WEBVTT 00:00:00.000 --> 00:00:02.500 Imagine an AI that doesn't just tell you things, 00:00:02.740 --> 00:00:05.679 but actually does things for you, searching, 00:00:06.179 --> 00:00:10.039 buying, maybe even crafting content, all without 00:00:10.039 --> 00:00:13.199 you really lifting a finger. What if that future 00:00:13.199 --> 00:00:16.559 is, well, kind of already here? Welcome to the 00:00:16.559 --> 00:00:18.940 deep dive. Today, we're really digging into a 00:00:18.940 --> 00:00:21.679 concept that's rapidly reshaping our digital 00:00:21.679 --> 00:00:24.160 lives, AI agents. That's right. We're moving 00:00:24.160 --> 00:00:27.559 beyond AI is just this brilliant. brain in a 00:00:27.559 --> 00:00:30.079 box, you know. These agents are becoming our 00:00:30.079 --> 00:00:32.079 digital hands and eyes. They're interacting with 00:00:32.079 --> 00:00:34.060 the world for us. Yeah, and our mission today 00:00:34.060 --> 00:00:36.820 is to explore what these AI agents actually are, 00:00:36.979 --> 00:00:39.500 why there's such a potential game changer, and 00:00:39.500 --> 00:00:41.320 importantly how you can start getting a handle 00:00:41.320 --> 00:00:43.240 on them right now. We'll look at some real -world 00:00:43.240 --> 00:00:45.880 uses, understand this new idea of agent -engine 00:00:45.880 --> 00:00:48.740 optimization. VEO, yeah. And crucially, talk 00:00:48.740 --> 00:00:51.380 about how to navigate the risk responsibly. You'll 00:00:51.380 --> 00:00:53.420 get some practical insights, hopefully a clear 00:00:53.420 --> 00:00:56.710 roadmap for this evolving tech. Okay, let's untack 00:00:56.710 --> 00:00:59.710 this. For a long time, AI models, you know, like 00:00:59.710 --> 00:01:02.590 the early versions of Chach -PT, they were brilliant 00:01:02.590 --> 00:01:04.189 conversationalists. You'd ask a question, it 00:01:04.189 --> 00:01:06.549 gives you an answer. But it was always sort of 00:01:06.549 --> 00:01:09.150 contained, right? A mind without a body, almost. 00:01:09.329 --> 00:01:12.090 Exactly. And the core idea with AI agents is 00:01:12.090 --> 00:01:14.709 this really fundamental shift. We're going from 00:01:14.709 --> 00:01:18.090 responsive AI to proactive AI. An agent doesn't 00:01:18.090 --> 00:01:20.819 just answer, It acts. It's an autonomous system. 00:01:21.159 --> 00:01:22.939 It perceives its environment. It makes plans. 00:01:23.140 --> 00:01:25.560 It makes decisions. And then it executes actions 00:01:25.560 --> 00:01:28.659 to achieve specific goals you give it. Think 00:01:28.659 --> 00:01:32.090 of them like digital assistants. but with virtual 00:01:32.090 --> 00:01:34.489 hands and eyes. So it's like we've handed them 00:01:34.489 --> 00:01:36.769 a mouse and keyboard in a virtual world. The 00:01:36.769 --> 00:01:39.030 big breakthrough really came when platforms like 00:01:39.030 --> 00:01:42.390 OpenAI implemented these agent modes. This lets 00:01:42.390 --> 00:01:44.890 the AI actually control a virtual computer. It 00:01:44.890 --> 00:01:47.090 can browse websites, fill out forms, click buttons, 00:01:47.390 --> 00:01:49.590 just like a person would. It's kind of wild to 00:01:49.590 --> 00:01:52.730 watch. It really is. And this matters, like... 00:01:52.439 --> 00:01:55.659 a lot. First, you get this exponential efficiency. 00:01:55.900 --> 00:01:58.459 Agents can work 247 on repetitive stuff. Then 00:01:58.459 --> 00:02:00.480 there's new accessibility. They can reach information 00:02:00.480 --> 00:02:04.000 behind login pages or inside proprietary databases, 00:02:04.599 --> 00:02:06.760 stuff that traditional APIs, that's application 00:02:06.760 --> 00:02:08.639 programming interfaces, basically how software 00:02:08.639 --> 00:02:11.759 talks, just can't get to. We're seeing really 00:02:11.759 --> 00:02:14.639 deep automation now, not just single tasks, but 00:02:14.639 --> 00:02:17.460 entire multi -step workflows. And early adopters, 00:02:17.639 --> 00:02:20.719 they're gaining unparalleled speed, scalability, 00:02:20.800 --> 00:02:22.900 and innovation. We're already seeing agents ordering 00:02:22.900 --> 00:02:25.699 goods, sending LinkedIn messages, even doing 00:02:25.699 --> 00:02:28.039 bulk video creation and deep market research. 