WEBVTT 00:00:00.000 --> 00:00:01.659 You know, I was sitting at my desk this morning 00:00:01.659 --> 00:00:05.099 just staring at the Google product page for all 00:00:05.099 --> 00:00:07.960 their new AI stuff. And I have to say, I felt 00:00:07.960 --> 00:00:12.339 this very specific kind of exhaustion. I'm looking 00:00:12.339 --> 00:00:14.240 at this whole landscape, trying to see the strategy, 00:00:14.359 --> 00:00:16.059 and it doesn't really look like a product lineup. 00:00:16.960 --> 00:00:19.699 It looks like a labyrinth. Oh, it is absolute 00:00:19.699 --> 00:00:21.480 chaos. I mean, if you're just trying to get up 00:00:21.480 --> 00:00:24.940 to speed now, good luck. You're staring at Gemini, 00:00:25.120 --> 00:00:30.129 Gemma, Nano, Jules. Opal, Vertex, AI Studio, 00:00:30.670 --> 00:00:32.630 Notebook LM. I mean, I could probably keep going. 00:00:32.649 --> 00:00:34.409 I think I'd run out of breath before I actually 00:00:34.409 --> 00:00:36.250 finish the list. And the source material we're 00:00:36.250 --> 00:00:38.590 diving into today, it points out something really 00:00:38.590 --> 00:00:42.570 crucial right at the top. Google has over 30, 00:00:42.590 --> 00:00:46.710 30 different generative AI tools. They overlap. 00:00:46.890 --> 00:00:49.350 The names are constantly changing. It confuses 00:00:49.350 --> 00:00:52.159 the experts, not just. you know, the beginners. 00:00:53.039 --> 00:00:54.579 So if you're listening to this and you're feeling 00:00:54.579 --> 00:00:56.840 overwhelmed or like you missed the memo somewhere. 00:00:57.100 --> 00:00:59.100 It is not a skill issue. That's maybe the most 00:00:59.100 --> 00:01:00.780 important thing to hear. It's a clarity issue. 00:01:00.840 --> 00:01:03.020 It's on Google's end. So let's try to unpack 00:01:03.020 --> 00:01:05.439 this. Our mission today is not to just do a feature 00:01:05.439 --> 00:01:07.920 dump. We're not going to read you a manual. We 00:01:07.920 --> 00:01:10.340 want to build a kind of mental model for this 00:01:10.340 --> 00:01:13.000 whole ecosystem. We need to figure out which 00:01:13.000 --> 00:01:15.939 tools actually matter for the work we do and 00:01:15.939 --> 00:01:19.620 which ones are frankly Just noise. Exactly. We 00:01:19.620 --> 00:01:22.260 need to stop looking at 30 tools and start seeing 00:01:22.260 --> 00:01:25.040 like five buckets. The best way to organize this 00:01:25.040 --> 00:01:27.439 whole mess is to categorize it. So you've got 00:01:27.439 --> 00:01:29.959 the core, then the Swiss Army knives for productivity, 00:01:30.340 --> 00:01:33.540 then the developing tools, which has this really 00:01:33.540 --> 00:01:35.640 interesting lab versus factory thing going on. 00:01:36.060 --> 00:01:38.700 Then there's the future stuff in labs. And finally, 00:01:38.840 --> 00:01:40.680 this invisible layer. I like that structure. 00:01:40.959 --> 00:01:43.439 OK, so let's start at the center, the core. This 00:01:43.439 --> 00:01:47.090 is the Gemini ecosystem. Now, I think most people, 00:01:47.090 --> 00:01:49.129 they open Gemini, they type in a prompt and they 00:01:49.129 --> 00:01:50.849 think, okay, this is Google's version of chat 00:01:50.849 --> 00:01:54.230 GPT. But the source suggests that's kind of the 00:01:54.230 --> 00:01:56.709 wrong way to frame it. It is. It totally is. 00:01:56.769 --> 00:01:59.250 If you treat Gemini like just another chat bot, 00:01:59.790 --> 00:02:03.189 you're missing its actual superpower. The source 00:02:03.189 --> 00:02:05.650 material really highlights that Gemini, especially 00:02:05.650 --> 00:02:07.969 the free version, shines as a research engine. 