Reacting to Google's "Gemini at Work" Launch: BEST AI PLATFORM!!

0:00All right, we're live. Okay. Try live panel.

0:47What going on here?

1:14past custom browser. See, I'm figuring this out, man. We got the chat here. What up, guys? Just uh doing a bit of stream maintenance here and then we'll get going. Go back here on that web page. Look at the right hand column where the live chat box is at the top of the Mac menu bar docs.

2:13Oh,

2:30come on. Yep. Got it.

2:42All right, guys. We're here with uh CTO Jesse. What up, guys? And we're just going to be having a bit of a vibe out today. Um, bunch of new releases this week we can go through. I got some interesting stuff on like the Higsfield video generation side that I want to have a jam on. Um, but I'll be answering your questions, vibing out, uh, and just cooking up. So, if you guys got any questions along the way, if you guys have anything interesting you want to share or you have a um you have a new release or something you want us to go through and inspect, then um just drop it in the

3:11chat. But I think good to get going.

3:27Um, what do I want to go with this?

3:47Boom. What? There we go. All right. So, first thing I want to go through is this Gemini at work, which is probably the biggest biggest thing. They got to like release video.

4:26So that's an hour long. Thank you too. Hey, there's your boy. Nice. Yeah.

4:45All right, let's watch a little bit of this. Can you guys hear that a little bit? CEO. My guy Sunda.

5:05Good morning everyone. Welcome Gemini at work. We glad you're here with us at hanger one now part of the NASA as research center. This building is so big that a cloud spawned inside it naturally on its own. No migration plans, no procurement cycle, no six month. This is also where scientists and engineers gathered to tackle problems that nobody have solved. So it's a very inspiring and exciting setting for today as we work together to enable a new frontier of work. At Google

5:37it

5:45moving from a period of experimentation to fully integrating AI into your core business. This rapid scale is only possible because of Google's full stack approach. You've heard me talk about it before because it's so integral to how we are able to deliver for you all our customers. For example, our research and models are so foundational to our products and we're getting together here at an exciting moment. Last week, we announced our new frontier model, Gemini 4. It's an extremely capable model that's been built with about I hadn't heard about it and we working to get it to more people as safely as

6:12possible. So this is like their fable. Yeah. Orgon delivers frontier performance in complex workflows across knowledge work, real world software engineering and cyber security defense. Going deeper on knowledge work where Argon really excels. Here you can see Argon leads on a balance index which measures economic impact across finance, coding, legal and tax world. On coding, you can see our lead here on deep street which measures real world long horizon software engineering. And it's not just what the benchmarks are showing. Googlers are

6:36feeling the difference internally too. It's so funny how these how these benchmarks work. It's like every week like they're just all trying to step on each other. It's not just improving velocity, it's improving quality out of the red rolls by nearly a third. It's also enabling Googan

6:56millions of lines of code. Normally this would take months. They've done it in weeks. Two major wins. First, it enables a new class of UI experiences in Chrome. So a live web page can move fluidly between a tab, a side panel, a floating window, or other experiences without being reload. Second, it helps eliminate an entire category of life cycle crashes

7:12to make the Chrome browser more stable. In addition to knowledge work and software engineering, Argon is state-of-the-art at cyber. It's so good in fact that this is already using Argon for protecting critical infrastructure. In one example, the model uncovered and helped fix a critical vulnerability in healthcare software used by hospitals worldwide. Something previous models have missed. I'm really glad some of our cloud customers, many of whom are here today, have been testing Gemini 4 argon alongside us. Thank you. We've heard great feedback that the model excels at processing huge unstructured inputs into more cohesive outputs and that it also executes complex enterprise workflows to completion. And today we are expanding

7:43to more customers around the world. As these examples show, we built our to be really capable across a range of domains important to how you all get worked up. These capabilities are also necessary to unlock the next era of agents for both businesses and consumers. We have 1 billion users now on the Gemini app and you're seeing a clear shift. They're not just asking Gemini to know things using Gemini to do things for businesses. You're already ahead of the third one agents. Last year at Gemini, we announced Gemini Enterprise, a front door for building agents for

8:08different purposes and tasks. Nearly 90% of Fortune 100 use Gemini Enterprise at work. So do thousands of startups, small businesses, and companies of all sizing it extensively. As the next step, we are evolving Gemini Enterprise to make how you get things done even simpler. Gemini will now become your universal agent for work. What up? Ready to go. Not only questions, you write the code complete longer and more complex. Great. You need to get the job

8:32done. And because it knows the organization, it works with the security administration and governance required by your company. Launching these types of powerful agents needs to be done safely. By bringing it to businesses first, we are able to solve the harder problems around security, scale, and performance. We're planning to roll this experience out to consumers soon. It's a really exciting time, and enterprise is once again at the forefront. To show you more, let me turn it over to Thomas Kuran, CEO of

8:52Google Cloud.

8:58They got closer to see. Oh yeah. Thank you, Sund. Now, Gemini is transforming how work is getting done. I want to start with what has already happened because it's further along than most people think. It's happening every second of the day somewhere around the world. Right now, it's evening in Europe. Nokia cuts telecom network

9:18troubleshooting time by up to 80%. Commerce Bank has made quality assurance for document development in starting manufacturing data agents reduce time to go from data collection to insight from two weeks to instantaneous which has given them back 450,000 hours a year all back to business. I mean once they start heading the enterprise bro they have so much more upside art and office works across the ocean halfway around the world it's early afternoon in Brazil where global reviews add creative three times faster personaliz communications 80% it's interesting to see what benchmark they use to try to validate every cure recursion Today the new interface for work but the

10:06window is the easy part. The question is what sits behind it. Today generally becomes an agent. You give it objectives not just instructions. It plans the work uses custom skills and tools connects your systems and brings back something finished. Six things make it different from anything else in the market. First and most importantly it is one agent. Gemini answers your questions does your knowledge work brings your images and media writes and runs your code all in one window all with one. I mean that that being so fluently multimodal give it any role you want teammate or a specialist needs you can assign taskly or have it start

10:43on a schedule or based on an event you know with other providers chat and code they live in separate tabs they do not share history Gemini does it all in one window Gemini is one agent doing all of it so they're going to like mono a phone the web not like scoping Gemini to be specific That's still like a very big open debate, right? Like Yeah. One of the most annoying things for people is obviously having to chat to different chats and feel like that context gets lost. Yeah. But like if you can have one agent you consistently talk to and it can it can keep all those

11:14different threads open and not mix them up and separate a share. I do quite like the thought of having special specialized agents though agent working on his own because it runs in the cloud. The size of the job is not limited by the size of your desktop. When the job gets bigger and more complex than a single agent can perform, Gemini can dynamically create a roster of agents, assign each of them tasks, message them directly, and coordinate

11:34them until the work is completed. Oh, so it coordinates for you. It just acts as an orchestrator. It sends the work to you don't need to interface with those guys. Yeah. So it manages the specialist for you. Everywhere the work happens in the Gemini app, in Google Workspace, Slack, Microsoft Office. It holds this information as a single personation record. The longer you work with it, the better it gets to know you and how you

11:53work. Damn. Fifth, it inherits the controls you already have. who it is, what it that's huge. So like with our Google separate from your with our Google Workspace set up all of the like people that I we add in their identities, their emails, they like access everything that I like control within the Google admin panel. I can just it's just natively built into this named

12:18huge what and on whose behalf and where. Finally, it's not tied to a single model. Gemini is the agent. The model underneath is a separate choice and it is yours. From Argon at the frontier to Flash for Speed and to the latest models, other leading proprietary open model, they're not locking you in. Dang, they they get they get it. Google's not going to lock you in to their own

12:40models. I wonder if you can use a Max plan though if it's API. Well, how does he frame? It's not tied to a single model. Gemini is the agent. The model underneath is a separate choice and it is yours from Argon at the frontier. The latest clock models. Other leading proprietary and open models are also

12:57coming soon. Dang, that's pretty massive. Open models always the the first thing I say about these platforms is that they're going to be like, "Oh, but like I don't want to get logged into one model." And I think that's so crazy. I think Google knows that because like they historically haven't necessarily been like at the absolute frontier like their coding models and they're right like Gemini hasn't been really fighting for that top spot apart from maybe on the image uh image gen and maybe like video understanding so they know that if they lock you into their models then know like why would I

13:26use that? Yeah. And if they could focus on that deep integration with uh a company's like Google Workspace basically. Yeah. all the data they have as like Google as a search engine. Yeah, this is what I I heard that from the All-In podcast is that Google's being like, look at like, yeah, we might we might win the model race like, but we're going to keep making trying to make frontier ones, but Google has more to gain from being like all these open source ones coming up and being able to run them on their as like Google Cloud and being able to run all these models and make money as the like the inference

13:59provider. biggest thing I see with Google is like um one of the biggest bottlenecks that most people have been encountering is context management. Like that's what all of them are trying to solve and all of them are trying to get better at. If they can truly do this where you can have like one chat panel that you talk to and it's intelligent enough to know everything about you from previous conversations and not get mixed up and be able to keep all these open threads open and like tie things up and then create specialized agents to handle the work. And like if it can actually do all

14:26the orchestration, that's what most people are trying to solve anyways. Like all these second brain things, all these like crazy orbs people make and all that, that's all trying to solve one problem, which is like how well does your agent understand you and how well can it juggle multiple threads of information. So like if they can really solve that moat, like that's a massive step in this

14:45whole AI use case. Yeah. I mean, Google's been one of my top picks for like a year or two now. That's one of my biggest holdings personally in my stock portfolio. Um, I I sold a bunch off like up here, but I think you could just hold this bad boy for I mean Warren like Google stock. There's a reason your boy Uncle Warren be dropping bags into it. Wait, where is

15:10it? Uh, holds nearly 106 million shares. So, they bought um expanded its stake by 83% in the second quarter 2026. Dang. Cuz they they see it. I do wonder how a guy like Uncle Uncle Warren is like if you're like is he struggling in this era? I guess we won't know for a while, but like I guess he's got the best analysts working for him. But for him not being able to maybe understand or grasp the technology, is he trusting people to make these calls from or has he got insider info from Google and they're like, "Trust me, bro. Like,

