WEBVTT 00:00:00.000 --> 00:00:02.359 We're told raw computing power is the ultimate 00:00:02.359 --> 00:00:05.519 sign of AI intelligence. We tend to think bigger 00:00:05.519 --> 00:00:08.599 servers always mean smarter machines. But I've 00:00:08.599 --> 00:00:10.980 been reflecting on this lately, and I wonder 00:00:10.980 --> 00:00:13.859 if we have it entirely backwards. What if true 00:00:13.859 --> 00:00:16.079 machine intelligence actually looks like self 00:00:16.079 --> 00:00:19.079 -doubt? Yeah, that's a really wild thought. Welcome 00:00:19.079 --> 00:00:22.059 to the Deep Dive. Today, we're looking at your 00:00:22.059 --> 00:00:25.120 curated sources. We're tracking a massive pivot 00:00:25.120 --> 00:00:27.960 in the AI landscape. We really are. We're moving 00:00:27.960 --> 00:00:30.059 away from models that just confidently guess. 00:00:30.280 --> 00:00:32.960 We're moving towards systems that pause and reflect. 00:00:33.240 --> 00:00:35.159 Right, systems that perceive the world the way 00:00:35.159 --> 00:00:37.460 we do. Exactly. First, we're going to explore 00:00:37.460 --> 00:00:40.679 Anthropic's new model. It prioritizes knowing 00:00:40.679 --> 00:00:43.359 exactly what it doesn't know. That shift is huge. 00:00:43.579 --> 00:00:45.899 It really is. Then we'll look at Google's Gemini 00:00:45.899 --> 00:00:48.899 embedding, too. That system is unifying the digital 00:00:48.899 --> 00:00:51.939 senses entirely. Into one single model. Yeah. 00:00:52.100 --> 00:00:54.939 And finally, we'll unpack a rather chaotic week, 00:00:55.020 --> 00:00:57.820 a week of billion -dollar valuations and shifted 00:00:57.820 --> 00:01:00.240 narratives. We'll look at those shifting realities 00:01:00.240 --> 00:01:03.159 on AI job losses. We've got a lot of ground to 00:01:03.159 --> 00:01:05.180 cover today. I'm incredibly excited about the 00:01:05.180 --> 00:01:07.439 stack of sources you brought. The anthropic news 00:01:07.439 --> 00:01:09.939 alone is completely game -changing. It challenges 00:01:09.939 --> 00:01:11.760 everything we thought we knew about scaling. 00:01:12.120 --> 00:01:14.340 Let's start right there with this concept of 00:01:14.340 --> 00:01:17.519 trust. Because raw capability really means nothing 00:01:17.519 --> 00:01:20.920 if the system hallucinates. Yeah. Exactly. Anthropic 00:01:20.920 --> 00:01:24.620 just dropped Claude Opus 4 .8. And looking at 00:01:24.620 --> 00:01:26.659 the timeline, this was a bit of a panic drop. 00:01:26.799 --> 00:01:29.260 Oh, it absolutely was. It came out only 41 days 00:01:29.260 --> 00:01:33.620 after Opus 4 .7. And honestly, that previous 00:01:33.620 --> 00:01:37.060 release was somewhat disappointing. People found 00:01:37.060 --> 00:01:39.400 it unreliable for power tasks. They really did. 00:01:39.560 --> 00:01:42.450 So Anthropic went right back to the lab. They 00:01:42.450 --> 00:01:45.170 focused heavily on messy data issues. They basically 00:01:45.170 --> 00:01:47.310 built a system that actively fights back. It 00:01:47.310 --> 00:01:49.170 fights against bad inputs from the user. The 00:01:49.170 --> 00:01:51.530 core feature they're touting is proactive fact 00:01:51.530 --> 00:01:54.530 -checking. Early testers note that Opus 4 .8 00:01:54.530 --> 00:01:57.109 is incredibly self -aware. It actively flags 00:01:57.109 --> 00:01:59.730 uncertainties in its own work. It calls out flawed 00:01:59.730 --> 00:02:02.290 logic before generating an answer. It even identifies 00:02:02.290 --> 00:02:05.510 gaps in your input data. This reminds me of a 00:02:05.510 --> 00:02:07.950 seasoned sous chef. Oh, I like that analogy. 00:02:08.189 --> 00:02:10.669 Yeah, imagine a recipe is missing a crucial step. 00:02:11.189 --> 00:02:13.090 Instead of just guessing and ruining the dish 00:02:13.090 --> 00:02:15.490 entirely, the chef stops cooking and turns around. 