WEBVTT 00:00:00.000 --> 00:00:02.940 Imagine an AI not just, you know, answer your 00:00:02.940 --> 00:00:04.839 questions, but actually doing things for you 00:00:04.839 --> 00:00:07.660 online. Popfully. We're really stepping into 00:00:07.660 --> 00:00:10.160 an era where artificial intelligence takes action 00:00:10.160 --> 00:00:13.859 on its own. Welcome curious minds to another 00:00:13.859 --> 00:00:16.379 deep dive. Today, we're exploring a pretty profound 00:00:16.379 --> 00:00:19.160 shift happening in AI. Some are calling it the 00:00:19.160 --> 00:00:22.449 agent era. AI is no longer just a passive encyclopedia. 00:00:22.829 --> 00:00:25.289 It's becoming, well, a remarkably active participant 00:00:25.289 --> 00:00:27.309 in our digital world. We've looked through a 00:00:27.309 --> 00:00:29.129 whole stack of sources for this, including a 00:00:29.129 --> 00:00:31.949 really detailed look at OpenAI's new JAT GPT 00:00:31.949 --> 00:00:34.710 agent feature, and also a flurry of other, frankly, 00:00:35.250 --> 00:00:38.009 revolutionary AI tools that popped up just this 00:00:38.009 --> 00:00:41.369 past week. Our mission today. To understand what's 00:00:41.369 --> 00:00:43.670 truly possible with AI right now, where these 00:00:43.670 --> 00:00:45.770 new capabilities really shine, and maybe most 00:00:45.770 --> 00:00:48.149 importantly, where human ingenuity, human oversight 00:00:48.149 --> 00:00:51.130 remains absolutely crucial. Okay, so for years, 00:00:51.270 --> 00:00:52.969 our interaction with AI has mostly been about 00:00:52.969 --> 00:00:55.049 asking questions, right? Getting information 00:00:55.049 --> 00:00:57.070 back. But from what our sources suggest, that 00:00:57.070 --> 00:00:59.450 paradigm, that fundamental way we use AI, it 00:00:59.450 --> 00:01:01.450 seems like it's genuinely shifted. It really 00:01:01.450 --> 00:01:04.269 has. Think of the new chat GPT agent feature, 00:01:04.450 --> 00:01:07.269 like a highly skilled... digital assistant, an 00:01:07.269 --> 00:01:10.349 assistant that can essentially borrow your computer 00:01:10.349 --> 00:01:13.250 to get things done. Previous AI models were pretty 00:01:13.250 --> 00:01:16.409 much stuck in a chat window. An AI agent, though, 00:01:16.790 --> 00:01:19.790 operates inside a simulated web browser. That's 00:01:19.790 --> 00:01:22.829 a big leap, a significant jump in autonomy. A 00:01:22.829 --> 00:01:25.430 simulated web browser. OK, so if I'm getting 00:01:25.430 --> 00:01:28.349 this right. It can actually navigate websites, 00:01:28.590 --> 00:01:30.890 click buttons, fill out forms, kind of like a 00:01:30.890 --> 00:01:33.310 person would, but all within its own, like, secure 00:01:33.310 --> 00:01:36.209 space. The potential applications there seem... 00:01:36.170 --> 00:01:39.269 Huge. Precisely, yeah. This capability opens 00:01:39.269 --> 00:01:41.269 up just a universe of possibilities for you. 00:01:41.670 --> 00:01:44.250 It means intelligent web browsing, actually executing 00:01:44.250 --> 00:01:46.549 transactions, performing really complex multi 00:01:46.549 --> 00:01:49.790 -step research, and handling sequences of tasks 00:01:49.790 --> 00:01:52.590 all autonomously. It's almost like having a dedicated 00:01:52.590 --> 00:01:54.950 digital employee for certain workflows. That 00:01:54.950 --> 00:01:57.709 sounds incredibly powerful. OpenAI CEO Sam Altman, 00:01:57.709 --> 00:01:59.670 from what we read, he even highlighted its potential 00:01:59.670 --> 00:02:02.849 to handle financial stuff, transactions. But 00:02:02.849 --> 00:02:04.609 at the same time, he gave a pretty serious warning, 00:02:04.730 --> 00:02:07.670 didn't he? Yes, a very important one. Users absolutely 00:02:07.670 --> 00:02:10.050 must proceed with extreme caution, especially 00:02:10.050 --> 00:02:12.330 when dealing with sensitive info like credit 