15:43we're going to crush it. Trust me, bro. Just trust me. Trust me this." Um, probably mix of both, I'm sure. You probably have to be a pretty cracked analyst to work at Brookshire Hathway. Yeah. But that's um that's interesting. And I mean if yeah if they're managing your whole second brain thing, they've got access to models. They can offer access to the cheaper ones and they're not going to absolutely rinse you on using the Frontier ones and you can say, "Look, I want to save 90% of my cost and use GLM or something." Uh then these guys are running the game. Plus all your workspace, all your perms, Drive, Docs,

16:18um Google Search baked into that. They've got the best like multimodal models. It's not not not looking too bad. Yeah. And they're multimodal stuff as well. Like that's Oh, I think they just dropped that, right? They basically dropped the 12 legs. Yeah, the Gemma the Gemma 21 or something like that. Um and it's small enough, I believe,

16:36that it can run on your phone, too. Like I was reading about it the other day and um that could be something really interesting. Mhm. Okay. time blending frontier models to drive sales insights across millions of merchants at PayPal ra 10 million multimodel requests every week. Now, let me show you what Gemini can do for people inside your company. Please

16:57welcome Adam Fisher. Thank you, Thomas. Great to be with you all today. Earlier this year, we made the lead from chatting with models, having agents do the work. Today, the Gemini agent brings that magic to everyone in the enterprise. Use a task. The Gemini agent does the work in its own secure computer. I'm going to walk you through a few examples of how I'm using Gemini in hopes it provides you some inspiration. This is the Gemini agent on my desktop. I [clears throat] can also work from web mobile CLI and other desktop or create code. I'll begin where many of us do. Let's for this hunting through files and folders,

17:33reading 20 chat spaces, and then massaging pixels around a slide for hours. But not with Gemini. A lot of our customers work in Microsoft. So, let's build there. And full disclosure, you're going to watch me paste my prompts today so you don't have to watch me type. Let's get into it. I tell Gemini what I need. PowerPoint, a

17:47revenue model, and a product video. Gemini understands my goal, builds a plan, and creates its own sub agency to get started. While that's running, let me show you around. Back to a new task. Where the hot box, but here I have the ability to pick which model I use. By default, auto mode picks the best model for each task to balance speed, quality, and cost. I can also choose Gemini or third party models. I can attach files or folders or even a project for specific work streams. A project is a combination of the right files, the right skills, and even the right agents that help do the

18:12job. For example, a Q4 product launch. Oh, that's skills. These are simple instructions repository or I can easily make these on my own. They just copy the whole copy like Google Workspace, Microsoft 365, Slack. It's all turning into the same thing right here in my test. That job we launched is still running. So, where does the where's the edge? Um, I think part of it is in the data like, and I think Gemini, I think Google has a pretty big moat with this is just the amount of data that they have. Um, it's interesting. I'm curious to see what they do with like custom spaces or

18:48web apps cuz um, when OpenAI did their thing recently, they've got this whole like notion copy that they're going for where you can like you can spin up a web page. It's all hosted there. Like that adds like a deep layer of customization that businesses I think want, but they want custom dashboards and all that. I'm curious to see if they try to do the same thing or what direction

19:06they go. They got Google Cloud, bro. They can just host your stuff. Like the whole point of reason why I invested in Google and I was like, bro, they own everything. They got they got all of it and they can they can make they've got all the YouTube data and all the search data and like so much distribution as well and everyone's already on all of their different apps. The workspace, you just plug this straight into the workspace. I mean, the one thing that the the OpenAI one it was a kind of a jump forward is the having the coding agent right next to

19:36like coding agent being able to run it. Um, and that was one of the gripes that I'd have is like I'm always working codeex. I'm working in Claude and then there's all this other stuff kind of built around but now this is like clearly just all in one place. If it's able to actually handle the orchestration layer pretty good and it can use coding agents to build stuff when you tell it to do it and it can understand that context properly, that's going to be really interesting because that kind of alleviates that whole need because most people who are building using this stuff, they're not like an

20:00engineer. Most people just like they maybe they want to spin up a quick demo or an MVP or a quick landing page and that doesn't require a crazy amount of um like a crazy strong coding agent. So if this can if this can orchestrate that properly and just spin something up, throw it on G um use that auto the auto like model

20:17routing. Yeah. Um yeah. I mean cuz this is this is essentially like the the base is now a a coding agent like in a harness. That sort of thing. It's like it's Gemini is the is the harness and then it just can swap in and out the model depending on the task. Um interesting. I mean the biggest thing I think in terms of this race is going to be cost as well, right? Like as long as if they can actually give me access to other

20:45model providers in there. Yeah. When I want to use Frontier Intelligence from other providers, but then I have like Gemini at a really freaking solid price that I know I can just be like cranking out a lot of my code with. Um it is going to come down to and these guys have so much cash they can undercut them, make their I mean Google models have always been cheap. Um, so I feel

21:06like cost is going to be a big thing. Like looking look at our business. I'm here like munching through the the business plan last weekend that I signed up for for the codeex for the um opening I released. I've already munched through that and a reset. Yeah. Uh, and in here I could be able to go a lot longer. So So let's diving into the output. I really want to

21:29show you Gemini's thinking process. Guys, I'll get to questions after we've skimmed through this with the whole session. So just keep keep putting your questions in and I'll rip through them afterwards. Um I just want to get through this presentation. We can see that understood my goal, built a plan and then started delegating the sub agents. It loaded skills like this corporate brand kit and it even wrote code to build these materials. All

21:51right, let's look at that. Huh. So Gemini didn't just write bolts. It used my brand guidelines to create a beautiful PowerPoint. See here, it pulled the right numbers from that Excel model that we built. Knew my timeline and even used Omni to generate product launch video. Let's take a listen. Dang. Omni is what? It's their Google. It's their

22:08new new V model. Not bad for they've had it for a little while, but I don't know. I think it's just a creative model. Taking me half a day to do, but it's like available right for me. This is Google Workspace. Let's head over in workspace. Gemini can work like a co-orker. It has its own workspace account, its own email address, its own context. Team works with it. The way they work with well, anyone. Tag it,

22:28email it, share with it. What the hell? For a second, wait, bro. I've been saying this ever since OpenCloud came out that it's like if you want to avoid things like this breaking your whole email, like you give it its own separate email address and let it like and just BCC it in like like an employee. So that's crazy. They natively

22:47added that. They've got See, look, they've still got distinct agents. This is built into the web app though of of Gmail or that's the chat. Maybe this is just Google Workspace. It has its own workspace account, its own email address, item

23:11and my team works with it in the way they work with well anyone. Tag it, email it, share with it, but it only knows when we can. To demonstrate, we've added an events agent right here in the group chat. I'll tag it in a comment now.

23:27I'm seeing help with our regional launch. Just like that, the agent jumps in to give us a hand. It's using skills and connectors that we provided to pull the right context from previous work. Looks like it pulled together the work stream status check. Let's see and the risk assessment. Let's look at the document. All right, I already see that I love it because it's using a template that I chose. I have exact summary. It calls out the action items owners and here even that detailed checklist, but I'm seeing it missed some of the recent meeting updates I want included. So, I'll drop a comment

23:53and this is the Open AI thing now. Yeah. watch as the agent goes to work. And that's notion. Yeah. I mean, here's the thing. Like, I've always thought Notion is a cleaner UI than like Google Docs. Like, Google Docs is that thing that it's just, bro, Google's like it's like GHL to some extent. Like GHL, I don't like everything. And I don't like what I don't Yeah, I don't really like Google

24:25Doc. That's what I'm saying. And like I just it doesn't feel like a good UI. So I don't think they have like the user experience with that. But like the deep integration across all their platforms. Yeah. 100%. It's already integrated with Google Drive. It gets its own Google Drive account. It seems like if you add it to it workspace, but like the Google Docs UI is kind of [ __ ] Yeah. I mean after Notion sort of showed us a new way of writing documents and like the block style, it was hard to go back, you know? I just look at docs and they're just [ __ ] ugly to me. I

24:53agree. Did you see that? Uh so now Gemini can delegate work to claude harness and worksheet. Uh no, Gemini is the harness. It can use the claude model to do its work. So if you like a lot of the concern people have about these platforms and I mean I've been talking about platforms and like the shift from harnesses to platforms where you've got multiple different harnesses and people and agents working together in one sort of space is that they would get vendor lock in and you'd be stuck at stuck using uh Google models or Gemini models and you put your whole business on there and take all this

25:26effort and time to to get on onto that platform and then Google drops the ball on the models and now you're using subpar intelligence while your competitors are absolutely [ __ ] munching you using the the best stuff that's coming out. But if you can have a platform like this and they can be smart and say, "Look, we're not going to we don't need to win the model thing. We don't need to force them to use our stuff. We can provide the

25:48everything around it." Yeah. It's like the deep integration with the way the companies already work. And because a lot of businesses rely heavily on Google Workspace, if they can just integrate their harness and make the agent, whatever AI model you're using, be able to touch and see everything going on in your business. um

26:04like nobody's going to want to leave. They're not going to want to leave Google Workspace. I think that's what they want mostly. They don't really care about people using their bottles. They want people to continue to use Google Workspace. Yeah. You know what would be interesting though is if um if it ends up being where like Google like the whole Gemini workspace thing is better for like enterprises and we find that smaller businesses use like different harnesses like maybe small mid-tier ones are using open AI and then like bigger companies end up using Google because they're already so deeply integrated and people who work spend a lot of time in

26:36corporate life are familiar with Google Docs they're using it all the time like oh this makes a lot of sense for how I already work but Yeah, smaller businesses just use like notion. That's what we realized early on with um when we're doing our dev projects at Morningside. were like, "Oh, you're not going to for most customers, you're not going to convince them to come off the ERP system or like the whole freaking infrastructure that the business is running on that they've spent millions of dollars to set up and get working on to manage the access and the emails and the permissions and the docs and all of

27:06that effort that's gone into this kind of clunky for us as like small businesses, digitally native, AI native seems ridiculous, but that's how a huge portion of like the companies generating most of the revenue on the on the planet run and small businesses who are like, "Oh, we use Slack and Notion and like all this." It's just not how most businesses run. And we're like, we're these companies were just much rather sit on their hands and wait for this stuff to come to them than to make I mean, we were able to like nibble off the edges of these bigger companies in mid-market that were like, "Hey, we want