00:02:08.189 --> 00:02:10.900 It's all about structure. You can be a concrete 00:02:10.900 --> 00:02:13.199 example of that. What does that look like in 00:02:13.199 --> 00:02:15.560 practice? How is it different from just asking 00:02:15.560 --> 00:02:18.639 for a list of things? Sure. So let's say you 00:02:18.639 --> 00:02:21.659 want to understand the market for AI tools for 00:02:21.659 --> 00:02:24.020 freelancers. If you ask a standard chat bot, 00:02:24.080 --> 00:02:26.180 you'll get a bulleted list. Here's tool A. Here's 00:02:26.180 --> 00:02:27.939 tool B. Right. I've seen that a million times. 00:02:28.240 --> 00:02:30.620 But if you ask Gemini, it's much, much better 00:02:30.620 --> 00:02:33.139 at mapping the categorization of that market. 00:02:33.580 --> 00:02:36.280 It'll break it down by workflow admin, creative, 00:02:36.620 --> 00:02:39.139 legal, and it'll explain why those categories 00:02:39.139 --> 00:02:41.280 matter. It builds a framework for you instead 00:02:41.280 --> 00:02:44.020 of just handing you facts. So it's helping you 00:02:44.020 --> 00:02:46.620 think, not just giving you an answer. It's almost 00:02:46.620 --> 00:02:49.599 like a junior analyst who organizes the data 00:02:49.599 --> 00:02:51.819 before they hand it over. That is a perfect way 00:02:51.819 --> 00:02:53.900 to put it. Yeah, it's distinct from just pure 00:02:53.900 --> 00:02:56.460 generation. OK, so that's the free version. But 00:02:56.460 --> 00:02:59.099 the source makes a really big deal about Gemini 00:02:59.099 --> 00:03:02.659 Advanced. And specifically, this idea of the 00:03:02.659 --> 00:03:05.400 context window. This is the game changer. We 00:03:05.400 --> 00:03:10.020 are talking about a 1 million token context window. 00:03:10.139 --> 00:03:12.699 OK, let's pause on that. 1 million tokens. For 00:03:12.699 --> 00:03:14.180 anyone listening who doesn't speak engineer, 00:03:14.259 --> 00:03:16.599 what does that actually mean in, like, human 00:03:16.599 --> 00:03:19.539 terms? Think of it as memory. like short -term 00:03:19.539 --> 00:03:22.300 working memory. Most models, they forget what 00:03:22.300 --> 00:03:24.120 you said 10 minutes ago, or they can only read 00:03:24.120 --> 00:03:26.539 a short article before they lose the thread. 00:03:27.159 --> 00:03:29.819 A token is basically a piece of a word. One million 00:03:29.819 --> 00:03:33.300 tokens is roughly 700 ,000 words. That is massive. 00:03:33.360 --> 00:03:35.120 That's what, multiple novels? It is massive. 00:03:35.159 --> 00:03:38.300 You can upload 30 competitor blog posts, or entire 00:03:38.300 --> 00:03:42.300 books, or a huge stack of legal documents, and 00:03:42.300 --> 00:03:44.199 you can ask questions against all of that data 00:03:44.199 --> 00:03:46.460 at the same time. That's fascinating. It's like 00:03:46.460 --> 00:03:49.139 having an analyst who who has perfectly memorized 00:03:49.139 --> 00:03:51.699 a specific library of books you just handed them. 00:03:51.960 --> 00:03:55.379 And it completely changes your workflow. This 00:03:55.379 --> 00:03:57.979 is what I mean about moving away from chat. You're 00:03:57.979 --> 00:04:01.020 not asking for a haiku or an email draft anymore. 00:04:01.159 --> 00:04:03.699 You're doing gap analysis. You're saying, here 00:04:03.699 --> 00:04:06.319 are 20 articles my competitors wrote. Tell me 00:04:06.319 --> 00:04:08.219 the one argument that none of them are making. 00:04:08.400 --> 00:04:10.800 That's a really powerful distinction. You're 00:04:10.800 --> 00:04:13.520 not using the AI to write. you're using it to 00:04:13.520 --> 00:04:16.259 synthesize. Synthesize is the word, exactly. 