00:02:28.259 --> 00:02:30.259 OK, so boiled down, what's the biggest difference 00:02:30.259 --> 00:02:32.270 then? between an agent and just a regular chat 00:02:32.270 --> 00:02:34.909 bot. Chat bots answer questions. Agents take 00:02:34.909 --> 00:02:37.050 action to achieve goals. That's clear. Action 00:02:37.050 --> 00:02:39.389 versus information. OK. One of the best ways 00:02:39.389 --> 00:02:41.229 to see this in action, maybe, is with something 00:02:41.229 --> 00:02:44.129 familiar, like online shopping. Imagine an AI 00:02:44.129 --> 00:02:46.349 agent handling that whole multi -step process 00:02:46.349 --> 00:02:48.770 for you. Yeah, and it's surprisingly straightforward 00:02:48.770 --> 00:02:52.629 to set up. But the precision is absolutely key. 00:02:52.750 --> 00:02:55.129 First, you activate agent mode. Sometimes it's 00:02:55.129 --> 00:02:57.270 called browse mode in whatever AI platform you're 00:02:57.270 --> 00:03:00.969 using. Then the really critical part. You have 00:03:00.969 --> 00:03:03.050 to give it extremely specific instructions. Don't 00:03:03.050 --> 00:03:05.189 just say, buy shoes. You need something like, 00:03:05.569 --> 00:03:08.009 search the Adidas Vietnam website for men's running 00:03:08.009 --> 00:03:11.789 shoes, black, size 42, ultra boost line, under 00:03:11.789 --> 00:03:15.370 3 ,500 ,000 VND. Add that to the cart and proceed 00:03:15.370 --> 00:03:18.490 to checkout. Really specific. And then you just 00:03:18.490 --> 00:03:20.490 watch it happen. Pretty much. You'll literally 00:03:20.490 --> 00:03:22.689 observe the virtual computer screen. You see 00:03:22.689 --> 00:03:25.189 the mouse cursor moving, text being typed, websites 00:03:25.189 --> 00:03:27.349 loading and being navigated. The agent might 00:03:27.349 --> 00:03:29.349 even ask you for more details, you know, like 00:03:29.349 --> 00:03:31.210 which version of UltraBoost, the Lite or the 00:03:31.210 --> 00:03:34.409 1 .0, that interaction is actually crucial. Watching 00:03:34.409 --> 00:03:36.389 it helps you figure out how to make your future 00:03:36.389 --> 00:03:38.889 prompts even better. Right. Now safety and security 00:03:38.889 --> 00:03:40.889 are obviously huge here. We're talking about 00:03:40.889 --> 00:03:43.090 potentially spending money. What's the absolute 00:03:43.090 --> 00:03:45.509 number one rule for payments? Okay, this is vital. 00:03:46.090 --> 00:03:48.409 Never share your full credit card details for 00:03:48.409 --> 00:03:51.629 autofill. Just don't do it. It's way too risky. 00:03:52.370 --> 00:03:54.849 Safer methods involve, say, instructing the agent 00:03:54.849 --> 00:03:57.689 to generate a QR code for you to scan and approve, 00:03:58.330 --> 00:04:00.509 or maybe using services like PayPal, where you 00:04:00.509 --> 00:04:03.409 do the final manual authentication step yourself, 00:04:03.729 --> 00:04:05.930 and always review the final order summary before 00:04:05.930 --> 00:04:08.330 you approve anything irreversible. Keep that 00:04:08.330 --> 00:04:10.409 control. I can definitely see some challenges 00:04:10.409 --> 00:04:13.409 cropping up. What are the common roadblocks people 00:04:13.409 --> 00:04:16.370 run into? Well, CAPTCHA barriers are one. Agents 00:04:16.370 --> 00:04:18.589 generally respect those, so they'll ask for human 00:04:18.589 --> 00:04:21.069 help. Website compatibility can sometimes be 00:04:21.069 --> 00:04:23.350 an issue, especially with really complex JavaScript 00:04:23.350 --> 00:04:26.029 -heavy sites. But honestly, the biggest hurdle, 00:04:26.189 --> 00:04:28.990 9 times out of 10, is prompt precision. Vague 00:04:28.990 --> 00:04:30.970 prompts just lead to unpredictable, sometimes 00:04:30.970 --> 00:04:33.509 weird results. You absolutely need to be highly 