27:35to reach out and try to build some out. Then you got to integrate into the system, but in reality, they're just going to sit on their hands and wait until this stuff comes." And it finally is here. Yeah. Yeah, this is what I'm thinking too when it comes to like how to set this up for businesses like how do you monetize this in an agency format like a lot of people have noticed uh if they don't understand how these businesses work you just want to build something and try to p like push [clears throat] it onto a business of any size like yo I built this thing use it where I think a

28:01lot of it is going to come from being like a one person we've talked about this before like a oneperson AI consultancy where like the goal is understanding how platforms like this how deeply they're integrated and understanding how a business works and figuring out what is the best fit Like maybe you're working with a bigger enterprise clients or you're talking to one and like it's better to try to get them to understand how to use the Gemini workspace and like deeply integrate that, understand their business, figure out how to optimize it with the platform they're already on rather than trying to get them to move to some custom thing or

28:29try to move to OpenAI. Mhm. Yeah. Being real good on these platforms and I think I claim I called it at the start of the year. You guys can go to my 2026 predictions. Uh, I said that I think it's going to keep shifting towards like training being the they use like the analog is like a flying fox. So like you're coming down a flying fox. Is that what you call them

28:48in America? You told me. No. Yeah. Like when you at a on a playground, put that like thing under your and you jump off and it goes no [ __ ] clue. Do you not have those on the playground? No, I don't think so. It's like a It's a zip line, but you put

29:01you sit on it. I guess you sit on You like jump in like water or something like that? No, no, no. That's just like a line. Look this up. Flying fox. New Zealand's got some crazy [clears throat] things. I know. Like playground. Playground. Yeah. This B. Yeah. Yeah. No. One of these flying fox. Yeah. So, you know, at the end, this must just be a New Zealand thing or something. Um, Mel, Australia, New Zealand thing. No, I

29:34never seen one of these. Damn, that childhood must have been whack, bro. [laughter] But, you know, at the end of it, it's got like the It like slows you down. Oh, yeah. I feel like with how fast development and consulting's moving, it's like AI is helping consult and audit the business way faster and build out proposals and

29:53sort of map the opportunities. Oh, [ __ ] I went on one of these in Cape Town like 6 months ago. Must be a sudden hemisphere thing. But as the auditing becomes faster with agents, you can just plug it into their Google platform audit. Um, I think one of the biggest use cases for people trying to run an agency is like figure out a system that can audit like you were saying like understand where you can like give a company or like whoever whatever person the company you're talking to have them interview them in depth and be able to extract as much information as possible with AI and

30:25also have that understand these platforms and where the biggest use cases are to like tie that together and then operate as a as a consultant and like help them string it up. Mhm. But like what after the the consulting and the auditing piece gets heavily assisted by agents and then the development like you can basically go from if you fully map the context of the business and it understands all the integrations and stuff you can go from auditing press a button using an implement loop or or specking sheet create a PR fully plan the build out figure out all the testing and evals and data it's going to need to

30:54be able to push it to the like a decent ready for a review standpoint. Yeah. And then you could just kick off the development of of those pieces and then the auditing and the development is moving very fast but then at the end of the day as we see in the makeovers as well it hits the training and like the the human piece of like okay now you've actually got to get humans to use these systems is where it like slows down and it goes like it's it's no longer AI pace it's human pace and so that's like the last bastion of value for AI agencies is can you sure you can do all

31:24the auditing and development that's kind of like that's part of the value you provide But that should be very easy for you now. And a lot of your time is going to be spent on refining that that end system to get it working really well. And if in conjunction with training the team and go back and forth between like training the team, making it work better, training the team, making it work better, uh, and getting it to actually stick and teaching them how to do the new workflows. So I think training like claimed at the start of the year is is going to be one of the

31:49kind of focuses of or last bastions of value for AI agencies as as the rest of the stack gets heavily AI assisted. Yeah. Yeah. And a lot of this, a lot of it really comes down to the whole like the projects tab you shared earlier with all the connectors, the skills. Like that's where a lot of the work goes into with something like this is just

32:05configuring all that. Yeah, it made both the suggested edit and commented reply under its own name. I didn't even have to touch the keyboard. So, this document is now ready for review. Let's ask Gemini to get the leads together. Gemini works like a well-informed assistant. I don't have to get names or addresses. It knows who my team is, what time zone they're in, and whether I usually work with their calendar delegate. It also gets my approval before it contacts anyone. Let's see what it's drafted. Nice and short. I love it. All louder. This is workspace intelligence in action. Behind the scenes, Gemini emails the leads on my

32:33behalf, suggesting dates and times that work, and finds one that is right for the internal and external teams. Now, normally we have to wait a long time for all these replies to come in, but I've got a great team backstage helping us out so we can move this along more quickly. And just like that, the meeting

32:44is confirmed. Gemini handled all the scheduling. I have to do and thank goodness because I've got a lot going on today. And this is why I'm also happy Gemini is proactive. Let's head over to the inbox and check it out. Sorting through my inequality

33:04leadership meeting and our key stakeholder Matt. He's out of the office the day before. We need to make sure you have the right. It knows my calendar, my team, my files. I don't have anything to connect and no automations to set up. And it doesn't just surface this important information. If I click start with Gemini, it prepares an executive summary for Matt with a light context and highlights the risks and updates he needs to know. Now, this is going to take a few minutes to finish, so we're not going to wait around. But if I need to step away from my desk, I can pick this project up

33:26directly from my phone. Let me show you. We already looked at chat, so I'm going to show you this in Slack now. Uhoh. I get a [clears throat] lot of important customer feedback. So, I'm going to have Gemini. That's all the same. Gemini is on it. Oh, here we go. 35 comments across three threads. This is a lot. We need to make sure this is

33:41included. Add this to the executive summary we worked on earlier. Great. Gemini's got it from here. Now, we've seen a few examples how Gemini works with existing tools, but sometimes the one that I need, it doesn't exist. Let's go back to the desktop. Here we go. All right. Every week, I spend hours pulling together updates across different sheets and Jira blockers and other data sources to make a status report. How can I help save some time on this? Let's see what thinks. It's got three suggestions for me, but the one

34:09recommends a live micro. Sounds good. Let's try it. I want this to be fast, so let's just go with the HTML markup. Without writing or even reading a single line of code, I can quickly prototype a tool that gives me insight that was previously available only across multiple other data sources. Oh, there it is. It's exactly what I was talking about earlier. That's crazy. Single use or

34:26like short-term use software. Definitely going to be easier for the team to manage. Well, Gemini can write all the code and deploy this micro site to production from here. I want to show you the same thing from the terminal. Same Gemini, same work just in the CLI. Gemini will spin up a dynamic team of sub agents. It'll use the skills I have to read the sheets, connect the data,

34:44and write the code. Once it finishes, I can also test and debug the system. It can keep the data up to date. And why are you showing me this? The bottom line here is you pick the surface agraph

35:04who I am and where I fit in my organization. Okay. I took notes on how and how that communication varies by stakeholder. And I learned my preferences. For me, this is never use an M dash. Obviously, obviously, executive summaries always have the above line up front. And this is where I think things get really powerful. Let me show you an example of

35:20how I'm using this with a schedule job. Every [clears throat] morning, Gemini reads my new emails and drafts are reply as part of a wake up routine. If I review these emails and edit one over coffee and then send it, Gemini reviews the differences and improves the next day. As Gemini learns my style, it needs fewer nudges and can do more on my behalf in the background while I focus on other priorities. This is a

35:36compounding benefit for your business. But if this room has any CIOS, then I'm sure there are at least a few breaking out in no cold sweat. Which is why we need to talk about one final component, governance. While Gemini is personalized to me as a builder, have complete visibility and control across the entire

35:49company. In the console, everything users and agents do. Each agent has unique identity, so ads my agents take. For example, here's a trace that shows exactly what an agent did on previous run. You can see here the agents that were invoked calls a policy tool calls. Holy moly, that's so nice that they have that deep level of observability. across the organiz

36:13platform. It looks like it. Yeah, I guess you're not going to be looking at it. You'll probably have agents at a random. That's what I was going to say. As long as it's at least there and like if agents can read it, then any external email has a human in the loop review or like automatically blocking the sharing

36:25of sensitive data. So, they've got policies. This is such a massive like um opportunity for people who want to like do some kind of AI consultancy business or AI agency because like this shit's complicated for a lot of people in businesses to try to figure out. So like the what policies they should be putting in

36:42place. Yeah. So the real moat for people that I see right now if you want to like make money with this is like be the subject matter expert that figures this stuff out and you can like talk to businesses on it and help them figure it out. Yeah. I think you'd have to roll in with a It's just [ __ ] hard cuz it keeps changing so much. But I think now that

36:57we've got these platforms taking shape. Yeah. But it it's like you need to have a playbook of saying, "Okay, here's how we audit your business. Here's how we like prioritize those and here's things we set up first. Okay, we plug your whole team in onto this." And I still think from our experience, just getting the

37:12entire team onto a workspace. Yeah. fully contextualized, trained on how to use uh agents and build their own skills and build short-term like single use software. I think just that's the the first bit of lift and get all of that working and then you can build these layers in and say okay next we need to uh build our our governance policies and stuff and we need to build the layers of management on top of this before you can start to crack into the the big builds and custom builds on top. I I still think this installs this is essentially like a productized or like platform based AIOS. This is what

37:46we've been building and duct taping together for the past 6 months and now it's coming in the form of a platform that where these big like this is the stuff we were having to like the permissions and access and stuff we're having to build in GitHub repos and like using these like priming commands to determine who's in which seat who role and it's all going to be wiped out and it's going to come through stuff like this and so you need to be have a have a playbook to roll into a company and say we take this approach first we set you up on here we set up these agents we

38:14have this training program. Um, and yeah, being able to effectively like having a mini course as part of your agency that you plug in uh, and put people through you've got testing or some way of but again, how do you scale that up to like 100 employees is the question. How would you train an

38:30organization of that size? You probably have to do like train like weekly training sessions for a certain amount of time. Yeah. I I talked to a guy that did something like this with uh, I think it was a co-orker. He just did he ran basically like training sessions. Mhm. Um he's making ha bread doing it too. Yeah. I mean cooworkers cooworkers was the first one really where people could go on there and and they actually had some governance policies and stuff in there I think. Um but this is this is interesting and this is again like as an agency do you specialize in we work with Google based