00:04:16.459 --> 00:04:18.120 There's another part of this core stack mentioned, 00:04:18.259 --> 00:04:21.199 though, and that's search. The AI overviews, 00:04:21.199 --> 00:04:23.439 we've all seen this, where the answer just appears 00:04:23.439 --> 00:04:26.079 in the top of Google now, the source calls them 00:04:26.079 --> 00:04:29.040 zero -click searches. Right. And if you're a 00:04:29.040 --> 00:04:31.180 content creator or you run a business that depends 00:04:31.180 --> 00:04:34.339 on SEO, this is, well, it's both terrifying and 00:04:34.339 --> 00:04:36.120 exciting. You aren't just writing for keywords 00:04:36.120 --> 00:04:38.279 anymore. You're writing so that the AI understands 00:04:38.279 --> 00:04:41.120 you well enough to summarize you. It makes me 00:04:41.120 --> 00:04:44.699 wonder then, if the AI has this perfect memory, 00:04:44.740 --> 00:04:47.519 this one million token context, and it can summarize 00:04:47.519 --> 00:04:50.939 the whole internet, does having that change how 00:04:50.939 --> 00:04:53.019 we actually think about doing our work? Totally. 00:04:53.079 --> 00:04:55.279 You stop summarizing and you start synthesizing 00:04:55.279 --> 00:04:57.819 massive data sets instantly. Right. The value 00:04:57.819 --> 00:05:00.720 shifts from, I found it, to, I connected the 00:05:00.720 --> 00:05:03.759 dots. That summary versus synthesis idea is key. 00:05:04.800 --> 00:05:06.839 Okay, let's move to that second layer you mentioned, 00:05:06.980 --> 00:05:09.420 the Swiss Army knives, the productivity layer. 00:05:09.550 --> 00:05:11.589 This is where it gets really practical, and we 00:05:11.589 --> 00:05:13.910 have to talk about Notebook LM. I've heard this 00:05:13.910 --> 00:05:17.290 one mentioned so many times, but the source says 00:05:17.290 --> 00:05:20.230 it's the most misunderstood tool in the entire 00:05:20.230 --> 00:05:23.709 lineup. Why is that? Because people look at it 00:05:23.709 --> 00:05:26.350 and they think, oh, it makes podcasts or it summarizes 00:05:26.350 --> 00:05:29.110 docs, and it does do those things. But the killer 00:05:29.110 --> 00:05:31.819 feature... The thing that makes it unique is 00:05:31.819 --> 00:05:34.439 grounding. Grounding. Define that for us. Grounding 00:05:34.439 --> 00:05:37.199 means Notebook LM does not look at the open internet 00:05:37.199 --> 00:05:39.420 to answer you. It only looks at the documents 00:05:39.420 --> 00:05:42.079 you gave it. It basically puts on blinders. So 00:05:42.079 --> 00:05:45.300 it refuses to hallucinate. Exactly. And hallucination 00:05:45.300 --> 00:05:47.699 is just the polite term for when an AI makes 00:05:47.699 --> 00:05:49.300 things up because it doesn't know the answer 00:05:49.300 --> 00:05:51.660 but it still wants to be helpful. Notebook LM 00:05:51.660 --> 00:05:53.990 won't do that. If the answer isn't in your source 00:05:53.990 --> 00:05:56.589 material, it just won't invent one. That's actually 00:05:56.589 --> 00:06:00.009 pretty rare, because usually these models are 00:06:00.009 --> 00:06:02.250 trying to be these helpful creative partners. 00:06:02.310 --> 00:06:04.110 They want to give you an answer, even if they 00:06:04.110 --> 00:06:06.470 have to fudge it. Precisely. So if you're a lawyer 00:06:06.470 --> 00:06:08.509 or a researcher or maybe you're studying for 00:06:08.509 --> 00:06:11.129 a med school exam, you don't want creativity. 00:06:11.329 --> 00:06:13.910 You want accuracy. Yeah. You want a tool that's 00:06:13.910 --> 00:06:16.069 willing to say, I don't know, instead of lying 00:06:16.069 --> 00:06:18.879 to you. Trustworthy over creative. Yeah. I think 00:06:18.879 --> 00:06:20.660 that's a trade -off a lot of us would make for 00:06:20.660 --> 00:06:24.100 serious work. Speaking of consistency, the source 00:06:24.100 --> 00:06:27.439 also brings up gems. Right. Gemini gems. These 00:06:27.439 --> 00:06:30.879 are basically roles or agents you can save. I 00:06:30.879 --> 00:06:33.480 have to admit, and this is a bit of a vulnerable 00:06:33.480 --> 00:06:36.319 thing to say, but I still wrestle with prompt 00:06:36.319 --> 00:06:39.060 drift myself. I'll get a prompt working just 00:06:39.060 --> 00:06:41.319 