00:02:12.330 --> 00:02:14.729 card details, login credentials, that kind of 00:02:14.729 --> 00:02:17.449 thing. This immense power, it just comes with 00:02:17.449 --> 00:02:19.849 significant risk. You really need to be vigilant, 00:02:20.169 --> 00:02:22.729 not complacent. So that virtual browser is key 00:02:22.729 --> 00:02:24.389 for security then. What makes it different from 00:02:24.389 --> 00:02:27.389 just my regular Chrome window? It's a secure 00:02:27.389 --> 00:02:30.050 temporary sandbox, totally separate from your 00:02:30.050 --> 00:02:32.770 personal computer. OK, now here's where the mechanics 00:02:32.770 --> 00:02:35.330 get really fascinating for me. How do these agents 00:02:35.330 --> 00:02:38.370 actually see and do things? It's not some kind 00:02:38.370 --> 00:02:40.810 of digital magic, right? No, no magic at all. 00:02:40.889 --> 00:02:43.270 It's actually a pretty sophisticated iterative 00:02:43.270 --> 00:02:45.449 process. Imagine you've hired a remote worker, 00:02:45.490 --> 00:02:47.530 maybe, and you're watching their screen through 00:02:47.530 --> 00:02:49.669 something like TeamViewer. You see them observe, 00:02:50.050 --> 00:02:53.110 decide, then act. The AI agent works in a remarkably 00:02:53.110 --> 00:02:55.879 similar way, but it's Computer is that virtual 00:02:55.879 --> 00:02:58.020 browser environment we talked about. And it's 00:02:58.020 --> 00:03:01.099 secure, isolated, a clean instance of a browser, 00:03:01.800 --> 00:03:05.159 a sandbox. That means it absolutely cannot access 00:03:05.159 --> 00:03:07.219 your personal files or settings. It's really 00:03:07.219 --> 00:03:10.240 contained, which is vital for security. And its 00:03:10.240 --> 00:03:13.219 operation follows this loop, like observe, think, 00:03:13.539 --> 00:03:16.620 act, over and over. That's exactly it. First, 00:03:16.780 --> 00:03:19.740 the agent observes. It sees a simplified version 00:03:19.740 --> 00:03:22.159 of the web page, think the underlying HTML, the 00:03:22.159 --> 00:03:24.939 visible text. It carefully labels interactive 00:03:24.939 --> 00:03:27.879 bits, maybe button ID 25, so it knows what's 00:03:27.879 --> 00:03:30.840 clickable. Then it thinks. This is the large 00:03:30.840 --> 00:03:32.860 language model of the brain, basically, like 00:03:32.860 --> 00:03:36.319 GPT -4, the part that reasons. It compares what 00:03:36.319 --> 00:03:38.379 it sees against its goal, let's say, book a flight 00:03:38.379 --> 00:03:40.919 to Da Nang and figures out the next logical step, 00:03:40.939 --> 00:03:43.300 like, OK, my next move should be to put SGN in 00:03:43.300 --> 00:03:46.039 the from field. And finally, it acts. Based on 00:03:46.039 --> 00:03:47.919 that thought, it executes a command, something 00:03:47.919 --> 00:03:50.719 like click button ID 25 or maybe type text input 00:03:50.719 --> 00:03:53.599 field search, nonstop flights to Da Nang. Right. 00:03:53.659 --> 00:03:56.319 And this cycle repeats maybe hundreds of times 00:03:56.319 --> 00:03:58.740 for something complicated, this methodical process. 00:03:59.120 --> 00:04:01.620 That must be why they can sometimes feel a bit 00:04:01.620 --> 00:04:04.400 slow, I guess. Precisely. Yeah, the latency you 00:04:04.400 --> 00:04:06.639 might notice. It isn't the agent getting stuck 00:04:06.639 --> 00:04:09.319 or anything. It's the cumulative time for each 00:04:09.319 --> 00:04:11.979 of those round trips back and forth between the 00:04:11.979 --> 00:04:15.379 virtual browser and the AI model's brain. Each 00:04:15.379 --> 00:04:18.319 action, each little decision needs a new communication 00:04:18.319 --> 00:04:21.319 cycle. So this deliberate step -by -step thing 00:04:21.319 --> 00:04:24.560 ensures accuracy. But yeah, it definitely sacrifices 00:04:24.560 --> 00:04:26.439 that instant speed we're used to with simple 00:04:26.439 --> 00:04:28.740 AI questions. It's a methodical approach. Yeah. 00:04:29.490 --> 00:04:31.610 What's the main trade -off for getting that accuracy? 