39:00businesses which is probably going to be the biggest category if you want to work with like SMBs. Um I mean cuz what other like personally I don't want to go into Microsoft side. Yeah. [ __ ] that. Uh, so I'd rather work with Google. I'm very familiar with it. Yeah. I mean, um, that's I I don't know if that's quite determined yet because it's going to be it's going to be a combination of like you could become a subject matter expert just in the Google Workspace and like really drill down on that and try to figure out what like what the TAM is like what um how many

39:30businesses use this already or like what scale fits this right and then just target that specifically. Or you could become just more a generalist between the different platforms like OpenAI and Google and whatever else is out there like things like Victor and like understand them all and try to figure out what fits a specific company size. But like things like this like their documentation is probably super thorough and like if you were to just tell like Claude whatever like hey what's going on with this it probably won't know like but you could take all the documentation compile it into a skill where anytime you want to understand this you just run

40:01that skill and it primes your session context with all the documentation for how this works and you could like query your questions about like how business works and where it fits together. I mean the the exciting part I I saw like a glimpse of where this stuff can go when when I was doing the Kylon setup and had a Kylon Kylon has a CLI you was oh we can't really be like but um yeah we'll be maybe 30 40 um it looks good sign me thank you um where was it Kylon CL LI means that programmatically you can manage your client setup. So you could have a dashboard that basically

40:45you build a custom dashboard for managing your different clients from afar and then you're able to via some CLI be able to hook in and basically set up new policies or analyze things how they've been going and instead of having to come in here and do all of this stuff or use computer use it's like programmatically you get to tap like hook into your client system fully understand what's going on make decisions analyze it. I remember when we when we were doing the setup for Josh, that was super convenient having the

41:10whole CLI baked in. Finally, admins control project and user spend. So tasks and background routines never overrun the budget. For example, here I'm getting pretty close. You can see what portion of the budget spends, what my fixed caps are. Oh, that's still a project. Yeah, that's crazy. So every check this is like the whole AI ROI thing that has been an issue for so

41:31long. Yeah. Is that you couldn't scope it or quantify it. Now you say okay all your work related to this project is done in this folder then you can start to quantify the spend okay image gen video gen coding all of that grouped into one and then you can say okay well that say if we're launching a new offer or we're putting out say Sean editing a a a makeover video we can say work in this project we see you spend $1,000 in credits okay well what was the ROI on that like can we start to track that better so I'm not surprised in retrospect that

42:02they've got such deep observability already like baked into it. Yeah, I think they just need to make it Well, Google Cloud's already always been so deep. Yeah, it's just making it a bit more like as long as that data is there, you can pull it and do nicer stuff with it. Hopefully can operate it programmatically through

42:16Gemini as well as like an admin. Every action is trapped, every policy enforced, every budget is kept in check. That's how you unlock the full power of Gemini. Gemini has completely changed the way I work. The Gemini agent is available across different services, learns my preferences, and runs. And it's not just for me. Everyone on my team can use the skills, tools, and preferences they need. And whatever the role, it starts the same way. You spread the task. The agent does the work. Back

42:39to you, Thomas. Thank you so much, Ed. One agent, five services, desktop, web, CLI, mobile, browser. One agent, three types of work. Chat, pass, and code. Over 50 companies are running it today. adopting an agent-led strategy on successfully transition 24 core enterprise services to Google cutting migration time from three months to two weeks per service. In addition, they're testing the demi agent in early access to boost process optimization selecting the best model for each use case. This adjective first approach has teams reporting meaningful productivity gains keeping the teams focused on innovating the sportsware industry with a customercentric approach. Now speed is just one kind of number. The one I get asked most about

43:22is revenue. Alta Beauty used Gemini Enterprise to build Alta AI to guide guests through more than 30,000 different products. Customers can ask questions as they shop. They receive personalized guidance and tailored product recommendations. As a result, this has driven a three times higher conversion rate than traditional browsing on their site. But the prompt window is becoming the interface, not one of several, but the interface for getting done. Everything we build from here assumes it. You just watch the Gemini agent in action, already knowing your business, your data, and the answers you wrote months ago. Nobody

43:51manually loaded everything. That's that's just it. We were ahead of the game six months ago. You know your business, your data, and the answer. What you say? But the interface for getting work done. This is the interface for getting work done. Pull [clears throat] up my video from like 6 months ago. And I was like, this is we've been doing this for 6 months. This is the only place we work with include code of codeex. Plug all your stuff and get your data in one

44:13place context. Yeah. And been ripping in in a in a coding agent kind of thing like this. So, uh I wonder like when the next vision of the future's going to hit us. Like where does it where does it go? I guess this is the this is the next leg for the next couple of years I'd say. Um it'll just improve on this format. This

44:37is this is the new work format clearly. I think it'll just get more and more aentic where you've got agents that are able to do more work in the background to each other as well. the way of it surfacing the stuff that needs human approval is going to get better and better and like your workflow of how to manage all of these like small decisions being made and stuff and how that gets approved and the the human and agent collaboration is just going to have to

45:02get more and more dialed. Everything we build from here assumes it. You just watch the Gemini agent in action already knowing your business, your data and the answers you wrote months ago. Nobody manually loaded anything. The Gemini agent comes with three important capabilities to connect to your enterprise systems, understand the information in them and let it run a

45:23process end to end on your behalf. First, skills. The Gemini agent comes with a broad set of skills which are reusable sets of instructions, knowledge or workflows stored as markdown files or modular prompts that teach it how to perform specific multi-step tasks. Gemini ships with a broad skills library and each person can build their own skills and publish

45:39them. That's actually mega helpful. I found like with Kylon they had wasn't a skills library but it had an integrations library. So that already built in actually did have skills but when they've done the work of being like okay here's like the aptify skill here's the Apollo skill here's and they bake in a lot of stuff in the default agent that's what goes from just some like open claw based thing to like this thing is really capable at doing a lot of things like generating video or like UI frontend design um copywriting all these kind of things that you'd want this general employee to be good at

46:13this regry or any other user or agent can find them. Each time Gemini uses a skill, it learns from it, improving its consistency and following the same standard procedure every time. It also improves efficiency by using the right skills, saving you tokens and improving accuracy. Second, tools. Tools are how the demo reaches the software your company already. Workspace, Microsoft Office, Teams and Slack for collaboration, Gro Confluence and Git for building, Salesforce and Service Now for running the business. It can also connect securely to any MCP server inside your company or the agent. And as with skills, your teams can build their own tools and publish them in the

46:45register. Third, context. Generally learns from each question you ask it, each tool it uses, each objective you assign it, and uses your personal context and your organization context to personalize and improve the quality of it responses. Nowhere is that clearer than Google Workspace. Jet works directly inside Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar. The same memory, skills, and controls it has everywhere else because it already knows your calendar, your team, and your project. It is brief before you start and it puts that context to work suggesting what you can hand off and surfacing the email and chat messages that matter most to you. As Gemini

47:19works, it maintains four kinds of memory. Section memory for the task directly in front of it, semantic memory, which is a structured knowledge base that builds as it works. Procedural memory for how a job gets done. And episodic memory of everything that it has done before. They've really thought through the memory there. That's I wonder. Romantic is the crazy one that they're like picking certain data types and semantically storing them and allowing the agent to retrie like that's what we built in the the Daniel Maker and we're like okay some of this data like your competitor post needs to be

47:48stored semantically. Well, this was we were literally just talking about this yesterday about like the whole second brain thing about how much of it is just marketing hype and how much of it is like once companies like this really figure out the memory it's not going to be some crazy visual crazy thing that people are looking at. It's going to be something you don't think about because when you talk to an agent, it just knows what to say next. It knows what's going on. It's not going to be some crazy integrated

48:11cortex with this 3D visual flashing orb. We might still sell that, Jason. Just flexing on that. Just lost us 100 bands. You go, "Damn, that was my offer. I was launching next week, bro." No, it's just so cool, though. It's cool to be able to see it. It's a good visual hook, you know? like, "Yeah, this is what we're going to build, but like you really don't need

48:33any of that." Yeah. I mean, having just a visualization like could be cool, but Yeah. Um, yeah. I mean, my question is just like if I'm like, can I bring my own is it just like if you want any other fancy stuff on the side, get a Superbase account set up, connect the MCP, and then it as long as it's indexed at and

48:52knows it's available. Yeah. Um, just like how you you could bring like any kind of external data store and plug it into the into the workspace. It'll be interesting to see how good this actually is though. Like is this just something they're like, "Hey, this is this is what we're releasing. It's going to be great and then like maybe it sucks or is it actually going to be pretty good?" Back

49:10up. catch brother. In addition to providing you a powerful Gemini agent, we're also specializing each of these capabilities for specific domains, starting with data analytics to help everyone in your organization get reliable operational reports to monitor your business. Users can ask Gemini their questions in plain language. The Gemini agent uses specific skills with BigQuery and a knowledge catalog to generate and save the query. Once done, you can run the report without incurring the cost of tokens and get a reliable answer every single time. We're also extending the Gemini agent with machine learning skills and tools to back every data engineer with a team of data engineers describe the outcome they

49:44want. Gemini generates Pispark for the machine learning model provides local notebooks to edit the code, brings the model and automatically fixes issues. What? Damn. So I could spin up for like the content system or something. Yeah, auto machine learning. You could just chuck it in there and it's going to like just plug like ML. That's pretty tough, bro. Vibe machine learning is going to be like that's

50:05that's mad. Wow. To process data accurately from your many enterprise databases and applications, we've integrated a knowledge catalog with the knowledge catalog maps your business data definitions with no reind. That's just they're on a heater here, bro. Agent reads net margin. It knows exactly what your company means by that metric. It helps improve accuracy by knowing your schemas and business rules, making responses more reliable. You define those metrics just once and all Gemini agents use them. By the way, if that metric lives in data bricks in DBT and log ML or in SAP, Gemini will read those metrics wherever they're defined. Let's show you these data capabilities in

50:45action. Please welcome Gabe Weiss. Damn, I did say in my video like I said give Google like 3 to 6 months to respond. I thought that' take a bit longer like at least at the end of the year. They took a week and clap back on that. That is crazy. Um but yeah, they kind of mopping up OpenAI at this point. You could just see like the difference between like, hey, you got chat GPT has now got spaces and pages versus like org level access and perms and tracing and