perfectly, and then two days later I type it 00:06:41.319 --> 00:06:43.100 just slightly differently and get a totally different 00:06:43.100 --> 00:06:46.000 result. It drives me crazy. That is exactly what 00:06:46.000 --> 00:06:48.879 gems are built to solve. you basically save the 00:06:48.879 --> 00:06:51.259 prompt as a kind of persona. So you can have 00:06:51.259 --> 00:06:52.879 a proposal, write, or gem. You don't have to 00:06:52.879 --> 00:06:54.560 explain all the rules every single time. You 00:06:54.560 --> 00:06:56.740 just open that gem, and it already knows the 00:06:56.740 --> 00:06:59.660 tone, the format, all the constraints. It's all 00:06:59.660 --> 00:07:02.800 about repeatability. And then there's Opal. This 00:07:02.800 --> 00:07:06.100 one sounds new to me. Opal is the lightweight 00:07:06.100 --> 00:07:08.779 automation tool. Think of it like a very, very 00:07:08.779 --> 00:07:12.779 simple Zapier, but it's only for the Google ecosystem. 00:07:13.079 --> 00:07:15.279 So like if I get an email with an invoice, put 00:07:15.279 --> 00:07:18.620 it in a specific doc. Exactly that. Or summarize 00:07:18.620 --> 00:07:20.839 this new Google doc and email it to my boss. 00:07:21.139 --> 00:07:23.519 It's for the non -technical person who just wants 00:07:23.519 --> 00:07:25.500 to connect two Google things without learning 00:07:25.500 --> 00:07:28.100 to code. So going back to Notebook LM for a second, 00:07:28.740 --> 00:07:31.540 if it refuses to invent facts and it's strictly 00:07:31.540 --> 00:07:33.819 bound to the source material, is it actually 00:07:33.819 --> 00:07:36.439 creative at all? Maybe not creative in an artistic 00:07:36.439 --> 00:07:38.560 sense, but it's the only one that's trustworthy. 00:07:38.879 --> 00:07:42.120 And in a world of infinite AI -generated junk, 00:07:42.300 --> 00:07:44.600 trust is becoming the most valuable currency 00:07:44.600 --> 00:07:47.079 there is. That's a really good point. OK, let's 00:07:47.079 --> 00:07:49.839 shift gears to the builders, the developer tools, 00:07:50.259 --> 00:07:51.759 because this seems to be where the landscape 00:07:51.759 --> 00:07:54.199 gets even more fragmented. Yeah. And this is 00:07:54.199 --> 00:07:57.199 where we see that lab versus factory distinction 00:07:57.199 --> 00:08:00.439 really clearly. Explain that. OK, so you have 00:08:00.439 --> 00:08:03.769 AI Studio. The source describes this as the lab 00:08:03.769 --> 00:08:06.470 bench. This is where you go to prototype things. 00:08:06.569 --> 00:08:09.610 You want to test a new prompt. Go to AI Studio. 00:08:09.689 --> 00:08:11.990 You want to see if the newest model is any smarter. 00:08:12.860 --> 00:08:15.040 AI Studio, but you don't build a business there. 00:08:15.100 --> 00:08:17.519 It's a sandbox. In the factory. That would be 00:08:17.519 --> 00:08:19.819 Firebase Studio. That's where you have databases, 00:08:20.300 --> 00:08:23.000 authentication, user logins. That is where you 00:08:23.000 --> 00:08:25.019 build the real app that you actually ship to 00:08:25.019 --> 00:08:27.480 customers. I see. So AI Studio is for the idea. 00:08:27.680 --> 00:08:29.639 Firebase is for the product. And then we've got 00:08:29.639 --> 00:08:31.540 these coding tools with these really sci -fi 00:08:31.540 --> 00:08:35.110 names. Anti -gravity. Jules. Jules is fascinating. 00:08:35.309 --> 00:08:38.250 It represents a shift from helping me code to 00:08:38.250 --> 00:08:40.830 doing the code for me. It's an agent. You tell 00:08:40.830 --> 00:08:43.210 Jules, hey, refactor the entire authentication 00:08:43.210 --> 00:08:45.190 system, and it just goes off and tries to do 00:08:45.190 --> 00:08:47.110 it. Yeah. You're delegating, not just typing 00:08:47.110 --> 00:08:50.529 faster. But the source material is surprisingly 00:08:50.529 --> 00:08:54.409 honest here. It compares Google's tools to things 00:08:54.409 --> 00:08:57.789 like Cursor and Claude. And it's not exactly 00:08:57.789 --> 00:09:00.330 a glowing review for Google. No, it doesn't pull 00:09:00.330 --> 00:09:02.309 any punches. It says pretty straightforwardly. 