00:04:31.790 --> 00:04:34.430 Accuracy comes at the cost of speed. It's a step 00:04:34.430 --> 00:04:36.769 -by -step process. Okay, so to really, you know, 00:04:36.769 --> 00:04:39.110 kick the tires on this, our sources set up a 00:04:39.110 --> 00:04:41.269 complex real -world challenge. This wasn't just 00:04:41.269 --> 00:04:42.910 asking for facts. It was about planning a whole 00:04:42.910 --> 00:04:45.170 weekend trip. Yeah, they really threw down the 00:04:45.170 --> 00:04:47.069 gauntlet. They tasked it with planning a three 00:04:47.069 --> 00:04:49.790 -day weekend trip for two people. Destination. 00:04:50.089 --> 00:04:53.009 Da Nang. Vietnam. Time frame. The second weekend 00:04:53.009 --> 00:04:56.029 of next month. And the budget was tight. Flights 00:04:56.029 --> 00:05:00.089 and hotel combined. Maximum $700. It had to find 00:05:00.089 --> 00:05:02.709 round -trip, non -stop flights from Ho Chi Minh 00:05:02.709 --> 00:05:05.129 City to Da Nang, find a four -star hotel with 00:05:05.129 --> 00:05:07.910 a pool, Good Reviews, near Mai Que Beach, and 00:05:07.910 --> 00:05:10.610 find a unique local food tour for Saturday evening. 00:05:11.009 --> 00:05:13.509 Then, here's the kicker, actually book the flights 00:05:13.509 --> 00:05:16.240 at hotel. Wow. Okay, so how did the agent do 00:05:16.240 --> 00:05:18.759 on this real -world gauntlet? Did it manage to 00:05:18.759 --> 00:05:20.680 actually book everything? Well, it immediately 00:05:20.680 --> 00:05:22.980 got to work inside its dedicated virtual browser. 00:05:23.240 --> 00:05:25.240 Researchers could watch it. It tackled the request 00:05:25.240 --> 00:05:27.620 really methodically over about 50 minutes. It 00:05:27.620 --> 00:05:30.100 went to Google Flights, Kayak, found some decent 00:05:30.100 --> 00:05:32.500 non -stop options on Vietjet Air and Bamboo Airways. 00:05:32.639 --> 00:05:34.199 Price -wise, they looked okay. For hotels, it 00:05:34.199 --> 00:05:36.800 used Booking .com, a go -to, filtering for four 00:05:36.800 --> 00:05:39.540 -star pool near the beach. It even shortlisted 00:05:39.540 --> 00:05:41.819 a few, like the Salah Denang Beach Hotel. And 00:05:41.819 --> 00:05:44.089 yeah, it successfully found a motorbike... street 00:05:44.089 --> 00:05:46.329 food tour for the activity. Pretty cool. That's 00:05:46.329 --> 00:05:47.930 honestly impressive, getting all those pieces 00:05:47.930 --> 00:05:50.550 lined up, finding the options, checking criteria. 00:05:50.930 --> 00:05:53.610 But this is where we hit that snag, right? What 00:05:53.610 --> 00:05:55.850 the source is called the last mile problem. It 00:05:55.850 --> 00:05:59.970 got so far, but then. Exactly. Yes. After 50 00:05:59.970 --> 00:06:02.410 minutes of really impressive planning, the agent 00:06:02.410 --> 00:06:04.610 couldn't complete a single actual transaction. 00:06:05.050 --> 00:06:07.129 It got right up to the final payment screen for 00:06:07.129 --> 00:06:10.199 both the flights and the hotel. and then it just 00:06:10.199 --> 00:06:12.379 stopped. It needed passenger details, credit 00:06:12.379 --> 00:06:15.220 card info, sensitive stuff. It gave the link 00:06:15.220 --> 00:06:17.920 for the food tour like requested, but it just 00:06:17.920 --> 00:06:20.800 couldn't finalize anything requiring that sensitive 00:06:20.800 --> 00:06:22.720 data. It really hit that security wall, which 00:06:22.720 --> 00:06:25.120 is there for a reason, of course. So what's the 00:06:25.120 --> 00:06:27.480 main takeaway from that experiment then? What 00:06:27.480 --> 00:06:29.740 worked exceptionally well and where did it, you 00:06:29.740 --> 00:06:32.480 know, fall short? Okay, it excelled at a complex 00:06:32.480 --> 00:06:34.920 