51:19machine learning. Vibe machine learning pre-built skills. So let's say I'm a data scientist at a retail bank and I need to build an ML model to help marketing identify the customers most likely to switch their high rate credit card to new low rate personal loans. So I can just start by asking Gemini, what data do we have on customer accounts, card spending and past loan offers? Visualize my data landscape. Gemini makes this easy with a set of skills including pre-built data cloud skills to work across my data state. Our knowledge catalog has already registered my Google cloud and crosscloud data sets including rich metadata schemas and glosseries. Not

51:47only that, our developers can easily create, register and describe new data sets and business logic in knowledge catalog making both those data sets and their descriptions available to humans. Okay. So AWS and Salesforce CRM, Google Cloud, Lakehouse. So it's able to it's linking the the keys between all of them. I wonder if you can connect superbase to

52:07this as well. I mean, I guess for small businesses. Yeah. Um like us would have a super base cuz I'm think all the content data that I already have like the [ __ ] 80,000 posts I have and thousands of accounts. Like if I can string this up cuz it's slap like a bit of J action on top. Yeah. There's like 70 tables and 10 schemas I've got with a bunch of data classification I've done on Jev. Like if I can take that, run it through here and be able to like build a couple models,

52:32run machine learning on it. Mhm. Like that's actually crazy. I'm I'm going to experiment with that. Yeah. I mean, I guess you could have already done this beforehand. It's like I I did machine learning with Python when I was learning to code and you just get the data sets and then you just like can't remember what the what the libraries were for it but um yeah you just pass it in you'd run a function and it would then train and train and train and you'd see like the the accuracy of it improve and then you could just chuck any kind of input in as a as an input say like what's the

53:07prediction for this and it would predict it. So maybe this is a bit of like a a nothing burger, but we could have already done it for massive data sets and like companies that just have a bunch of like spread out data, especially because if they're mentioning AWS, they're not talking about small businesses. They're talking about more like enterprise things with massive data sets and and Gemini agents. So now let's put this

53:31data to work. Oh, hang on. And I might need to find the customers who could save money by switching our high rate card balance. Now we're going to watch Gemini as it generates highspar code that queries our customer and spend data sets from the lake house and pushes zero coffee joins to lightning engine using a pre-built

53:46data cloud sales. Here we are. Okay. So it's I I like that it's able to understand where that data is in the different store stores and then it's joining them creating a new I don't know what you say lighthouse or lakehouse I don't know above my selftaught pay grade this environment is live I can do stuff like edit the pipar code live line by line right in Gemini or if I want dude's out here editing code in 2026 can change this and follow the charts below it. Recompantly with this, we've already identified over 50,000 customers

54:23that can save $3,400 on average. Great. Great. Now I just need 52,000 customers. Build a machine learning model to identify which of these customers have a high likelihood of switching to our offering. Gemini query the knowledge catalog campaign history from the past three years stored in BigQuery using the pre-built data cloud skills. Gemini,

54:38this is not going to lie. Excuse me. Now we get this right. Just as an example, I can see here it knows the columns from the data that I need to reconstruct the features for training. Evaluates this model on historical test data. And in this case, I can see we've achieved 90% model accuracy with an 85% precision rate. In addition, it's all explainable here. Gemini is actually telling me which signals drove it, confirming that high interest credit card rates and steady monthly payroll deposits are the are the strongest propensity drivers. I love the quality of this model. I'm going to go ahead and push it here to knowledge catalog. What

55:04this does is this will allow my business users and my agents to use it. So, knowing that, let me switch gears for a minute. Let's say now I'm at marketing manager and I want to use that machine learning model. So I can ask it identify the customers in Illinois most likely to take this offer and create an email campaign including some digital marketing assets. Gemini finds the model in knowledge catalog and invokes it because this is all built on the same tools and memories to draft compets generate a CSV for the target segment and the whole campaign in Google marketing platform. This is great for Google for the target segment and stage

55:31the whole campaign in Google marketing platform. What's a Google marketing platform? This is great. Fast forward a week. my campaign's been running and to track the business impact. I can ask it, show me how our personal loan campaign is performing by the top five counties. And in just seconds, we get a dashboard on live data. Jax also makes it really easy to build on our data story. Let's go ahead and add an additional KPI just so I can really understand the total

55:50impact. Add a KPI for the median time for how fast the campaign is turning into loans. This is going to easily add that metric to my dashboard, which supports interactive parameter controls that I can modify without. So let me show you what that means. First, I can see here, great, they've added me the tile with my new KPI. And when I click into these the different counties behind the scenes, that dashboard is leveraging query templates that are stored in knowledge catalog. These context templates allow me to run the chart queries with zero outline tokens and no compromise on accuracy. Business intelligence dashboards also stay fresh and secure

56:15grounded in your enterprise data. And from here, I can go ahead and share the live results with the rest of my team. What used to take months of manual work just happened in minutes with an enterprisegrade governed transparent workflow. We empowered our marketers to visualize data in real time leveraging it and did so with less token spend and high accuracy all grounded in knowledge

56:30catalog. That's huge. Thank you. That is pretty huge for like medium and large companies like that. That's huge. Thank you. Especially ones that have like crazy large data sets. Well, there's still such a moat though in understanding like really becoming a subject matter expert on how to how to leverage these tools because there's so many things to it that often times you're like you don't realize how deep it goes until you see the people who work on it and build it every day like showcase it and then you're like, "Holy [ __ ] this can do a lot." Yeah, but like really getting deep and understanding how to make all that

56:59happen for for companies, I think is going to be a huge uh huge wedge. I just can't help like as as someone who's been in online businesses where the the structure of the business is just constantly ch like running an ARI agency like it doesn't run like a freaking unless you've niched down, you got a a standardized offer, the business the systems are not like as rock solid

57:21as say like an accounting company. Yeah. Um, and when you've got these systems that are constantly changing, the data you build up is kind of all like amorphous and doesn't really like the data from the last three months isn't kind of as relevant to the system as the previous 3 months or 6 months because you've had major changes in how things work or the content strategy change or something. So, I can't help thinking that all of these changes, all of these deep database benefits and being able to build these models and have it all available to the agents and building on top of it, um, is really

57:51going to benefit, say, guys like us. If we went in and bought an accountancy business or that partnered with one of these more traditional slowmoving businesses that at the end of the day, the business model is [ __ ] it's set. It's concrete. Unless you're going to fully rethink the model from the ground up, which is a bit more of a gamble, if you just go in there and say, "Let's audit this whole business, the whole sales pipeline." Um, and then even after they've closed into the into the delivery and customer satisfaction data and all the key steps and how long it takes to get like if you can get a clean

58:22data set across a set of systems that don't really move that much and you have the ability to go in and and and do the stuff on top, I think is really where the lift is. But if you look at a lot of these newer companies that are constantly moving, constantly shifting, um it's going to be a bit harder. Um yeah, I see what you're saying though about like buying companies that are more traditional and haven't updated and just

58:47Yeah. understanding how to rework from the ground up with AI with things like this and making it crazy, tracking everything, storing it, and then just letting these letting these agents tell you how to run it basically. Um, it's like what's our average time to uh average time to close? Okay, what's the based off all of our customer data that we collect on the onboarding form, what's the let's create a machine learning model that would tell us the likelihood to of them to convert or something. Or let's analyze the sentiment in the discovery call and turn that into a a row in the database. And then we're going to create a machine

59:23learning model based off that in terms of discovery call sentiment to closing call closed or not. And now you've got like interesting bits of data that could help you to write a much you know it's that's really exciting to me. What I like what they said about this, a lot of this is that they're they're removing the the failure modes of AI when analyzing data and they're making a lot of it deterministic because they keep mentioning like no token spend on

59:45certain pieces. And that was one of the issues that we talked about on Josh's makeover was like we want to be very careful to leave the the deterministic things to deterministic code. Yeah. And not have the AI try to like generate where it's going to elucidate and cause issues. And it seems like they've

59:58thought through this as well. Yeah. uh correct correct me if I'm wrong but this is more focused on big companies and not too much SMBs right I'd say still like mediumsiz businesses for sure um they are talking about a lot of bigger ones as their as their examples but um I still think a lot of these tools will will trickle down to us as uh as small business owners yeah I mean the reality if you if you're handy with cloud code and like you use their documentation to like have your agents be able to manage like a Google cloud server you could spin this up as a

1:00:30oneman team for like medium-sized businesses, even small businesses. Mhm. Um, what up, Mr. Pink? Um, what's your take on staying relevant as an AI educator in times of AI influencers taking over? Um, I had an interesting take on this the other day, which is I feel like more and more people aren't watching the video, they'll just like copy the link and give it to their CL or they'll like you'll get scraped up in a in a scraping run in order to like deep research a topic. And it's like if if you're not going to watch the video, do

1:01:05I need to film it? You know what I mean? If it's a talking head video, do I really need to sit there and film it? I think creators can sort of take that back at at some point and say, "Hey, look, I've got a message to share. Bang, I'm going to use my avatar to create it." And if you use something like Jenjitsu and it's like chopping and changing through the angles and it looks

1:01:23it's better produced than it would be. It's faster to produce. It's cheaper to produce. Um, I guess cheaper in terms of my time as a founder and it's better. That's like the that's the holy trinity of AI. Faster, cheaper, better. So, if I can use jinjitsu and not I guess cheaper maybe questionable if it's like 10 15

1:01:40minute video. How much is sea dots too? Uh it's pretty expensive. I think it when we did it three bucks per Yeah, when we did it for Isaac it was like $2 for a sixsecond video or something like that. Uh yeah, I suppose I suppose I suppose. Uh but staying relevant as an AI educator,

1:02:01that's a good question. What I'm starting to see a lot of is like it's a bit of like inside a baseball kind of stuff where certain people now have access to the labs and they get early access to stuff and then it's it like as a smaller creator you're just not going to be able to keep up like they've already had the new thing for a week and so a lot of the the views come around the the new releases. Um and

1:02:26so how do you stand out? I I just see the with all of these new new advancements and developments in using AI and how simple it can be to to set a business up on say this Google platform, how the tools of of tracking the data and and building systems around that are getting so much easier. I heard this I saw an interesting tweet from Alex Becker and it was like I I think like being a content creator is like less valuable than ever. This is like the This is the era of the sweaty like neck beard dude in his basement just like absolutely ripping Claude code and

1:03:03and Codeex and building this completely faceless thing using AI creative generation and all that stuff. And I honestly think even with my own content, I'm looking at like I could just stop trying to do all of this stuff and just build a really good business driven by paid ads, pump it into a funnel, use the existing credibility I have, which I guess is kind of a thing that you don't have. But um if I've got a big social media, I can just keep posting stuff and not necessarily try to drive the business through it, but just have it as kind of like proof of life where I share

1:03:33about me or my opinions or takes. doesn't need to be absolutely driving the entire business when I can build a proper like diversify my company away from being purely content driven which is I can tell you bro you're going to burn out if your whole business is based off the leads coming from your content cuz eventually the algorithm will stop like shining its light on you or something will happen or you got competitors or whatever it is and then you're just in this like depressive state because you can't get the performance you used to and that's just the the the the curve of a creator like

1:04:03a hero's journey and you You better hope that by the time you have that period, then you've you've figured out how to run this business properly. And then you can go set up a a paid funnel or something on the other side so that you can run the business properly and build a business that doesn't really need your content as much, but your content is more of just like the the credibility where if they're in your salesunnel,

1:04:25they'll go, "Oh, let me check it out." And they see Liam's active on on Instagram, LinkedIn, and YouTube. and he's got a big following or even just some following is going to help um that you're not a that you're not a nobody online. And I think that's the that's the balance now. Ideally, the content doesn't need to drive all the all the conversions. It's just like a proof of

1:04:42life and an authority thing.