00:09:02.629 --> 00:09:05.169 Right now, if you're doing complex coding, Cursor 00:09:05.169 --> 00:09:07.210 paired with Claude is better. It's just faster. 00:09:07.250 --> 00:09:09.450 It's more mature. So Google's playing catch up. 00:09:09.710 --> 00:09:12.210 In the dev environment, yeah, for sure. But they 00:09:12.210 --> 00:09:14.210 have the data and the scale, which is their big 00:09:14.210 --> 00:09:16.389 advantage. So that really begs the question. 00:09:16.649 --> 00:09:19.870 Why would anyone invest time in Google's dev 00:09:19.870 --> 00:09:23.470 tools right now if something like Cursor is faster? 00:09:23.789 --> 00:09:26.289 Why learn a tool that's currently second best? 00:09:26.509 --> 00:09:29.529 It's a long game. Google's whole thesis is integration. 00:09:30.190 --> 00:09:32.649 Eventually, having your database, your authentication, 00:09:32.730 --> 00:09:35.830 and your AI all in one ecosystem, that might 00:09:35.830 --> 00:09:38.730 just beat the fragmented best -in -class approach. 00:09:39.149 --> 00:09:41.110 They're betting on scale. The long game. Yeah, 00:09:41.110 --> 00:09:42.809 that seems to be a recurring theme with all of 00:09:42.809 --> 00:09:44.929 this. Okay, we've covered the core, the productivity 00:09:44.929 --> 00:09:47.350 tools, the dev stack. Now let's look at the crystal 00:09:47.350 --> 00:09:51.610 ball. Google Labs. Ah, I love labs. This is the 00:09:51.610 --> 00:09:54.250 experimental playground. The source says you 00:09:54.250 --> 00:09:56.309 should watch the space not because the tools 00:09:56.309 --> 00:09:58.649 are perfect, but because they show you where 00:09:58.649 --> 00:10:00.769 the ship is steering. One of the tools mentioned 00:10:00.769 --> 00:10:04.610 is called Pameli. It analyzes business DNA. What's 00:10:04.610 --> 00:10:07.190 that? This is so cool. So instead of you telling 00:10:07.190 --> 00:10:10.070 the AI right in a professional but friendly tone, 00:10:10.629 --> 00:10:13.769 Pameli scans your website. It looks at your fonts, 00:10:14.009 --> 00:10:16.909 your copy, your layout, and it extracts the vibe. 00:10:17.450 --> 00:10:19.850 Then it writes copy that actually sounds like 00:10:19.850 --> 00:10:22.789 you. It's way better than a generic persona prompt 00:10:22.789 --> 00:10:25.309 because it's based on your actual data. And then 00:10:25.309 --> 00:10:27.970 there's something called gen tabs. Wow. Which 00:10:27.970 --> 00:10:30.929 sounds kind of chaotic. Or brilliant, depending 00:10:30.929 --> 00:10:33.070 on how your brain works. It takes all those open 00:10:33.070 --> 00:10:34.970 browser tabs. You know, when you have 20 tabs 00:10:34.970 --> 00:10:37.809 open doing research, and it turns them into an 00:10:37.809 --> 00:10:40.190 interactive dashboard. It's like an instant app 00:10:40.190 --> 00:10:41.990 made out of your browsing session. These are 00:10:41.990 --> 00:10:43.629 just experiments, right? They could kill them 00:10:43.629 --> 00:10:45.669 off next week. I mean, Google is famous for that. 00:10:46.070 --> 00:10:48.149 So is it really worth using tools that might 00:10:48.149 --> 00:10:51.509 just disappear? Yes. But not for the utility 00:10:51.509 --> 00:10:54.299 today. You use them to spot patterns for tomorrow. 