00:02:15.810 --> 00:02:18.789 They ask you for clarification instead. That's 00:02:18.789 --> 00:02:21.349 a fundamental shift in AI behavior. It really 00:02:21.349 --> 00:02:23.689 is. And I gotta admit, I still wrestle with prompt 00:02:23.689 --> 00:02:27.210 drift myself. Two sec silence. Prompt drift is 00:02:27.210 --> 00:02:29.990 when the AI slowly loses the original context. 00:02:30.250 --> 00:02:32.969 It loses track of your instructions over a long 00:02:32.969 --> 00:02:35.270 conversation. Which is so deeply frustrating. 00:02:35.740 --> 00:02:38.120 It is, especially when an AI confidently lies 00:02:38.120 --> 00:02:40.780 to you. It just hallucinates facts to keep the 00:02:40.780 --> 00:02:43.280 conversation moving. Right. And that is exactly 00:02:43.280 --> 00:02:46.219 the core problem they've solved here. It completely 00:02:46.219 --> 00:02:49.659 changes relying on AI for serious, complex work. 00:02:50.139 --> 00:02:52.819 The mechanism behind it is absolutely fascinating, 00:02:53.020 --> 00:02:55.139 too. How so? Well, it doesn't just guess anymore. 00:02:55.300 --> 00:02:57.819 It builds an internal confidence score as it 00:02:57.819 --> 00:03:00.750 processes. If that score drops below a mathematical 00:03:00.750 --> 00:03:03.610 threshold, it halts. It triggers a halt in query 00:03:03.610 --> 00:03:05.669 response immediately. That makes perfect sense. 00:03:05.849 --> 00:03:08.530 Plus, Anthropic launched a dynamic workflows 00:03:08.530 --> 00:03:11.530 research preview alongside this. That preview 00:03:11.530 --> 00:03:14.009 orchestrates hundreds of parallel subagents all 00:03:14.009 --> 00:03:16.750 at once. And they have massive code migrations 00:03:16.750 --> 00:03:19.349 working now. Wait, you mean pairing it with Claude 00:03:19.349 --> 00:03:23.129 code? Yeah. If you pair Opus 4 .8 with Claude 00:03:23.129 --> 00:03:25.960 code today. It handles architectural migrations 00:03:25.960 --> 00:03:29.599 across hundreds of thousands of lines. That scale 00:03:29.599 --> 00:03:31.740 of coding autonomy is staggering to think about. 00:03:31.860 --> 00:03:34.099 It maps the entire architecture in its working 00:03:34.099 --> 00:03:36.879 memory. It spots dependencies and asks questions 00:03:36.879 --> 00:03:39.620 when conflicts arise. It does. And they also 00:03:39.620 --> 00:03:42.599 hinted at their looming mythos model. Ah, right. 00:03:43.180 --> 00:03:45.719 It's currently on ice due to strict cybersecurity 00:03:45.719 --> 00:03:48.780 guardrails, but they mentioned these final safety 00:03:48.780 --> 00:03:50.419 checks are wrapping up. We should see it roll 00:03:50.419 --> 00:03:52.219 out in the coming weeks. I'm definitely watching 00:03:52.219 --> 00:03:54.379 that one very closely. So looking at this whole 00:03:54.379 --> 00:03:56.780 landscape, I have a question. Is this safety 00:03:56.780 --> 00:03:59.020 -first self -awareness the only viable path? 00:03:59.259 --> 00:04:01.379 Is this how we build production -grade autonomous 00:04:01.379 --> 00:04:04.280 systems, systems that don't need constant human 00:04:04.280 --> 00:04:07.219 babysitting? It absolutely is, and here is why. 00:04:07.719 --> 00:04:10.460 Enterprise adoption completely stalls without 00:04:10.460 --> 00:04:14.509 deep, undeniable trust. Major companies will 00:04:14.509 --> 00:04:17.629 not deploy an AI that just guesses. A confident 00:04:17.629 --> 00:04:20.470 hallucination in a legal brief is dangerous. 00:04:20.790 --> 00:04:23.189 In a medical diagnosis, it can cost millions 00:04:23.189 --> 00:04:26.170 of dollars. Nobody wants to explain that an algorithm 00:04:26.170 --> 00:04:28.389 hallucinated financial metrics. Yeah, that's 00:04:28.389 --> 00:04:31.250 a total boardroom nightmare. Exactly. Big businesses 00:04:31.250 --> 00:04:33.899 need systems that ask for help. They need models 00:04:33.899 --> 00:04:36.339 that independently verify facts before acting. 