1:04:49So yeah, would I get into AI and start ripping trying to compete in the in the market as it is right now making AI influency content. I think honestly if you figure out how to crack like AI creatives and and ads that's going to be a there's a there's a window here where if you could do that

1:05:15you're gonna get much quicker results. Yeah creatives is some of the best performing content in the entire niche right now. Mhm. But I mean if you were to try to get into the AI space would I recommend making a a YouTube channel and trying to compete with just how fierce the competition is on the content side? Uh some people if you're completely newbie Ah yeah I feel like if you just figured out how to I mean you're going to need some

1:05:37money to do the ads thing. It's I think it's always better like if you're trying to decide between doing organic or ads like getting good at ads if you have the money to because the nice thing about organic is that it's h I can't even say it's free because of the competition. I mean the scraping that I'm doing I spend a [ __ ] ton of

1:05:54money on that. So like if you're happen to be a really creative person, sure. Otherwise, ads are nice because you can kind of track whether something works right away because you get metrics and you can use metrics. With creatives, you have no metrics. You have to like just throw [ __ ] at a wall sometimes, see if it sticks. If it doesn't, you got to figure

1:06:11you have no idea what went wrong. And I do like the scientific nature of like how in ecom or something, it's all about, okay, what was the cost per click? What's the clickthrough rate? Okay, what's the conversion rate on the landing page? You could really scientifically build your business. You need to know roughly who you're trying to talk to, what they want to hear, what you're going to sell them, of course, and how to deliver on something like that. Um, or you could just be smart about and say, "Hey, look, I'm going to target this niche, and I'm going to run five different landing pages and offers,

1:06:37and I'm going to run traffic to them. I'm going to figure out how to get the clickthrough rate looking good. Then, I'm going to figure out which offer is converting better. Then, I'm going to try to sell that." So, you don't even need to and then you figure out what the thing is you need to sell uh off the back of that which is converting. You don't actually need to have the thing ready. if they're just converting or clicking through and you have like an email capture or something that's showing you intent that hey I want this thing. Uh but don't get me this is weird coming from like content

1:07:02guys saying that like the grass is greener on the other side. So keep that in mind. Um this is something that we're only just getting into now for our business and I will definitely keep you guys in the loop on these on how that goes. But uh yeah, it definitely is pretty fierce competition these days. Um, if you can be technical and build your own content systems to help you stay on top of the news and post quickly and edit quickly and do that uh in a way that's cost effective, that's also a a very very very like people are are blowing up rapidly. And I mean,

1:07:37if you told me if I was starting again now and I'm like, I'm just going to make myself sort of grow on Instagram by doing essentially like kind of AI slop just refeeding the news to people on Instagram who are like a bit further down the thing. That's all it is. Like and getting paid sponsorships. That's

1:07:54that's a pretty valid way. Everybody's scraping each other's posts. They're actually kind of crazy. The one thing I would say about short form that I I probably didn't appreciate is that a lot of my videos would be on on long form on YouTube would be like building the whole thing out and showing them how to do it and having this end result. On short form, you just make the result and say, "Hey, here's the thing. If you want

1:08:13it, click comment down below and start." It's way harder to do YouTube for sure. Like you got to have the editing, the thumb, like YouTube has thumbnails, titles, and editing. Well, you also have to like learn how to capture attention for a much longer period of time. And it's it like you have to be more articulate more articulate on YouTube as

1:08:32well. Mhm. Bro, you fall off and your school is full of bots. Yeah. Shut up, bro. Damn. Okay. 90% of enterprise data is unstructured like images, PDFs, reporting, scans. For most companies, that information is dark. You store it, you pay for it. No system ever reads it. Smart storage changes that. It uses Gemini to enrich those objects in place, writing context back under the object itself. So the intelligence stays where the bites are and inherits the security postures you already have. Snap connected its prison

1:09:05agent straight to it storage archives. Cut diagnostic troubleshooting from 30 minutes to 30 seconds. Snap. Snapchat prison prison data. Pause. Some soap dropping soap dropping action. give everybody in in in prison Snapchat account. Using our data enrichment agent, we're able to get Snapchat on 90% of prison phones. Is dark. You store it, you pay for it. No system ever reads it. Smart storage changes that it uses Gemini to enrich those objects in place, writing context back onto the object itself. So the intelligence stays where the bites are and inherits the security postures you already have. Snap connected its prison agent straight to its storage archives and cut diagnostic troubleshooting from

1:09:4830 minutes to 30 seconds. And once Gemini can read an image, it can check and validate work that used to depend on someone remembering. Itachi for example with the help of Gemini Enterprises transforming global power and utilities by putting agents in the hands of their frontline workers using HMAX their portfolio of AI powered solutions. Technicians at the custom AI agents for over 200 use cases. For example, instantly verifying photos taken of every tool and wire before and after a job. This helps to keep the team safer while boosting productivity by up to 30%. Your data moves across clouds in

1:10:17SAS applications and even on premises. We're not asking you to move your data. Gemini uses our borderless lakehouse to query Amazon S3 and Azure data lake. No variable egress fees. It reads Salesforce SAP service now without copying anything. And it fed e borderless lakehouse. Salesforce SAP service now. I'm trying to get a lakehouse. Yeah, without copying anything. and it federates open iceberg across data bricks unity snowflake polaris and AWS glove. Deutsche.com has made data collection collation and analysis three to four times faster. Etsy unified three pabytes of e-commerce data on open iceberg to run it pipeline queries 60%

1:10:53faster. Wow. Bloomberg media grounded its data agents in the knowledge catalog and lifted query access accuracy by 63%. Here's how a leading industrial automation company is protecting mission critical systems by predicting maintenance problems before they happen using Jenna.

1:11:13We are automation company. So what mission critical segment means is that our control system cannot fail. It has to work 24/7 365 days a year. That's a fundamental principle. For example, a hospital system, they want only one thing. It should never stop. My name is Kapoor and I'm chairman and coal. So what we do today is to serve our customer to automate their mission critical operations. Honeywell technologies has partnered with Google cloud to build force intelligence layer on top of our control system. China enterprise is a key enabler for honey's

1:11:39predictive maintenance capability. Predictive maintenance is the latest trying to predict what can fail with predictive maintenance. We take all that data in the cloud analyze it and then when it predicts that something is going to go wrong, it produces a service case for the operators. The first challenge we have is to get easier access to data, frictionless access to the data. This is our AR sensor, predicted maintenance, reliability sensor. It can measure many data points such as temperature, magnetic floods, vibrations, sound. These are pretty much the brain of the control system in the middle and they transfer the data from the actual units to our pure systems

1:12:04here. So we used an enterprise to contextualize the data. You know what specific part before it goes down. So you're taking an unplanned outage and turn it into a PL outage. The second problem is to make this data relevant to the customer and an outcome and that outcome can be automated with a system. Vender build has more users. Think of university campus 100. So this is a massive scale where this capability is being deployed because it's a learning system. It is telling them some failures they may not have observed before. And because they're able to deal with potential issues earlier, the equipment target with predictive maintenance that's

1:12:32enabling our customer to become more and more autonomous. We made a decision to replatform our forge platform uh on Google cloud. The decision is to give our customers choices on the architecture they want to choose. Gen enterprise is a critical tool in our partnership with Google cloud is an important part of our solution. The mission criticality really becomes a hard to enable our to be successful,

1:12:52bro. Indians be running the show. Like I think if you do the stats, they're just like crushing it in terms of like high positions. I mean, when I worked at Amazon, pretty much my entire team on both the teams I worked at were Indian. I mean, I guess when you have enough Indians joining as like low-level engineers, like you're going to have people work their way up if they're actually a killer, you My entire team was either

1:13:14Argentinian or Indian. Yeah, like Sunda, this guy and the Honeywell guy. I mean, I don't I don't know what exactly they are, but they they look like look and sound vaguely Indian. Yeah. Crushing everything. And this dude's here. What a G. We're specializing the Gemini agent for individual industries with the skills, tools, and knowledge each one needs.

1:13:40Gemini. Oh, it's hellbirth. my guy. What up, man? Hope you're doing good. This is like a I think he's Brazilian. No, he's Portuguese, but he moved to Brazil or he's selling to Brazilian clients or something. Um, and he he used to show up to all my free community Q&A in the in the community.