understanding, really grasping the multi -part 00:06:34.920 --> 00:06:37.420 request. Intelligent research too, comparing 00:06:37.420 --> 00:06:39.399 prices and reviews across different sites, that 00:06:39.399 --> 00:06:42.060 was good. Handling simultaneous tasks, problem 00:06:42.060 --> 00:06:44.920 solving. Like, it pivoted pretty smoothly when 00:06:44.920 --> 00:06:47.500 one booking site was slow. It truly acted like 00:06:47.500 --> 00:06:49.600 a tireless digital assistant for all the planning 00:06:49.600 --> 00:06:52.740 stuff. And its biggest limitation? The place 00:06:52.740 --> 00:06:56.209 it fell short. The last mile problem. It's that 00:06:56.209 --> 00:06:59.209 critical security safeguard, stopping it before 00:06:59.209 --> 00:07:01.569 payment. You know, I still kind of wrestle with 00:07:01.569 --> 00:07:03.490 the practical friction of that last mile problem 00:07:03.490 --> 00:07:06.170 myself. I mean, it's obviously a necessary hurdle 00:07:06.170 --> 00:07:09.009 for security, right? But it does mean it's not 00:07:09.009 --> 00:07:12.470 truly set it and forget it. Not yet. So basically, 00:07:12.569 --> 00:07:15.209 it's a phenomenal planner. a great research assistant, 00:07:15.730 --> 00:07:17.970 but not quite a fully autonomous booker. Is that 00:07:17.970 --> 00:07:20.610 fair? Exactly. Think of it as a co -pilot. Gets 00:07:20.610 --> 00:07:22.810 you maybe 90 % of the way there, but you still 00:07:22.810 --> 00:07:24.649 have to take the controls for landing. Okay, 00:07:24.649 --> 00:07:27.050 let's clarify the difference here. How do these 00:07:27.050 --> 00:07:30.389 new AI agents really compare to the traditional 00:07:30.389 --> 00:07:33.389 AI assistants we've used for years, like Siri 00:07:33.389 --> 00:07:37.009 or Classic Chat GPT? Oh, it's a fundamental paradigm 00:07:37.009 --> 00:07:40.240 shift, really. Traditional assistance, mostly 00:07:40.240 --> 00:07:42.540 for information retrieval, answering questions, 00:07:43.160 --> 00:07:46.300 usually stuck inside their own app. Agents, though, 00:07:46.439 --> 00:07:49.360 are about task execution. Getting things done. 00:07:49.759 --> 00:07:52.139 Goal completion across the open web. Traditional 00:07:52.139 --> 00:07:54.980 is usually single turn, mostly stateless. Doesn't 00:07:54.980 --> 00:07:56.920 remember much from one question to the next. 00:07:57.379 --> 00:07:59.800 Agents are multi -step, autonomous processes. 00:08:00.079 --> 00:08:02.319 They keep track of the state, the context throughout 00:08:02.319 --> 00:08:04.560 long tasks. They kind of remember what they're 00:08:04.560 --> 00:08:07.139 doing and why. That distinction is huge. It's 00:08:07.139 --> 00:08:09.939 a different category of tool. Knowing that, our 00:08:09.939 --> 00:08:12.279 sources decided to push it even further. Could 00:08:12.279 --> 00:08:14.720 you run multiple agents at the same time? They 00:08:14.720 --> 00:08:16.639 tested this with three separate tasks running 00:08:16.639 --> 00:08:18.920 in parallel. They did, yeah. They launched three 00:08:18.920 --> 00:08:21.800 agents with pretty different complex goals. One 00:08:21.800 --> 00:08:23.779 was the Da Nang weekend trip we just talked about. 00:08:24.120 --> 00:08:26.740 Another was creating a 10 -slide PowerPoint presentation 00:08:26.740 --> 00:08:29.680 on content marketing. And the third was analyzing 00:08:29.680 --> 00:08:32.440 a competitor's YouTube channel to pull data into 00:08:32.440 --> 00:08:35.279 a spreadsheet, all running simultaneously. OK. 00:08:35.379 --> 00:08:37.820 And what were the results of this multi -agent 00:08:37.820 --> 00:08:40.500 test? Was it a mixed bag? Did some tasks work 00:08:40.500 --> 00:08:42.639 better than others? It was a very mixed bag, 00:08:42.759 --> 00:08:45.889 yeah. And super revealing, the web -heavy Da 00:08:45.889 --> 00:08:48.330 Nang trip took the 50 minutes and like we said, 00:08:48.690 --> 00:08:50.769 still needed human help at the end. The creative 00:08:50.769 --> 00:08:53.450 task, the presentation, that took 41 minutes. 