00:04:36.480 --> 00:04:39.819 If you want to deploy AI across a Fortune 500 00:04:39.819 --> 00:04:42.600 company, baking that hesitation into the foundation 00:04:42.600 --> 00:04:45.079 is the only way. So building enterprise trust 00:04:45.079 --> 00:04:48.079 means programming doubt directly into the model. 00:04:48.240 --> 00:04:50.980 Spot on. Doubt is the absolute foundation of 00:04:50.980 --> 00:04:52.879 corporate reliability. If enterprise companies 00:04:52.879 --> 00:04:56.269 are finally trusting AI because of doubt. The 00:04:56.269 --> 00:04:58.569 next hurdle is how it perceives the real world. 00:04:58.649 --> 00:05:01.129 Quiet. From an AI that thinks more clearly. We 00:05:01.129 --> 00:05:03.550 naturally transition to an AI that perceives 00:05:03.550 --> 00:05:06.329 more seamlessly. Because a text -only AI is essentially 00:05:06.329 --> 00:05:09.069 blind. Right. Let's talk about Google's Gemini 00:05:09.069 --> 00:05:11.829 embedding too. This is basically the one model 00:05:11.829 --> 00:05:15.050 to rule them all. It processes text, audio, video, 00:05:15.250 --> 00:05:18.290 and images simultaneously. Usually you need a 00:05:18.290 --> 00:05:20.889 completely different model for each format. You 00:05:20.889 --> 00:05:22.610 have one brain for reading and one for seeing. 00:05:22.970 --> 00:05:26.279 But this new model. handles all four formats 00:05:26.279 --> 00:05:29.459 perfectly. It lives inside one single unified 00:05:29.459 --> 00:05:32.600 system. This replaces what developers know is 00:05:32.600 --> 00:05:35.220 an absolute nightmare. Developers used to have 00:05:35.220 --> 00:05:37.500 to stitch three different databases together. 00:05:37.660 --> 00:05:39.439 Yeah, it was basically digital duct tape. Exactly. 00:05:39.819 --> 00:05:43.500 They did this just to achieve basic multimodal 00:05:43.500 --> 00:05:46.120 AI. You'd have to translate a video into text 00:05:46.120 --> 00:05:48.339 tags just so the search engine could understand 00:05:48.339 --> 00:05:50.980 it at all. Right. But now you can search with 00:05:50.980 --> 00:05:52.879 a simple image. You take an image of a broken 00:05:52.879 --> 00:05:55.579 pipe. You get a specific repair video back as 00:05:55.579 --> 00:05:58.220 an answer. You don't rely on text tags at all. 00:05:58.680 --> 00:06:02.480 Beat. Whoa, imagine scaling to a billion queries. 00:06:02.920 --> 00:06:04.639 Doing that across different media types used 00:06:04.639 --> 00:06:07.259 to meld servers. It's completely wild. And it's 00:06:07.259 --> 00:06:09.379 hitting number one on the leaderboards. Oh, it 00:06:09.379 --> 00:06:12.560 dominates image and video search benchmarks globally 00:06:12.560 --> 00:06:16.420 now. It's winning at complex coding tasks and 00:06:16.420 --> 00:06:19.259 text translation. And what's crazy is it works 00:06:19.259 --> 00:06:22.139 out of the box. On incredibly niche topics, right? 00:06:22.199 --> 00:06:24.439 Yeah, topics it was never explicitly fine -tuned 00:06:24.439 --> 00:06:27.240 on. We're talking about deep space astronomy 00:06:27.240 --> 00:06:31.220 imagery. Or even fine dining plating techniques. 00:06:31.560 --> 00:06:34.519 It actually beats Google's older text -only models. 00:06:34.779 --> 00:06:37.300 It beats them at their own text -only game, which 00:06:37.300 --> 00:06:39.660 is counterintuitive. It really is. And developers 00:06:39.660 --> 00:06:41.600 can start building with this unified system today. 