00:04:33.509 --> 00:04:36.290 detailed. OK, so if there's one core takeaway 00:04:36.290 --> 00:04:39.029 for making purchasing safe with agents, what 00:04:39.029 --> 00:04:41.649 is it? Keep financial control yourself. Let agents 00:04:41.649 --> 00:04:44.420 do the searching. Got it. OK, beyond shopping, 00:04:44.779 --> 00:04:47.660 these agents seem incredibly powerful for, say, 00:04:48.120 --> 00:04:50.259 streamlining professional outreach, especially 00:04:50.259 --> 00:04:52.420 on platforms like LinkedIn, expanding networks, 00:04:52.660 --> 00:04:55.639 finding clients. But you've got to do it responsibly, 00:04:55.800 --> 00:04:58.600 right? Oh, absolutely. Rule number one, non -negotiable. 00:04:59.139 --> 00:05:01.519 Use a test account first. Seriously. You don't 00:05:01.519 --> 00:05:03.699 want to mess up your actual professional reputation 00:05:03.699 --> 00:05:06.040 while you're figuring things out. Then define 00:05:06.040 --> 00:05:09.009 your target audience with laser precision. No 00:05:09.009 --> 00:05:11.490 vague stuff. Like, find marketing managers at 00:05:11.490 --> 00:05:14.129 software development companies in Vietnam, 201, 00:05:14.269 --> 00:05:17.230 500 employees, using LinkedIn Sales Navigator. 00:05:17.490 --> 00:05:19.550 That level of detail. That makes total sense. 00:05:19.649 --> 00:05:21.569 It's like giving it very specific coordinates. 00:05:21.810 --> 00:05:23.790 And personalization is key, too, I imagine. Not 00:05:23.790 --> 00:05:25.829 just blasting templates. Mystically. You don't 00:05:25.829 --> 00:05:28.389 want to sound like a robot. Prompt the agent 00:05:28.389 --> 00:05:30.949 to maybe visit their recent activity feed first. 00:05:31.410 --> 00:05:33.870 Find something specific, like compliment their 00:05:33.870 --> 00:05:36.649 recent article on Axe or mention their company's 00:05:36.649 --> 00:05:39.379 latest funding round. Then introduce your service 00:05:39.379 --> 00:05:42.379 like a conversion rate analysis offer and always 00:05:42.379 --> 00:05:46.160 always set clear limits tell it explicitly Execute 00:05:46.160 --> 00:05:48.699 this process for exactly 10 people and then stop 00:05:48.699 --> 00:05:51.319 report the results back to me in a table Then 00:05:51.319 --> 00:05:53.720 you watch, see what works, and tweak your messages 00:05:53.720 --> 00:05:55.779 based on the response rates. You know, I still 00:05:55.779 --> 00:05:57.879 wrestle with prompt drift myself sometimes. It's 00:05:57.879 --> 00:05:59.699 funny, I was trying to get an agent to draft 00:05:59.699 --> 00:06:02.519 a nuanced email last week and it just kept defaulting 00:06:02.519 --> 00:06:05.259 to this really like overly formal, almost stiff 00:06:05.259 --> 00:06:07.600 tone. No matter how I tweaked it, it's a real 00:06:07.600 --> 00:06:10.120 challenge. You realize pretty quickly how a slightly 00:06:10.120 --> 00:06:12.439 off prompt can just send the whole thing sideways. 00:06:12.779 --> 00:06:15.120 That's a great point and a perfect example. A 00:06:15.120 --> 00:06:17.230 poorly written prompt especially for outreach 00:06:17.230 --> 00:06:20.629 like that, can have genuinely disastrous consequences 00:06:20.629 --> 00:06:23.290 for your brand image. It's not just about telling 00:06:23.290 --> 00:06:25.990 it what to do, but also what not to do. Always 00:06:25.990 --> 00:06:28.110 include stopping conditions, like we said, but 00:06:28.110 --> 00:06:31.250 also negative constraints. Things like, do not 00:06:31.250 --> 00:06:34.490 contact anyone at company X, Y, or Z. That gives 00:06:34.490 --> 00:06:36.910 you much tighter control, prevents embarrassing 00:06:36.910 --> 00:06:39.829 mistakes. So thinking about social media automation 00:06:39.829 --> 00:06:42.850 then, what's the golden rule? Test on a separate 00:06:42.850 --> 00:06:45.639 account. use extremely precise instructions. 