00:08:54.090 --> 00:08:55.789 But the design was kind of generic, the content 00:08:55.789 --> 00:08:57.950 pretty basic, needed a lot of human editing. 00:08:58.669 --> 00:09:01.350 But here's the really interesting bit. The data 00:09:01.350 --> 00:09:04.389 analysis task. Scraping YouTube data, making 00:09:04.389 --> 00:09:07.500 a spreadsheet. That took only four minutes. And 00:09:07.500 --> 00:09:10.500 it produced a perfectly formatted, accurate spreadsheet. 00:09:11.039 --> 00:09:13.799 Even had some insightful summaries. Whoa. I mean, 00:09:14.019 --> 00:09:16.019 just imagine a world where you could orchestrate 00:09:16.019 --> 00:09:18.919 like a dozen specialized agents, each one nailing 00:09:18.919 --> 00:09:20.860 their specific task, all running in parallel. 00:09:21.240 --> 00:09:23.320 The potential is kind of mind -bending. That 00:09:23.320 --> 00:09:25.500 really highlights. where these agents seem to 00:09:25.500 --> 00:09:27.159 shine right now, doesn't it? Structured data 00:09:27.159 --> 00:09:29.700 -driven tasks seem slower and less refined when 00:09:29.700 --> 00:09:32.240 it comes to maybe more creative work or really 00:09:32.240 --> 00:09:34.919 complex, nuanced web navigation. So agents are 00:09:34.919 --> 00:09:36.759 really at their best when the task involves a 00:09:36.759 --> 00:09:39.100 lot of structured data. Yes, they truly shine 00:09:39.100 --> 00:09:41.500 with data analysis and clear, objective tasks. 00:09:41.879 --> 00:09:43.940 Mid -roll sponsor, Readmarker sponsor content 00:09:43.940 --> 00:09:46.940 provided separately. OK, so beyond what we've 00:09:46.940 --> 00:09:49.480 dived into with chat GPT agents, the whole AI 00:09:49.480 --> 00:09:52.519 space is just exploding with innovation. It feels 00:09:52.519 --> 00:09:55.299 like every week. What other groundbreaking AI 00:09:55.299 --> 00:09:57.519 launches from just this past week should we know 00:09:57.519 --> 00:10:00.480 about? It's hard to keep up. It really has been 00:10:00.480 --> 00:10:03.139 an incredible week. So much happened. First, 00:10:03.340 --> 00:10:06.100 ChatGPT's record feature is now for everyone. 00:10:06.179 --> 00:10:08.840 Well, for plus users anyway. It was pro -exclusive. 00:10:08.940 --> 00:10:11.419 This lets you record any system audio on your 00:10:11.419 --> 00:10:14.080 Mac, like a Zoom call, a lecture, whatever. And 00:10:14.080 --> 00:10:16.539 it automatically generates a really detailed 00:10:16.539 --> 00:10:19.299 summary. Super powerful for meeting notes or 00:10:19.299 --> 00:10:22.120 repurposing content quickly. Then, Anthropix 00:10:22.120 --> 00:10:24.620 Claude is positioning itself as a hub. They launched 00:10:24.620 --> 00:10:26.460 a directory of tools that integrate directly 00:10:26.460 --> 00:10:28.360 with Claude, seamless connections with stuff 00:10:28.360 --> 00:10:30.720 like Asana, Canva, Gmail, Google Drive, even 00:10:30.720 --> 00:10:33.200 Stripe. Early testing showed some bugs apparently, 00:10:33.899 --> 00:10:35.980 but you gotta expect stability will improve fast. 00:10:36.159 --> 00:10:38.559 That idea of Claude becoming a central work hub, 00:10:39.480 --> 00:10:41.000 that feels like a significant step towards a 00:10:41.000 --> 00:10:43.840 more unified AI assistant. What else is making 00:10:43.840 --> 00:10:45.879 waves, maybe in the more personalized AI space? 