00:06:41.839 --> 00:06:43.439 Yeah, it's available right now on the Gemini 00:06:43.439 --> 00:06:46.060 API. It's also live on Vertex AI. If you've ever 00:06:46.060 --> 00:06:48.040 built a multimedia application, you know. The 00:06:48.040 --> 00:06:50.670 pain is very real. Running three separate databases 00:06:50.670 --> 00:06:53.529 and praying they play nice is awful. This release 00:06:53.529 --> 00:06:56.350 is a massive win for CrossModalArchie. Let me 00:06:56.350 --> 00:06:58.310 just define CrossModalArchie for you quickly. 00:06:58.470 --> 00:07:01.769 AI retrieving facts across text, audio, and video 00:07:01.769 --> 00:07:03.970 to answer. Exactly. Wait, but I want to push 00:07:03.970 --> 00:07:06.470 back on the architecture here. Why wouldn't three 00:07:06.470 --> 00:07:09.329 specialized databases be better? One built for 00:07:09.329 --> 00:07:11.949 video, one for text, and one for audio. Usually 00:07:11.949 --> 00:07:14.389 a jack of all trades is a master of none. That's 00:07:14.389 --> 00:07:17.009 a really fair point. So why does this generalized 00:07:17.009 --> 00:07:20.670 multisensory model win at text? That's the perfect 00:07:20.670 --> 00:07:23.310 question to ask. It comes down to how the model 00:07:23.310 --> 00:07:26.430 builds conceptual maps. Understanding the relationship 00:07:26.430 --> 00:07:29.750 between an image and text is powerful. It builds 00:07:29.750 --> 00:07:32.129 a much deeper conceptual map than text alone. 00:07:32.540 --> 00:07:34.600 Because it actually sees the connection. Right. 00:07:34.680 --> 00:07:37.579 By training on images and text in the same mathematical 00:07:37.579 --> 00:07:41.019 space, the AI learns intrinsic deep connections 00:07:41.019 --> 00:07:44.000 between concepts. When it sees the visual context 00:07:44.000 --> 00:07:47.319 of a word, it understands. It doesn't just translate. 00:07:47.399 --> 00:07:49.579 It truly understands the underlying reality. 00:07:49.800 --> 00:07:51.519 That makes perfect sense. It's like stacking 00:07:51.519 --> 00:07:54.139 Lego blocks of data. The visual data anchors 00:07:54.139 --> 00:07:57.079 the text data perfectly. It learns the word spherical 00:07:57.079 --> 00:07:59.439 and an image of a baseball. They share the exact 00:07:59.439 --> 00:08:01.920 same coordinates in its brain. A single brain 00:08:01.920 --> 00:08:04.920 understanding all media sees patterns that disconnected 00:08:04.920 --> 00:08:08.139 databases miss completely. Precisely. It connects 00:08:08.139 --> 00:08:10.240 dots across disciplines that separate systems 00:08:10.240 --> 00:08:12.660 cannot see. Building these foundational brains 00:08:12.660 --> 00:08:15.730 requires massive leaps in reasoning. It also 00:08:15.730 --> 00:08:18.430 requires staggering amounts of cash to pull off. 00:08:18.649 --> 00:08:20.670 We need to zoom out and look at the business 00:08:20.670 --> 00:08:23.050 landscape. Unbelievable amounts of money are 00:08:23.050 --> 00:08:24.870 fueling these breakthroughs right now. Yeah, 00:08:24.930 --> 00:08:26.990 and massive corporate realignments are happening 00:08:26.990 --> 00:08:29.790 as we speak. OpenAI is making some major lineup 00:08:29.790 --> 00:08:31.689 changes, for instance. They're officially retiring 00:08:31.689 --> 00:08:35.929 GPT -5 .2 and GPT -5 .3 codecs. They're cleaning 00:08:35.929 --> 00:08:39.529 house a bit. Moving forward, GPT -5 .5 will become 00:08:39.529 --> 00:08:42.049 the default model. For all free users, right? 00:08:42.230 --> 00:08:44.529 Exactly. The older models stay available through 00:08:44.529 --> 00:08:47.409 the API for developers. Yeah. But the consumer 00:08:47.409 --> 00:08:50.230 facing side is getting a massive upgrade. Meanwhile, 00:08:50.490 --> 00:08:53.330 CEO Sam Altman made a surprising admission this 00:08:53.330 --> 00:08:55.789 week. He publicly admitted he was pretty wrong 00:08:55.789 --> 00:08:58.690 about AI job losses. That definitely caught my 00:08:58.690 --> 00:09:01.570 eye. He meant near term job losses specifically, 00:09:01.970 --> 00:09:04.389 acknowledging they haven't