsible for listening. s169 00:15:27.440 --> 00:15:32.720 It allows the model to listen to the outside world and set up its own triggers for things that might happen. s170 00:15:32.720 --> 00:15:34.720 For example, this could be SMS messages. s171 00:15:34.720 --> 00:15:40.000 This could be web hooks, cron jobs, hooking it up to GitHub PRs, the the list goes on and on. s172 00:15:40.000 --> 00:15:42.639 But this is a generic plug-in primitive. s173 00:15:42.639 --> 00:15:45.920 Anti-gravity already uses sidecars for things that are timebased. s174 00:15:45.920 --> 00:15:48.959 This is where the the scheduled task cron concept comes from. s175 00:15:48.959 --> 00:15:52.000 Um, but under the hood, this is all this new sidecar primitive. s176 00:15:52.000 --> 00:15:57.279 So, we'll be releasing the spec for this so that you all can build on top of this new primitive um later this summer. s177 00:15:57.279 --> 00:16:02.800 But there are some really, really cool ways that people internally have been using this sort of concept. s178 00:16:02.800 --> 00:16:05.839 And the third and final primitive is generative UI. s179 00:16:05.839 --> 00:16:12.240 So we hypothesize that human written specialized UIs are kind of dead. s180 00:16:12.240 --> 00:16:16.639 Gemini Flash on anti-gravity clocks in at almost 900 tokens a second. s181 00:16:16.639 --> 00:16:19.839 This is 10x faster than a lot of the other frontier model experiences. s182 00:16:19.839 --> 00:16:29.920 And in a matter of seconds, you're able to go from whatever you thinking inside of your head into a prompt into a use case that is designed and embedded inside of your conversation view perfectly. s183 00:16:29.920 --> 00:16:36.880 And rather than rely on templates or even HTML files, anti-gravity can render your generated UI in line. s184 00:16:36.880 --> 00:16:39.199 So you can do things like this and play Doom. s185 00:16:39.199 --> 00:16:52.079 But this also extends to things like bar charts, graphs, um, tables, anything that you would want to interact with and maybe inspect a little bit further than just a markdown file or just a conversation. s186 00:16:53.199 --> 00:16:58.240 And generative UI in many ways reminds me of the quote that the late Steve Jobs said when unveiling the iPhone. s187 00:16:58.240 --> 00:17:02.560 He justifies the removal of the keyboard and says they all have these keyboards. s188 00:17:02.560 --> 00:17:05.439 They are there whether you need them or not. s189 00:17:05.439 --> 00:17:10.720 And they all have these control buttons that are fixed in plastic and are the same for every application. s190 00:17:10.720 --> 00:17:16.000 In an analogous way, we built our product to dynamically scale with the needs of the agent. s191 00:17:16.000 --> 00:17:22.160 We skipped the heavy infrastructure and mechanical UIs in favor of sidecars and generative UI. s192 00:17:22.160 --> 00:17:27.600 And that creates a product experience that is not fixed in plastic. s193 00:17:27.679 --> 00:17:33.919 So, sub aents, sidecar triggers, and generative UI are the latest primitives that are powering anti-gravity. s194 00:17:33.919 --> 00:17:37.280 We've tried our best to stay out of the way and let the model cook. s195 00:17:37.280 --> 00:17:45.520 And if you're building a product around an agent, you should consider what are the primitives that are in my product and how might they scale with the model's intelligence. s196 00:17:45.520 --> 00:17:50.559 We all are familiar with shipping features is now quite easy with all of these new tools. s197 00:17:50.559 --> 00:18:00.240 And it's about deciding what features to actually add um so that the model so that your product can scale with the next release of the next model which will inevitably be faster, s198 00:18:00.240 --> 00:18:01.840 better, and cheaper. s199 00:18:01.840 --> 00:18:08.720 And so with the right primitives, you as a builder or you as a product owner, you might be surprised at what the models can do. s200 00:18:08.720 --> 00:18:13.919 And in classic fashion, I'm going to keep using the slide until we've actually conquered the TPU crunch. s201 00:18:13.919 --> 00:18:15.520 So you can find me on Twitter. s202 00:18:15.520 --> 00:18:17.039 Uh you can DM me for feedback. s203 00:18:17.039 --> 00:18:20.720 We're always looking for new ideas on how to build the latest and greatest. s204 00:18:20.720 --> 00:18:21.600 Thank you for watching. s205 00:18:21.600 --> 00:18:23.039 Thank you Swix and Ben for having me. s206 00:18:23.039 --> 00:18:24.000 It's always a joy to be here. s207 00:18:24.000 --> 00:18:26.640 And I'll be at the anti-gravity booth if you want to talk further. s208 00:18:26.640 --> 00:18:29.360 If you want to get to know the product a bit more, some team members will be there. s209 00:18:29.360 --> 00:18:30.799 So, thank you so much for your time. s210 00:18:30.799 --> 00:18:33.600 Excited to meet you all. s211 00:18:43.039 --> 00:18:45.039 Yeah. s212 00:18:53.760 --> 00:18:55.760 Heat.