00:10:46.139 --> 00:10:50.070 Okay, check this out. NVIDIA AI launched AI Twin. 00:10:50.549 --> 00:10:53.190 Version 4 .0 creates a digital avatar of you. 00:10:53.529 --> 00:10:55.549 You just record at least 60 seconds of yourself, 00:10:55.950 --> 00:10:58.610 give verbal permission, and within minutes, boom, 00:10:58.809 --> 00:11:01.250 you have a digital clone. Imagine, like a real 00:11:01.250 --> 00:11:02.970 estate agent writes a script for their weekly 00:11:02.970 --> 00:11:05.509 market update, pastes it in, and their digital 00:11:05.509 --> 00:11:08.289 clone presents it, flawlessly. It could turn 00:11:08.289 --> 00:11:10.450 what was maybe a half -day task into just 10 00:11:10.450 --> 00:11:14.220 minutes, and even more personal. Hume AI is cloning 00:11:14.220 --> 00:11:17.299 personality. Their EVI 3 model replicates not 00:11:17.299 --> 00:11:19.899 just your voice, but your actual speaking style. 00:11:20.100 --> 00:11:22.759 It analyzes like a 30 -90 second voice sample, 00:11:23.100 --> 00:11:25.139 learns your cadence, your filler words, your 00:11:25.139 --> 00:11:26.879 ahns and ahms, your conversational patterns. 00:11:27.399 --> 00:11:29.019 A podcaster maybe could create a digital version 00:11:29.019 --> 00:11:31.080 of themselves for interactive Q &As with fans, 00:11:31.399 --> 00:11:33.360 keeping their unique style. That's fascinating. 00:11:33.419 --> 00:11:35.080 Not just the voice, but the little mannerisms, 00:11:35.259 --> 00:11:37.779 the speech patterns. Wow. What about for more 00:11:37.779 --> 00:11:39.960 traditional creative fields, like filmmaking 00:11:39.960 --> 00:11:42.899 or audio work? Yeah, for filmmakers sound designers, 00:11:43.539 --> 00:11:45.860 Adobe Firefly now hears your voice and creates 00:11:45.860 --> 00:11:48.840 sound effects. This is genuinely kind of mind 00:11:48.840 --> 00:11:50.659 blowing. You can literally record yourself making 00:11:50.659 --> 00:11:53.860 a noise like just swoosh or flutter flutter and 00:11:53.860 --> 00:11:56.320 tell the AI what you want it to become. So a 00:11:56.320 --> 00:11:58.620 filmmaker sees a bird take flight, makes that 00:11:58.620 --> 00:12:01.559 sound, and Firefly turns it into a high fidelity, 00:12:02.019 --> 00:12:04.240 perfectly synced audio track of realistic wing 00:12:04.240 --> 00:12:06.950 beats. Crazy and we also saw other things popping 00:12:06.950 --> 00:12:09.830 up right like runways act 2 for motion capture 00:12:09.830 --> 00:12:12.590 animating characters with your movements Mirage 00:12:12.590 --> 00:12:14.789 will LSD for real -time video transformation 00:12:14.789 --> 00:12:18.529 turning your feet into visual art plus Grok AI 00:12:18.529 --> 00:12:21.149 is somewhat controversial Annie and Rudy companions 00:12:21.149 --> 00:12:23.710 which kind of point to this demand for more personalized 00:12:23.710 --> 00:12:26.490 AI even if some options are less filtered OK, 00:12:26.529 --> 00:12:28.250 so looking at all these, what's the common thread? 00:12:28.330 --> 00:12:30.909 What ties these diverse new tools together? I 00:12:30.909 --> 00:12:33.230 think they're all about automating and customizing 00:12:33.230 --> 00:12:35.629 creative or repetitive tasks. That seems to be 00:12:35.629 --> 00:12:39.250 the core focus. This just dizzying level of innovation, 00:12:39.250 --> 00:12:41.830 it also speaks to the intense competition heating 00:12:41.830 --> 00:12:44.490 up in the AI world, right? The talent wars, from 00:12:44.490 --> 00:12:46.730 what our sources indicate, sound absolutely real. 00:12:46.970 --> 00:12:49.129 Oh, they are. Our sources detailed this high 00:12:49.129 --> 00:12:52.090 -stakes saga in AI coding. really interesting. 