materialized the way 00:09:04.389 --> 00:09:06.950 he previously predicted. Let's challenge the 00:09:06.950 --> 00:09:09.700 timing of this admission, though. Is it just 00:09:09.700 --> 00:09:12.220 pure coincidence this reality check happens now? 00:09:12.360 --> 00:09:15.440 This is right before a massive, highly rumored 00:09:15.440 --> 00:09:18.960 IPO push? Or is this strategic table setting 00:09:18.960 --> 00:09:21.460 for big investors? So they don't get spooked, 00:09:21.460 --> 00:09:24.100 you mean? It definitely feels like careful expectation 00:09:24.100 --> 00:09:26.980 management to me. When you're asking Wall Street 00:09:26.980 --> 00:09:30.860 for billions, you clear the air. You don't want 00:09:30.860 --> 00:09:33.240 unexpected job loss controversies before going 00:09:33.240 --> 00:09:35.659 public. Right. You don't want congressional hearings 00:09:35.659 --> 00:09:38.399 disrupting your roadshow. You want the narrative 00:09:38.399 --> 00:09:42.000 to be about productivity, not unemployment. But 00:09:42.000 --> 00:09:44.379 honestly, look at the staggering money flowing 00:09:44.379 --> 00:09:47.320 elsewhere. Anthropic just raised $65 billion. 00:09:47.759 --> 00:09:50.740 Yeah, they're sitting at a mind -bending $965 00:09:50.740 --> 00:09:54.139 billion valuation. That is also happening right 00:09:54.139 --> 00:09:56.320 ahead of their expected IPO. Their enterprise 00:09:56.320 --> 00:09:58.620 business is what fascinates me the most. That 00:09:58.620 --> 00:10:01.340 valuation is not just startup vaporware anymore. 00:10:01.620 --> 00:10:05.039 Their enterprise side surged to a $47 billion 00:10:05.039 --> 00:10:08.679 revenue run rate. That is real, undeniable corporate 00:10:08.679 --> 00:10:11.509 adoption. We're moving way past just venture 00:10:11.509 --> 00:10:13.710 capital echo chambers now. And look at Cognition, 00:10:13.789 --> 00:10:16.049 the maker of the Devin AI coder. Right. They 00:10:16.049 --> 00:10:18.289 just raised a billion dollars. At a $25 billion 00:10:18.289 --> 00:10:21.629 valuation. That valuation is up from $10 .2 billion 00:10:21.629 --> 00:10:24.330 in just eight months. The growth curve is almost 00:10:24.330 --> 00:10:28.669 vertical. Devin now drives a $492 million revenue 00:10:28.669 --> 00:10:32.529 run rate. Enterprise usage is growing 50 % every 00:10:32.529 --> 00:10:34.960 single month. People are actually paying real 00:10:34.960 --> 00:10:37.919 money for autonomous AI coders. They're augmenting 00:10:37.919 --> 00:10:40.100 their engineering teams actively with these agents. 00:10:40.279 --> 00:10:42.320 And the legacy tech giants aren't sitting still 00:10:42.320 --> 00:10:45.600 either. Apple and Amazon are making huge, aggressive 00:10:45.600 --> 00:10:48.340 moves. The recent Apple leaks show a brand new 00:10:48.340 --> 00:10:51.440 Siri chatbot app. It has long -term memory and 00:10:51.440 --> 00:10:53.860 document uploads built natively. It features 00:10:53.860 --> 00:10:56.259 Gemini -powered search directly inside iPhones. 00:10:56.750 --> 00:10:58.769 They're baking this intelligence right into the 00:10:58.769 --> 00:11:00.590 hardware level. Yeah, and Amazon is preparing 00:11:00.590 --> 00:11:03.690 to add SpaceX's Grok AI. They're integrating 00:11:03.690 --> 00:11:05.870 it deeply into their flagship enterprise service. 