1:13:57That's sick. Um, and he started crushing cuz he kept showing up. And is now in preview for financial services and legal and coming soon to government, healthcare, and retail, each with its own connector, skills, and agents. Whoa. See, this is if you talk about like being a general agency, you just roll in there with one of these, it's going to take a bit longer. But I think if you have this legal base and then you have your own customs things on top that you bring in financial services, Gemini ships with I can't help thinking, bro, it's got to

1:14:25go towards being partners with them. That's what I said again in in the in my my predictions for this year is it's it's all about being their their AI partner. like it's going to take some time for this stuff to soak in. Getting them up on this, monitoring it, being like, "Hey, looks like we've got an opportunity for a new skill here or new new custom app or this has been done kind of inefficiently or you've got these people who are not really using the system at all. Probably need to let them go my guy." Stuff like that. Yeah, that's more of like consultancy

1:14:56basically. And trusted financial data from Fax, Elsag, and others. Let me show you Gemini for financial services. Please welcome. One day we're going to be sitting here and it's [snorts] going to go like one of the plugins is like AI consultancy and development. We're like they got us eventually, man. It's like damn Claude Claude just be killing all these industries, man. Yeah. Damn. And then one day it's going to

1:15:22Yeah. going to come fast. Your own AI auditor and consultant and development team managing your entire AI stack and strategy. [laughter] Let's say I manage a $100 million global multi-asset fund. An email delivers a YouTube video breaking the news of an unexpected 48. Now Indian crushing it billion dollar all stock buyout of a primary energy holding. Now that news immediately impacts my fund because we hold a substantial position in the target company. Our client contract has one strict rule. No more than 25% of our portfolio can be in energy and I only have until the market closes in a few hours to evaluate our risk, verify

1:16:00compliance and credit our capital. That is a lot to do here. I turn to my Gemini agent which has been given specific skills, tools and knowledge for financial services all while so plugins again are like a bundle of connector skills and agents. Yeah, Gemini to assess how this announced energy. Everything is going to plugins, bro. Plugins are the new like

1:16:19Apple app store basically. Mhm. Like when Apple first came out with the iPhone and everybody was like, "Yo, this is like such a crazy new thing." Like apps weren't a thing. It wasn't until like the App Store came out and everybody's like, "Holy [ __ ] I can build an app." Yeah. And like there was like an app revolution for a while. And I feel like right now we're going to get into this like plug-in revolution because you've got OpenAI that just released their plug-in marketplace. You've got these

1:16:42ones that are talking about plugins. Obviously, Anthropic has had plugins for a while, but like this is going to continue to be like a a theme, but it so is a plugin they have is a mix of agents, connectors, and skills. Um, so us as an agency create our own plugin and we use it to deliver like we have our I don't know accounting firm plugin and we just roll set that up and it's got a lot of our our IP and our stuff already baked into it. But like I don't know about like as soon as you put it on the marketplace how like all of these like monetizations

1:17:16of of plugins and stuff has never really worked cuz as soon as it's out there basically handing them the code and they said it work. Yeah. Uh yeah I mean that's true. I don't know how it's going to work engineer it once it's once you've got it. Yeah. But who's going to be smart enough to do that though if you giving it to these businesses? I think it'd be interesting if we did a live stream one of these days on like just building plugins just to see what it looks like on all these different platforms like building all that on the back of a makeover if we

1:17:40delivered on one of these platforms where you could say hey look we've created this plugin for e-commerce businesses bam you know what I mean cuz it's it's more about like niche specific I think he's a Brazilian living in Portugal my guy good to see bro the breaking video I have an offer that I've generated. Can you validate it for me? You generated

1:18:03the offer. That already sounds sus. But yeah, send it through. Sounds sus. You're going to need to make it short cuz I don't think you can make like massive live positions to internally and the market data feeds from key data providers to carry out a deep analysis. Now, Gemini generates this result for me. So, I pre-wemini flags that our energy exposure is in breach. I click the confidence analysis to see that it has high confidence

1:18:28across multiple data sources. I also verified that it extracted the bio price of $80 per share correctly. Gemini says that they limit by $1.65 million. We can't just sell blindly. So I asked Gemini my next question which is what is the likelihood that regulators challenge this deal and what do I need to do if this portfoli maintains both short duration and cross session long duration persistent memories. I didn't need to provide context. Gemini picks up right where I left off. Now normally modeling market share across an entire operating basin would take hours but Gemini can do this for me in minutes. Again, I pre-war

1:19:00this, so let me show you the results. Geminine highlights that the threshold is indeed breached and the combined entity would involve an FCC second request or extended review. Under the hood, it is able to carry out a diverse analysis across multiple scenarios, highlighting the dollar P&L impact of each of those scenarios. From this analysis, Gemini calculates that if the deal goes through, then the fund can capture an additional side of $2.7 million. But if regulators challenge this, all these gains evaporate. I also receive a strategic recommendation to predict against the deal collapse downside. Now all I need is a concrete execution plan. So I give a final prompt

1:19:30to recommend trades to bring my exposure back with Gemini and how to set up an insurance policy to predict against downside prediction throughout. Finally, it hands me two deliverables. An executable PDF memo and an Excel sheet that entails the summary and the trades that I need to execute the story and rerun this portfolio difference automatically whenever an event occurs like a new SEC filing or a breaking fact. From breaking news to a mathematically hedge portfolio of minutes, that is Gemini for financial

1:19:55services. Back to you. Initutions like CME Group and Deutsche Bank are already using Gemini for financial services. In legal, Gemini works alongside the agents your firm already uses. Everlaw, Harvey, and I manage matter level permissions and ethical walls hold across every single agent handoff. Harvey automates complex legal work while meeting the firm's confident standards and on it writes contract life cycle management without exposing privilege. We're also working with a premier global law firm, Koulie, an early adopter of Gemini Enterprise for legal. Kulie is building a confidential information reduction agent on Gemini Enterprise for legal that helps them to automate reductions including personally identifial automate reductions confidential

1:20:33informational law firm Kulie an early adopter of Gemini enterprise for legal. Kulie is building a confidential information reduction agent on Gemini Enterprise for legal that helps them to automate redactions including personally identifiable information and confidential material before legal filings go public. Interesting. In high stakes litigation, this can cut days of document review while maintaining the firm's rigorous standards for client confidence. In out there, the hardest part is finding

1:20:56something that nobody knows yet. I mean, the the interesting thing there is that like the Honeywell thing, they have all these dashboards and thing like they're building that on top of all of this. Yeah. Um, which I think is interesting thing of them migrating their custom internal operations to this stack and having agents be able to interact

1:21:17with it. Yeah. Is working to accelerate molecular research without compromising data privacy

1:21:29is probably our biggest enemy. We have revolution in technologies abilities. We know more than ever about human biology. But what do we manufacture as time? Mark has an incredibly special mission to use in science to be able to develop novel therapeutics and vaccines for the treatment and prevention of human disease. So every decision reverberates not just for day or week some cases for the collaboration between marketable bio but bring together high resolution data sets purpose algorithms and extraordinary power to be able to solve problems that even 5 years ago were solved. How do we actually not yet research medicine application of our drug? Developing the right drug for the

1:22:02right patient is really hard. Then we have to design a drug that will prove to be public efficacious and safe. Gemini Enterprise agents really have been helping us in every stage to let the drug describe from interrogating biology to better designing [clears throat] our molecules. We can now leverage all of these foundational mods use Gemini Enterprise as the orchestration layer to give these models our own data. Now ask the questions that are specific to the drugs and compounds that we're building. We're already seeing impact on those molecular design cycles being induced in lab at [clears throat] Mercy 75,000 employees have access to holy from discovery all the way to

1:22:35manufacturing commercial clinical I use it every single day the ability to use Gemini enterprise to distill key concepts key data sets in just minutes in some cases in seconds and do it probably better in many cases than I ever could is quite remarkable everybody wants to break through faster than the timeline right now they're 10 to 12 years to invent develop a new therapeutic or vaccine. Imagine a world where we could do that in 2020. That's

1:23:00just tremendous impact. We're at a point now where precision medicine, it's no longer our dream. It's a reality. Rise right now to all of our scientists. This actually gives them more time to

1:23:18these examples we're sharing today address real business problems, but sometimes it's hard to know where to start. Deoid's agentic AI blueprint helps companies achieve measurable impact up to three times faster. The solution first identifies the most appropriate operations to automate with AI using heat maps based on Deoid's industry benchmark and then design aente. Deote helped achieve a 30% cash flow improvement for a telecom client by redesigning their corporate billing

1:23:44process with this. So now they starting to have proper case studies coming out of even Deote. So [ __ ] the rush to that like true ROI for your client still on you know skills tools and context with those three an agent stops assisting a person and in addition starts running the process. Gemini can now work across your entire business. To make this agenda program successful however you need to govern it and determine whether you can afford it. Let's start with governance. Four questions typically matter. Who the agent is, what it's allowed to do, what it did, and what it must never touch. The first three are

1:24:20answered by identity, policy, and observability. Every Gemini agent gets its own identity, cryptographically attested and governed like an employee with least privileged permissions. Okay, watch this. Next thing, it's going to be an external agent that other business agents can interact. Like I've been saying agent across businesses. Yeah. Like you go on your Google

1:24:42profile. Yeah. And instead of having to click on their website or click on the like reserve, communicate with their agent is their s like front, it's like the receptionist, but instead of having to call over the freaking phone lines, it's just like here's the agent ID, external agent ID for I mean you could have a department like the finance department how your agent can can go in and talk to them directly. There's a bunch of security stuff that needs to go in there. But each business on Google My Business, like the profile they have on maps is going to have like a thing where you can

1:25:09scan it or an agent can scrape it and that's like this is the entry point to our business. If you have a question about a reservation or anything, your personal agents like Muse and Dots and all of these other ones are going to be able to chat to that. And so you'll have something that you set up in on on the Google platform here. It'll be your external business agent. That's kind of like your receptionist that uh anyone

1:25:29can call or interact with. Yeah. Users can give each agent fine grain role-based access control permissions and get them approved by organization security administrator. Every action that Gemini takes is written to an audit trail attributed to the agent rather than the person. That identity follows the agent to any virtual machine it spins up to run code allowing security teams to watch it and identify anomalous behavior before they become problems. The fourth question is answered at the perimeter. All Gemini agents execute tasks safely inside an agent sandbox with its own network boundary and all traffic in and out between agents passes through agent gateway, an AI network firewall that

1:26:07enforces your policies and checks for malicious prompts and two calls in real time. You write the policy just once. For instance, agents may not open documents classified need to know. The gateway enforces the policy on every agent in your company rather than being checked one agent at a time. This matters most where the administrative work is regulated. Here's one of Britain's largest banks putting Gemini to work for customers and colleagues in a very

1:26:31highly regulated industry.