00:10:54.679 --> 00:10:57.000 If you use gen tabs, you start to realize, oh, 00:10:57.000 --> 00:10:59.600 in the future, websites won't just be static 00:10:59.600 --> 00:11:03.019 pages, they'll be dynamic dashboards. You're 00:11:03.019 --> 00:11:05.559 training your intuition. So you're seeing the 00:11:05.559 --> 00:11:08.080 ghost of a future product. Exactly. All right, 00:11:08.080 --> 00:11:11.539 let's move to the fun stuff. Creative and media, 00:11:12.580 --> 00:11:15.159 the wow factor. This is where Google has been 00:11:15.159 --> 00:11:17.779 quietly solving some really massive problems. 00:11:17.899 --> 00:11:20.679 The source brings up Nano Banana, which is a 00:11:20.679 --> 00:11:22.480 great name, by the way. I'm assuming it's not 00:11:22.480 --> 00:11:25.539 about fruit Incredible name But the tech behind 00:11:25.539 --> 00:11:28.740 it is even better. It basically solves the morphing 00:11:28.740 --> 00:11:31.360 problem in AI images. Right, because usually 00:11:31.360 --> 00:11:33.799 you generate a character and then you try to 00:11:33.799 --> 00:11:36.019 generate them again in a different pose and they 00:11:36.019 --> 00:11:37.399 look like a completely different person. It's 00:11:37.399 --> 00:11:39.320 impossible to make a comic book or a storyboard 00:11:39.320 --> 00:11:42.980 that way. Exactly. Nano Banana locks the character's 00:11:42.980 --> 00:11:45.340 consistency. You can edit the image, change the 00:11:45.340 --> 00:11:48.179 background, change the pose, but the person still 00:11:48.179 --> 00:11:50.860 looks like the same person. That's huge. That 00:11:50.860 --> 00:11:52.940 is the difference between a toy and a production 00:11:52.940 --> 00:11:55.879 tool. the V3 video model. This one just glue 00:11:55.879 --> 00:12:00.299 my mind. It generates video with dialogue and 00:12:00.299 --> 00:12:02.659 audio. Whoa, hang on. So we're not just talking 00:12:02.659 --> 00:12:04.799 about a silent clip of a dog running on a beach. 00:12:05.019 --> 00:12:07.399 We're talking about characters actually speaking. 00:12:07.860 --> 00:12:10.779 Yes. Imagine generating a full film where the 00:12:10.779 --> 00:12:13.120 characters look the same in every single shot, 00:12:13.200 --> 00:12:15.919 thanks to that nano banana tech, and they talk 00:12:15.919 --> 00:12:18.639 with synchronized lip movements. That's the dream, 00:12:18.679 --> 00:12:20.980 isn't it? That's the whole Hollywood in a box 00:12:20.980 --> 00:12:23.490 idea we've been hearing about for years. And 00:12:23.490 --> 00:12:25.850 Google calls the workflow for combining all this 00:12:25.850 --> 00:12:28.269 stuff flow. They're really trying to make it 00:12:28.269 --> 00:12:31.649 a seamless studio experience. It feels like we 00:12:31.649 --> 00:12:35.110 are right on the edge of this being usable for 00:12:35.110 --> 00:12:38.750 actual work. So does this consistency, the nano 00:12:38.750 --> 00:12:41.669 banana stuff, does that finally make AI art a 00:12:41.669 --> 00:12:43.809 real production tool? Absolutely. It moves from 00:12:43.809 --> 00:12:46.850 a cool demo to a usable asset. Right. If you 00:12:46.850 --> 00:12:48.750 can't control the output, you can't use it for 00:12:48.750 --> 00:12:50.879 work. Google is solving the control problem. 00:12:51.059 --> 00:12:52.879 OK, finally, let's talk about the layer we don't 00:12:52.879 --> 00:12:55.059 really see, the invisible layer. This is the 00:12:55.059 --> 00:12:57.879 stuff that you don't open. It just happens. Like 00:12:57.879 --> 00:13:00.820 Gemini Nano. Yes. Which is confusingly named, 00:13:01.019 --> 00:13:04.500 but Gemini Nano runs on your device. It's on 00:13:04.500 --> 00:13:06.440 your phone. It doesn't go out to the cloud. Why 00:13:06.440 --> 00:13:08.860 does that matter? I mean, why should I care if 00:13:08.860 --> 00:13:11.360 the AI is on my phone or in some server farm 00:13:11.360 --> 00:13:14.659 in Oregon? Two big reasons. Privacy and speed. 00:13:14.940 --> 00:13:17.379 If it's on your phone, Google doesn't see your 00:13:17.379 --> 00:13:19.539 data. And because it doesn't have to travel to 00:13:19.539 --> 00:13:22.519 a server and back, it's instant. It's what powers 00:13:22.519 --> 00:13:25.360 things like smart replies or on -device summaries. 