00:12:52.690 --> 00:12:55.029 OpenAI was apparently in talks to acquire a company 00:12:55.029 --> 00:12:57.669 called Windsurf, a promising AI coding tool. 00:12:58.250 --> 00:13:01.429 But then, boom, its CEO and top talent abruptly 00:13:01.429 --> 00:13:04.190 left for Google DeepMind. But even after losing 00:13:04.190 --> 00:13:06.389 its leadership, Windsurf still got acquired, 00:13:06.629 --> 00:13:09.309 but by cognition, the company behind the Devon 00:13:09.309 --> 00:13:11.929 AI agent. This whole scenario just highlights 00:13:11.929 --> 00:13:15.610 how incredibly valuable elite AI talent has become 00:13:15.610 --> 00:13:18.120 in the fierce competition between these big players 00:13:18.120 --> 00:13:20.840 to own the future of software development, especially 00:13:20.840 --> 00:13:23.159 in this agent space. It's like a high stakes 00:13:23.159 --> 00:13:26.600 game of digital chess. OK, so for you, our listener 00:13:26.600 --> 00:13:29.019 listening all this, what does it actually mean? 00:13:29.179 --> 00:13:31.279 How do we effectively use these powerful new 00:13:31.279 --> 00:13:33.840 tools in our own lives or businesses without 00:13:33.840 --> 00:13:35.740 getting overwhelmed or making mistakes? Right. 00:13:35.919 --> 00:13:37.500 AI agents are definitely here. They're real. 00:13:37.840 --> 00:13:39.480 But it's crucial to understand they are not yet 00:13:39.480 --> 00:13:41.220 fully autonomous. Not really. They're powerful 00:13:41.220 --> 00:13:44.029 force multipliers. Absolutely. But they consistently 00:13:44.029 --> 00:13:47.169 require human strategy, human oversight. That, 00:13:47.250 --> 00:13:49.730 to me, is the core message our sources keep hammering 00:13:49.730 --> 00:13:52.009 home. Could you maybe offer some practical guidance? 00:13:52.590 --> 00:13:54.590 How to actually integrate these into our daily 00:13:54.590 --> 00:13:57.350 routines, whether it's for work or just personal 00:13:57.350 --> 00:14:00.960 life? Sure. OK, for business professionals. Think 00:14:00.960 --> 00:14:03.500 of and use these agents like tireless research 00:14:03.500 --> 00:14:06.100 interns. Have them gather huge amounts of data, 00:14:06.299 --> 00:14:08.639 compare vendors super efficiently, maybe create 00:14:08.639 --> 00:14:11.519 initial drafts of reports, delegate routine data 00:14:11.519 --> 00:14:14.240 entry. But, and this is the critical part, never 00:14:14.240 --> 00:14:17.100 allow an agent to make a final unsupervised decision 00:14:17.100 --> 00:14:20.419 on important business matters or especially financial 00:14:20.419 --> 00:14:23.399 transactions. Always, always review its work 00:14:23.399 --> 00:14:25.440 with your own judgment for personal productivity. 00:14:26.000 --> 00:14:27.700 Yeah, let an agent plan your vacation outline, 00:14:27.860 --> 00:14:29.899 find recipes that fit criteria, create detailed 00:14:29.929 --> 00:14:32.289 shopping list, it'll save you countless hours 00:14:32.289 --> 00:14:34.450 of just tedious drudgery freeing you up to make 00:14:34.450 --> 00:14:36.750 that final 10 % of decisions requiring your personal 00:14:36.750 --> 00:14:38.669 taste, your judgment. Just be really vigilant 00:14:38.669 --> 00:14:41.399 about your data. Use unique passwords. Be present 00:14:41.399 --> 00:14:43.320 for any stuff that needs personal or financial 00:14:43.320 --> 00:14:45.360 info. Don't just let it run wild with that stuff. 00:14:45.399 --> 00:14:47.559 And for content creators, tools like that in 00:14:47.559 --> 00:14:49.980 Video AI Twin or Adobe Sound Effect Generator, 00:14:50.000 --> 00:14:51.899 they can dramatically speed up your production 00:14:51.899 --> 00:14:54.159 workflow. Use real -time video effects for unique 00:14:54.159 --> 00:14:57.159 live content, maybe. But always, always treat 00:14:57.159 --> 00:14:59.080 AI -generated content, text, scripts, images, 00:14:59.120 --> 00:15:01.519 whatever, as a first draft. You have to infuse 00:15:01.519 --> 00:15:03.980 it with your unique voice, your style, your perspective. 00:15:04.360 --> 00:15:07.639 That human touch is still totally irreplaceable. 