00:11:06.320 --> 00:11:08.580 This gives cloud customers access to Elon Musk's 00:11:08.580 --> 00:11:11.399 data ecosystem. Even YouTube is becoming a serious 00:11:11.399 --> 00:11:13.200 competitor in this space. They're going after 00:11:13.200 --> 00:11:16.179 Spotify directly now. Right. They added AI podcast 00:11:16.179 --> 00:11:19.179 suggestions and adaptive playback speed. On -the 00:11:19.179 --> 00:11:20.899 -go listening just dropped for their premium 00:11:20.899 --> 00:11:23.379 users. But looking at all these massive numbers, 00:11:23.500 --> 00:11:26.580 I have to ask, are these astronomical valuations 00:11:26.580 --> 00:11:30.080 genuinely justified? Nearly a trillion dollars 00:11:30.080 --> 00:11:33.179 for Anthropic. Are they justified by the enterprise 00:11:33.179 --> 00:11:35.519 run rates we're seeing? I really think they finally 00:11:35.519 --> 00:11:37.919 are justified. For a long time, it was mostly 00:11:37.919 --> 00:11:41.399 just future potential. But a $47 billion run 00:11:41.399 --> 00:11:44.779 rate for Anthropic is massive. A nearly half 00:11:44.779 --> 00:11:48.399 billion dollar run rate for Devin is huge. Companies 00:11:48.399 --> 00:11:51.120 are seeing a direct, measurable return on investment. 00:11:51.299 --> 00:11:54.100 They're replacing incredibly expensive, clunky 00:11:54.100 --> 00:11:56.919 legacy software with these agents. That makes 00:11:56.919 --> 00:11:59.500 a lot of sense. If an AI migrates 100 ,000 lines 00:11:59.500 --> 00:12:02.940 of code, it does it over a single weekend. The 00:12:02.940 --> 00:12:05.159 cost savings justify the subscription price 100 00:12:05.159 --> 00:12:07.820 times over. The math is actually starting to 00:12:07.820 --> 00:12:09.940 check out. The massive valuations are finally 00:12:09.940 --> 00:12:12.000 being backed by undeniable enterprise revenue 00:12:12.000 --> 00:12:14.419 growth. Sponsor M. We spent a lot of time on 00:12:14.419 --> 00:12:16.000 foundational models today. We talked about the 00:12:16.000 --> 00:12:17.879 billions of dollars funding them. Right. But 00:12:17.879 --> 00:12:19.899 how does this trickle down to everyday tools? 00:12:20.139 --> 00:12:22.940 Let's ground this entirely in reality. There's 00:12:22.940 --> 00:12:25.759 a fascinating new iOS app making waves called 00:12:25.759 --> 00:12:29.100 Sesame. Sesame is brilliant. It lets you talk 00:12:29.100 --> 00:12:32.409 completely naturally with AI agents. These agents 00:12:32.409 --> 00:12:35.009 remember the deep context from previous conversations. 00:12:35.250 --> 00:12:38.029 They can search the live web in real time while 00:12:38.029 --> 00:12:41.169 speaking. It responds in a fluid human -like... 00:12:41.519 --> 00:12:43.720 way yeah it feels like a real conversation unlike 00:12:43.720 --> 00:12:45.860 older voice assistants then there's an enterprise 00:12:45.860 --> 00:12:48.580 app called pancake this one really caught my 00:12:48.580 --> 00:12:50.840 attention in the sources it acts a lot like the 00:12:50.840 --> 00:12:53.019 open cloud framework inside slack oh pancake 00:12:53.019 --> 00:12:55.779 is wild it essentially makes your entire company 00:12:55.779 --> 00:12:58.299 autonomous you can create unique agents with 00:12:58.299 --> 00:13:00.600 defined roles and goals then they have what they 00:13:00.600 --> 00:13:03.620 call a heartbeat right exactly they run a continuous 00:13:03.620 --> 00:13:07.039 background process to work while you sleep Pancake 00:13:07.039 --> 00:13:09.740 is literally like hiring a digital night shift 00:13:09.740 --> 00:13:12.580 worker. They never clock out and they never need 00:13:12.580 --> 00:13:15.059 a coffee break. They just keep executing tasks. 00:13:15.460 --> 00:13:18.159 It entirely changes what a lean startup can accomplish. 00:13:18.419 --> 00:13:20.879 It really levels the playing field for small 00:13:20.879 --> 00:13:23.720 teams. There is also a great new tool called 00:13:23.720 --> 00:13:26.879 Pitch Agent. I saw that one. It generates on 00:13:26.879 --> 00:13:29.159 -brand presentation slides from a simple text 00:13:29.159 --> 00:13:32.039 prompt. It reads complex file attachments and 00:13:32.039 --> 00:13:35.690 extracts the core narrative. Then, it refines 00:13:35.690 --> 00:13:38.049 the presentation via a simple chat interface. 