00:13:26.470 --> 00:13:29.549 Astra is multimodal. OK, define multimodal for 00:13:29.549 --> 00:13:32.129 us. It means it can see, hear, and speak all 00:13:32.129 --> 00:13:35.490 at once. Astra uses your phone's camera to see 00:13:35.490 --> 00:13:38.289 the world in real time. It's the AI looking at 00:13:38.289 --> 00:13:39.750 what you're looking at and understanding it. 00:13:39.950 --> 00:13:42.330 Like, where did I leave my keys? Astra saw where 00:13:42.330 --> 00:13:43.889 you put them down. And then there's the workspace 00:13:43.889 --> 00:13:46.429 integration. Thesaurus calls this the boring 00:13:46.429 --> 00:13:49.669 stuff, summarizing Gmail threads, formula help 00:13:49.669 --> 00:13:53.799 in sheets. Boring, maybe, but... super high value. 00:13:53.960 --> 00:13:56.220 It just reduces friction. It solves the blank 00:13:56.220 --> 00:13:58.860 page problem. You open a Google Doc and the AI 00:13:58.860 --> 00:14:01.000 is already there offering to help you get started. 00:14:01.299 --> 00:14:04.600 It makes me wonder, is the best AI the one we 00:14:04.600 --> 00:14:06.759 don't even realize we're using? Yeah, when it 00:14:06.759 --> 00:14:09.019 just becomes a feature instead of a technology. 00:14:09.879 --> 00:14:12.259 We don't say I'm using the spelling checker AI. 00:14:12.320 --> 00:14:14.799 We just say I'm typing. That's where all of this 00:14:14.799 --> 00:14:17.059 is going. It's just how the phone works. Exactly. 00:14:17.139 --> 00:14:19.360 So we've navigated the labyrinth. We've looked 00:14:19.360 --> 00:14:22.320 at the core, the tools, the dev side, the labs, 00:14:22.700 --> 00:14:25.899 the creative, and the invisible. Let's try to 00:14:25.899 --> 00:14:28.240 bring this home. What's the big idea here? I 00:14:28.240 --> 00:14:30.980 think the big idea is that Google is playing 00:14:30.980 --> 00:14:34.080 a long game. The ecosystem is messy because they're 00:14:34.080 --> 00:14:36.500 essentially building it all in public. So what's 00:14:36.500 --> 00:14:38.279 the strategy for the listener? If I'm sitting 00:14:38.279 --> 00:14:40.980 here, I have limited time. What do I actually 00:14:40.980 --> 00:14:43.240 do with all this information? Don't chase all 00:14:43.240 --> 00:14:46.580 30 tools. You'll just burn out. OK. Focus on 00:14:46.580 --> 00:14:49.600 research, that's Gemini Advanced. Focus on grounding, 00:14:49.899 --> 00:14:52.220 that's Notebook LM. Those are your high leverage 00:14:52.220 --> 00:14:54.620 tools. Keep one eye on labs to see what's coming 00:14:54.620 --> 00:14:57.740 next. And for everything else, use the best tool 00:14:57.740 --> 00:15:00.480 for the job. If cursor is better for coding right 00:15:00.480 --> 00:15:04.220 now, use cursor. Don't force loyalty to an ecosystem 00:15:04.220 --> 00:15:06.159 that's still under construction. That makes a 00:15:06.159 --> 00:15:08.340 lot of sense. The source had a line that really 00:15:08.340 --> 00:15:10.940 stuck with me. It said, being early doesn't mean 00:15:10.940 --> 00:15:15.090 being perfect. It means building intuition. I 00:15:15.090 --> 00:15:16.950 love that. And I think that's the real challenge 00:15:16.950 --> 00:15:18.649 for everyone listening, right? Are you waiting 00:15:18.649 --> 00:15:21.389 for all these tools to be perfect before you 00:15:21.389 --> 00:15:24.889 jump in? Or are you training your intuition right 00:15:24.889 --> 00:15:27.970 now while it's all messy so that when it is perfect, 00:15:28.009 --> 00:15:29.970 you're already an expert? That's the difference 00:15:29.970 --> 00:15:32.549 between consuming the future and actually helping 00:15:32.549 --> 00:15:35.710 to shape it. Well said. See you in the deep end.