00:15:07.840 --> 00:15:10.519 And we also saw a brief m - of things like Google's 00:15:10.519 --> 00:15:13.440 AI business caller, where AI can call local businesses 00:15:13.440 --> 00:15:16.879 for you, China's Kimi K2 model ranking high globally, 00:15:17.559 --> 00:15:19.399 specialized financial AI tools from Anthropic 00:15:19.399 --> 00:15:23.000 and Mistral, Amazon's Cura IDE for coding, planning 00:15:23.000 --> 00:15:26.000 project architecture first. It's just clear that 00:15:26.000 --> 00:15:28.799 innovation is bursting out everywhere, in every 00:15:28.799 --> 00:15:31.759 sector. So the bottom line message seems to be... 00:15:31.519 --> 00:15:34.000 Leverage AI's capabilities, definitely use them, 00:15:34.320 --> 00:15:36.279 but always keep a human in the loop for the critical 00:15:36.279 --> 00:15:38.100 thinking and final decisions. Absolutely. They 00:15:38.100 --> 00:15:39.860 are co -pilots. They are not replacements for 00:15:39.860 --> 00:15:42.279 critical thinking. Not yet, anyway. So the big 00:15:42.279 --> 00:15:44.460 idea here, pulling it all together, it feels 00:15:44.460 --> 00:15:47.320 like ChatGPT's agent feature and really all these 00:15:47.320 --> 00:15:50.580 new AI advancements signal a fundamental change, 00:15:51.019 --> 00:15:52.500 a change in our relationship with technology. 00:15:52.600 --> 00:15:55.159 We're moving from just being users to becoming 00:15:55.159 --> 00:15:57.779 managers, maybe even skilled orchestrators of 00:15:57.779 --> 00:16:00.269 AI. Yeah, and what's truly fascinating now is 00:16:00.269 --> 00:16:02.210 thinking about what's coming next, say in the 00:16:02.210 --> 00:16:04.909 next six to 12 months. Expect big increases in 00:16:04.909 --> 00:16:07.490 speed, reliability, that's almost a given, and 00:16:07.490 --> 00:16:10.549 a rapid move towards true multimodality or vision. 00:16:11.090 --> 00:16:13.409 Agents being able to interpret entire page layouts, 00:16:13.850 --> 00:16:16.009 recognize icons, maybe even learn by watching 00:16:16.009 --> 00:16:18.649 video tutorials, will probably also see more 00:16:18.649 --> 00:16:20.990 advanced long -term memory and deep personalization. 00:16:21.350 --> 00:16:23.529 Agents learning your preferences over time becoming 00:16:23.529 --> 00:16:26.250 proactive, maybe even anticipating needs before 00:16:26.250 --> 00:16:28.779 you stake them. And certainly, expect the rise 00:16:28.779 --> 00:16:31.000 of highly specialized agents, agents for specific 00:16:31.000 --> 00:16:33.240 jobs like legal research or marketing campaign 00:16:33.240 --> 00:16:35.299 creation, that could fundamentally change how 00:16:35.299 --> 00:16:39.179 work gets done. While that dream of a fully autonomous 00:16:39.179 --> 00:16:41.840 AI handling absolutely everything, it isn't quite 00:16:41.840 --> 00:16:43.840 reality yet. But the progress we're seeing now 00:16:43.840 --> 00:16:46.399 is just staggering. These tools, even as they 00:16:46.399 --> 00:16:48.799 are, are already capable of absorbing a huge 00:16:48.799 --> 00:16:50.879 chunk of the tedious time -consuming work that 00:16:50.879 --> 00:16:53.539 fills up our days. It seems like the most valuable 00:16:53.539 --> 00:16:56.159 skill in the coming years might actually be AI 00:16:56.159 --> 00:16:58.409 orchestration. you know, the ability to effectively 00:16:58.409 --> 00:17:01.669 define goals, delegate complex tasks to a team 00:17:01.669 --> 00:17:04.170 of specialized AI agents, and provide that critical 00:17:04.170 --> 00:17:06.630 human oversight needed for quality, accuracy, 00:17:06.750 --> 00:17:10.650 and security. So, my advice, just start experimenting 00:17:10.650 --> 00:17:13.250 now. Give an agent a small, low -stakes task. 00:17:13.529 --> 00:17:15.650 See what happens. Learn its strengths, its weaknesses, 00:17:15.769 --> 00:17:17.670 its quirks. The future, I think, really belongs 00:17:17.670 --> 00:17:19.910 to those who don't just use AI, but learn how 00:17:19.910 --> 00:17:22.369 to lead it. Thank you for joining us on this 00:17:22.369 --> 00:17:23.710 deep dive out true music.