1:26:41AI is substantially going to change the way we do business with our customers. My name is Roada. I'm the group chief operating officer for Lloyd Banking Group. Our big objective is to help companies, individuals and communities prosper all across the United Kingdom. We are applying AI in 12 big bets that range from our proposition in investments to fraud management or

1:27:01onboarding clients for the commercial. Anvoy is a pipeline to bring agent to base. Am I traveling? Fraud management for on boarding clients for Is he just reading off teleprompter or is he got back got an eye on the competition? And for range from our proposition in investments to fraud management or onboarding clients for the commercial bank. Envoy is a pipeline to bring agentic use cases in a secure and safe way fully managed into production. It's based on Gemini enterprise. So a full collaboration with Google definitely enterprise helps us offer a solid foundation to all of our engineers building use cases and so far we have onboarded close to a thousand engineers

1:27:37working on agentic use cases making use of so you said you had a use case that was looking for a orchestrator agent you can send them to the marketplace where we have a number of different agents that have already been built that can then be used by other teams that push. We invested together with Google Cloud into this tooling that helps us create a foundation, create a harness, provide guard rails to make sure that we focus on adding value in the individual use

1:27:58cases. One of the core components of what is the golden path that gives our users decal in the individual use cases. One of the core components of what is the golden path. Damn. Okay. So, Google's partnered with these guys given them dev support to build out all these pilots. Oh, yeah. This is clearly a pilot platform. a pilot that they've gone deep on. Um, but I mean like I guess when you're running stuff at scale these kind of this stuff matters where you want to like be like how could we reduce cost here or something like I use this scaffolding design build

1:28:33application and such a CCD pipeline. Data privacy and data security is at the heart of what banks do with Envoy. We have a road map to better make use of the security features of Google Cloud. Both the bank as well as our customers benefit from a use case for our agents to better detect and manage potential fraud. It saves a lot of time for them to help customers go through such a difficult situation. All of this is extremely relevant in a highly regulated industry. And that's where an agentic AI pipeline including all the guard rails and security play such an important role. What the future will look like is

1:29:02hard to predict, but from what I can tell today, everything is moving each day. You know, per token prices have fallen 98% since 2024, and enterprise AI spend is climbing anyway. The reason is volume. Single complex agentic task can consume 100,000 tokens. Every execution loop through a premium model and the budget breaks. Gemini helps you in three ways. First, you get to choose the right model for each task. Some tasks need the deepest reasonings. Others need a simple response. The best model for a task is always not the largest one. With Gemini, the agent chooses the right model to do a task. You get better results with less

1:29:39expense. Additionally, the leading model changes every few months. If you tie your context, your skills, your tools, and your policies to a single platform, you're forced to replatform every time the frontier moves. Yeah. Builds each job on a model that fits from Argon at the frontier to flash for speed to the latest models from Anthropic. Other leading proprietary and open models are coming very soon. One contract, one place to use the frontier. Second, to further optimize costs, earlier this month, we introduced smart routing and cost controls for you to manage your model usage costs. Gemini matches the model to the work. It starts on a more

1:30:12inexpensive model and escalates to a larger frontier model only when the job demands it. In our testing, that comes close to running everything on a Frontier model at a third of the cost and 40% faster. Third, you can set real-time spend caps per project. We give you tools to watch your agents and halt them before they overrun your budget. The cap covers the whole cost of an agent. The cost of the tokens, the cost of any code sandboxes, and the cost of the compute machines. Tracking spend the granularity of projects allows you to charge back cost to different

1:30:41departments or programs in your company. All of this, the agent, the data, the identity, the costs rest on two layers most people never see. These layers are also the reason the economics work. Start with silicon. We designed the chip, the interconnect, the data center, and the software as one system. Because we control all four layers, we can optimize workload routing directly from the hardware up. This full stack integration allows routine agents to run

1:31:05with minimal overhead. I've been saying that for so long that one of the biggest modes is the full on top of the silicon stack generally argon for frontier research, flash for speed, omnia, and nano banana for generative media and gem are open models for the workloads that will run somewhere else entirely. And when I say anywhere, I mean anywhere. NASA's Jet Propulsion Laboratory is running Gemma directly on a satellite in orbit, a first for a vision language model. JPL's software analyzes Earth imagery on board the satellite and sends summary of key information to ground operators. The imagery is paired with natural language text, allowing these operators to query

1:31:41the spacecraft directly about what it's seeing. The hardest part of all of this is really the technology. is changing how a company works and that often takes longer than any of you would like. As you build on this platform, we make three commitments. First, we take the friction out by giving you a future proof platform. Runs anywhere, does any task with your choice of models and robust controls. Second, we eliminate complexity and Google builds and integrates this whole stack, the agent, the tools, the skills, the models, the silicon, the controls. Third, what you build stays yours. We do not ask your data to move. We do not ask you to hand

1:32:19over what makes you special and how what you choose to make with our agent. We deliver this together with our global partner ecosystem. Inside Gemini, you have an extensive range of third party agents and connectors ready to use without building them yourself. You can find them now gathered as collections in the new AI solution funer and the firms

1:32:34that help you. You can find them now. Range of So these are deaf agencies. Oh yeah. And check buttons. Damn. Get on that list, man. Party agents and connectors ready to use more connectors. You can find them now gathered as collections in the new AI solution finder. And the firms that help you deploy it are investing too. I stood up a dedicated Gemini business group and several of our largest

1:33:02partners have made similar investments. Everything I showed you today already exists in someone's hands. Lloyds redefining commercial banking. A lawyer at Kulie protecting a client before a filing goes public. A satellite deciding for itself what is worth sending home. The techn is ready. We're looking forward to helping each of you put it to

1:33:20work. Thank you very much. I um I mean this is what I was I was saying. If we could get the the goal like with the makeover format would be if you could get Google and they're like yeah we just want you to go around and install install businesses on our platform consultant for that pretty much create a create a firm around it and then go out and do these deployments for them and show sort of like the the the

1:33:50results and the human side of it. Yeah. I mean, the more I see these announcements from like OpenAI, Google, like the more confident I am that building these custom workspace platforms, trying to compete with them, it's just it's never going to work. Like, yeah. Um, okay. There's been some questions spammed in here. I'll get to and then I'll bounce. Uh, 18 years old from India. I want to start an AI agency. What service should

1:34:16I offer first and how can I get started? Tell me the blueprint, please. According to 2026, what service is best for AI agency? I've said it many times before and I'll say it again. You need to learn how to build. So, get yourself a codeex or a Claude plan. Build some stuff for yourself. Watch a course online on how to build with Claude Code. Buy one on of Udemy or go on YouTube. Figure out how to build stuff. Get fluent. Build stuff for yourself. any little tools that can help you to reach out to people or to identify content opportunities or to automate cold email and enrichment or

1:34:52anything to help you build your agency. It's going to be cool little tools. Then try to go and help other people to build things for them. Say, okay, my nana needs a recipe generator tool. Okay, cool. That's your portfolio. You've got your own things that you made. You got the stuff you made in the course and you've got stuff you made for other people. Ideally, you could go to a business owner in your network or a friend or family or whatever and say, "Hey, can I I built you this website or I built you this little calculator tool or hey, I built you a content ideation

1:35:24platform or something. Anything. Build a portfolio." And then if you're from India, I'd recommend you go on something like Upwork. um unless you have access locally. Upwork you can put yourself uh put up your portfolio and then put your cost at basically nothing and try to land some deals that way. Um you'll need to optimize your profile and be grinded on Upwork and then you'll start to get a few deals here and there for a very low cost or free and then your reputation will start to build on there if you're

1:35:52actually good. Um yeah, check this out. I just download the Gemini app. They've got this thing called it's speak to Windows what they call it. included whisper flow. Damn. We can use this in any uh any chat window. Damn, right anywhere. Oh, so that comes with Gemini app. With Gemini. Yeah. Hold Fn key for literally any open window until it

1:36:12right. Dang, it's crazy. Uh last one. Uh the offer has helped busy business owners turn social media into a customer acquisition engine by creating and managing the the content building system to attract and qualify leads and booking appointments without them requiring to requiring them to record videos or appear on camera on camera. Um yeah I mean sure cool you can you can offer that. I think if you haven't done it for yourself I wouldn't trust you to do it for me. So the best proof would be that you built a content system that markets to these people like hey look this video that I'm watching right you're watching right now

1:36:54I didn't film any of it and this is the future of content for business owners if you want to see how I do it or if you comment such and such and you start to get followers and comments and people in your many chat list and then you can go making offers directly to them or you can just keep ripping that same thing it's like I made this video without filming you should be making content as a business owner if you're not DM me or comment this this and I'll send it to you and just keep spamming that on social media. You should be able to grow

1:37:20your account and get leads without you filming yourself and that will prove that you actually know what you're doing. And if you're like Indian or I don't know, Middle Eastern or something or something and you think that's going to prevent you from doing well in pitching these these clients, then choose a different character. you get to go on to Higsford or whatever and turn yourself into LeBron James and market yourself as LeBron James selling your AI content services. Um, you probably won't get away with that one, but you get what I mean. Um, you can just create a new, this is what Alex Beck was saying about

1:37:53like the basement, dude. You could just be sitting in your basement and like not show your face ever and just, you know, hide away from the world and just AI generate everything. And to be honest, while of course I'm very grateful for the platform I have, um, [ __ ] it, took a lot of work. And if you're talking about getting speed to money and being able to just like if you just want money, but if you want, I don't know, status or connections or something, obviously personal branding is great. But if you just want to make quit your job and make 10K, 20K a month, then yeah, basement

1:38:23straight, no content would probably work great. Um, never get in front of a camera ever. All right, guys. Got a bounce. That was I thought we'd get on here on Vibe Code. Then we just watch this whole thing. But I'm glad we did. Very interesting. If you haven't watched my video on dots already. Um it's the same OpenAI platform that's trying to compete with this. Absely just got mogged by this in my opinion. Google is the one to beat, the one to watch out for. As I said in my previous video on

1:38:49YouTube. Um Jesse, any last last minute? Nope. I'm uh I'm actually pretty bullish on this, especially for like large data sets. I mean, I've got like in in terms of like how I track content, what we're doing here, like I've got like 80,000 posts across platforms that I' I've been scraping and tracking. So, like I'm really curious to like throw that data into Gemini and see what kind of like Vibe ML stuff I can do to extract um insights. Like I I genuinely think this is going to be a great platform. I think like getting familiar with Google Cloud um is going to be in anyone's best

1:39:22interest if you like really want to stay ahead of this. Like I'm I'm pretty bullish on this. I'm I'm curious to see what I can what I can do. Okay. All right, guys. Have a uh have a good weekend. Catch you next week. Cheers, guys.