00:13:38.409 --> 00:13:40.649 Until it looks exactly how you want it. Google 00:13:40.649 --> 00:13:42.889 is also pushing hard on the creative front. They 00:13:42.889 --> 00:13:45.230 just released their ultimate video prompting 00:13:45.230 --> 00:13:48.090 guide. Specifically for mastering Gemini Omni, 00:13:48.110 --> 00:13:51.269 right? Yes. The guide covers five core strategies 00:13:51.269 --> 00:13:53.750 for advanced prompting. It includes cinematic 00:13:53.750 --> 00:13:56.190 control techniques and specific camera angles. 00:13:56.409 --> 00:13:59.129 It gives you pre -made prompts to generate incredibly 00:13:59.129 --> 00:14:01.909 realistic video. And finally, on the marketing 00:14:01.909 --> 00:14:04.830 side. There is spots now. Right. It meticulously 00:14:04.830 --> 00:14:07.450 tracks who is advertising on every single podcast. 00:14:07.629 --> 00:14:09.769 It shows exactly what they spend and where campaigns 00:14:09.769 --> 00:14:12.389 run. It is pure market intelligence. It's huge 00:14:12.389 --> 00:14:14.830 for ad buyers. But looking at tools like Pancake, 00:14:14.889 --> 00:14:17.470 I have to ask, what happens to human middle management 00:14:17.470 --> 00:14:20.029 when apps like Pancake have defined roles operating 00:14:20.029 --> 00:14:23.110 autonomously inside Slack? Management is completely 00:14:23.110 --> 00:14:25.750 shifting its fundamental focus right now. You 00:14:25.750 --> 00:14:27.730 have to stop managing human workflows entirely. 00:14:27.929 --> 00:14:30.759 You start managing digital outputs instead. Right. 00:14:30.940 --> 00:14:33.340 The role becomes much more like an editor -in 00:14:33.340 --> 00:14:36.120 -chief. The AI does all the tedious heavy lifting 00:14:36.120 --> 00:14:39.000 and drafting. Your job is just to verify the 00:14:39.000 --> 00:14:41.700 absolute quality. You ensure the final product 00:14:41.700 --> 00:14:44.139 aligns with strategic company goals. Exactly. 00:14:44.299 --> 00:14:46.600 You're guiding the ship, not rowing the oars. 00:14:46.779 --> 00:14:49.379 We are transitioning from using AI as software 00:14:49.379 --> 00:14:53.059 to managing AI as employees. That is a brilliant 00:14:53.059 --> 00:14:55.240 way to summarize the whole shift. If we look 00:14:55.240 --> 00:14:58.159 at the big picture today, the common thread weaving 00:14:58.159 --> 00:15:01.120 through all these sources is autonomy. Autonomy 00:15:01.120 --> 00:15:05.000 built firmly on undeniable trust. We saw Opus 00:15:05.000 --> 00:15:08.139 4 .8 actively doubting itself and asking questions. 00:15:08.340 --> 00:15:10.899 We saw Gemini embedding to understanding the 00:15:10.899 --> 00:15:13.159 world visually and orally, building deep conceptual 00:15:13.159 --> 00:15:16.379 maps of our shared reality. And we saw autonomous 00:15:16.379 --> 00:15:18.299 agents like Pancake working independently in 00:15:18.299 --> 00:15:20.139 Slack. We're no longer just building simple chat 00:15:20.139 --> 00:15:22.000 interfaces. We are building duply integrated 00:15:22.000 --> 00:15:24.639 systems meant to operate entirely independently. 00:15:25.230 --> 00:15:27.789 It is a profound shift in how we interact with 00:15:27.789 --> 00:15:31.450 machines. If Opus 4 .8 proves that the next great 00:15:31.450 --> 00:15:34.210 era of AI is about the model knowing exactly 00:15:34.210 --> 00:15:37.809 what it doesn't know, two sec silence, how long 00:15:37.809 --> 00:15:40.169 until these autonomous models start proactively 00:15:40.169 --> 00:15:42.610 interviewing us? Interviewing us to fill in the 00:15:42.610 --> 00:15:45.289 gaps of human logic. Wow. That's a heavy thought 00:15:45.289 --> 00:15:47.049 to leave on. Thank you for coming along on this 00:15:47.049 --> 00:15:49.330 deep dive with us today. Keep learning and keep 00:15:49.330 --> 00:15:51.110 questioning everything. Otiro music.