00:06:45.920 --> 00:06:47.839 Right. So with all these agents out there interacting 00:06:47.839 --> 00:06:49.879 with the web, this has got to change how we think 00:06:49.879 --> 00:06:52.339 about building websites too. We're seeing the 00:06:52.339 --> 00:06:54.620 start of something called Agent Engine Optimization, 00:06:54.720 --> 00:06:58.180 AEO. It's like SEO. But optimizing your site 00:06:58.180 --> 00:07:00.600 specifically for these AIIs, it sounds like a 00:07:00.600 --> 00:07:03.170 totally different game. It really is. Agents 00:07:03.170 --> 00:07:06.449 search based on logic and efficiency, not really 00:07:06.449 --> 00:07:08.189 aesthetics. They make very literal decisions. 00:07:08.750 --> 00:07:10.750 If your meta description says, step -by -step 00:07:10.750 --> 00:07:13.810 guide to XYZ, an agent will probably prioritize 00:07:13.810 --> 00:07:16.129 that over something vague like, discover the 00:07:16.129 --> 00:07:19.129 power of XYZ, the link choices. They look for 00:07:19.129 --> 00:07:21.930 clear anchor text, clean URLs, technical signals, 00:07:22.550 --> 00:07:25.189 not pretty buttons. And here's a huge one. Agents 00:07:25.189 --> 00:07:27.779 have an absolute priority on speed. a slow site, 00:07:27.899 --> 00:07:30.139 they'll just abandon it immediately. No patience. 00:07:30.480 --> 00:07:32.579 That's a stark difference from how humans browse, 00:07:32.860 --> 00:07:34.800 yeah. We might wait a bit longer for a nice design, 00:07:34.860 --> 00:07:36.439 so you can even sort of command agents through 00:07:36.439 --> 00:07:38.920 the website itself. Yeah, through meta tags, 00:07:39.079 --> 00:07:40.740 you can give them direct instructions, like you 00:07:40.740 --> 00:07:42.779 could have a tag that says, extract the table 00:07:42.779 --> 00:07:46.970 with the ID, hashtag data table 2025. So for 00:07:46.970 --> 00:07:49.750 optimizing, you need agent -friendly meta descriptions, 00:07:50.449 --> 00:07:52.970 direct action -oriented language. Websites probably 00:07:52.970 --> 00:07:55.209 need to optimize more for small screens because 00:07:55.209 --> 00:07:57.629 agents often run in these low -resolution virtual 00:07:57.629 --> 00:08:00.149 browsers. Loading speed, like I said, isn't just 00:08:00.149 --> 00:08:02.350 a ranking factor anymore. It's like table stakes. 00:08:02.509 --> 00:08:05.449 It's a basic requirement. Use large, clear buttons, 00:08:05.750 --> 00:08:08.420 too. Agents navigate partly by coordinates, so 00:08:08.420 --> 00:08:11.459 clear element IDs and good spacing are key. And 00:08:11.459 --> 00:08:13.600 implementing a reading mode, like a text -only 00:08:13.600 --> 00:08:16.160 version, can really help agents grab information 00:08:16.160 --> 00:08:18.319 faster. And what about personal branding and 00:08:18.319 --> 00:08:20.300 all this? Agents aren't just finding websites, 00:08:20.459 --> 00:08:22.279 they're finding information about people, too. 00:08:22.620 --> 00:08:25.000 Definitely. That's where personal SEO comes in. 00:08:25.100 --> 00:08:27.560 You need to optimize your LinkedIn profile, for 00:08:27.560 --> 00:08:30.319 instance. Keywords in your headline, your summary, 00:08:30.540 --> 00:08:32.860 your experience descriptions make your contact 00:08:32.860 --> 00:08:35.860 info easily accessible. Agents are designed to 00:08:35.860 --> 00:08:38.440 find phone numbers, emails, build your domain 00:08:38.440 --> 00:08:40.679 authority by creating content that really showcases 00:08:40.679 --> 00:08:43.500 your expertise, and get listed in professional 00:08:43.500 --> 00:08:46.139 directories specific to your industry. These 00:08:46.139 --> 00:08:48.399 are all signals that agents will pick up on when 00:08:48.399 --> 00:08:50.399 someone asks them to find an expert like you. 00:08:50.669 --> 00:08:53.409 This shift to AEO, it sounds like it could really 00:08:53.409 --> 00:08:56.269 overhaul web design philosophy. What do you see 00:08:56.269 --> 00:08:58.429 as the biggest change for how websites will need 00:08:58.429 --> 00:09:01.610 to be built going forward? Speed and clear literal 00:09:01.610 --> 00:09:04.309 structural elements will become paramount. Okay, 00:09:04.669 --> 00:09:07.090 let's shift gears to automated content creation. 00:09:07.590 --> 00:09:09.570 We're moving way beyond just, you know, getting 00:09:09.570 --> 00:09:12.490 help writing a blog post. Agents can apparently 00:09:12.490 --> 00:09:15.490 manage complex content workflows now, like creating 00:09:15.490 --> 00:09:17.870 entire daily news videos automatically. Yeah, 00:09:17.950 --> 00:09:19.769 it's quite a leap, isn't it? You can define the 00:09:19.769 --> 00:09:21.629 parameters, maybe something like, create a daily 00:09:21.629 --> 00:09:24.190 two -minute video summarizing the top three tech 00:09:24.190 --> 00:09:26.769 news stories from TechCrunch and The Verge. Make 00:09:26.769 --> 00:09:29.090 sure it has subtitles. The agent then handles 00:09:29.090 --> 00:09:31.450 the news aggregation part, scanning the sites, 00:09:31.950 --> 00:09:33.870 identifying the top stories based on whatever 00:09:33.870 --> 00:09:37.309 criteria you set. Then a large language model, 00:09:37.490 --> 00:09:40.450 an LLM, generates a concise script for the video. 00:09:40.620 --> 00:09:43.500 After that, the agent uses various tools, maybe 00:09:43.500 --> 00:09:46.360 some Python libraries like MoviePy or APIs for 00:09:46.360 --> 00:09:49.299 stock images like Pexels or Unsplash, to combine 00:09:49.299 --> 00:09:51.980 an AI voiceover with relevant visuals. And finally, 00:09:52.080 --> 00:09:54.500 the video gets compiled and can be automatically 00:09:54.500 --> 00:09:57.620 uploaded to, say, Google Drive or even a test 00:09:57.620 --> 00:09:59.639 YouTube channel. What's really fascinating there, 00:09:59.679 --> 00:10:01.259 I think, is that the agents aren't just using 00:10:01.259 --> 00:10:03.500 existing software in a fixed way. They're kind 00:10:03.500 --> 00:10:05.600 of creating bits of software on the fly. Like 00:10:05.600 --> 00:10:07.360 you said, they might write a short Python script 00:10:07.360 --> 00:10:09.820 just to generate that one video, run it, and 00:10:09.820 --> 00:10:13.320 then just discard the script. Whoa. OK, imagine 00:10:13.320 --> 00:10:16.580 scaling that. Like, a billion different queries 00:10:16.580 --> 00:10:19.659 coming in, and each one gets its own tiny bespoke 00:10:19.659 --> 00:10:21.919 piece of software created just for that task. 00:10:22.440 --> 00:10:25.649 Temporary, perfectly tailored. That's incredibly 00:10:25.649 --> 00:10:27.389 efficient when you think about it. It really 00:10:27.389 --> 00:10:29.610 changes the whole dynamic. Now, we should be 00:10:29.610 --> 00:10:32.250 realistic, right? The quality probably isn't 00:10:32.250 --> 00:10:34.429 going to win any Oscars just yet. The output 00:10:34.429 --> 00:10:37.169 we see now is often pretty basic, maybe more 00:10:37.169 --> 00:10:39.850 like animated slide shows than professional video 00:10:39.850 --> 00:10:42.190 productions. That's true. The quality level isn't 00:10:42.190 --> 00:10:45.070 always top -tier cinema. But the real value proposition 00:10:45.070 --> 00:10:47.710 here isn't necessarily artistic perfection, at 00:10:47.710 --> 00:10:50.169 least not yet. It's the sheer speed and scale. 00:10:50.399 --> 00:10:53.419 Imagine getting dozens of different content variations 00:10:53.419 --> 00:10:56.080 for A -B testing in just minutes, or being able 00:10:56.080 --> 00:10:58.539 to produce consistent daily content streams with 00:10:58.539 --> 00:11:01.340 basically zero ongoing human effort once it's 00:11:01.340 --> 00:11:04.159 set up. It allows for incredibly rapid prototyping 00:11:04.159 --> 00:11:06.779 of content ideas before you invest heavily. So 00:11:06.779 --> 00:11:09.100 based on that, are these AI agents really coming 00:11:09.100 --> 00:11:11.240 for creative jobs right now? Not yet, really. 00:11:11.360 --> 00:11:14.000 They create solid foundations, but human refinement 00:11:14.000 --> 00:11:16.580 adds the crucial value. OK, let's talk research 00:11:16.580 --> 00:11:19.639 because this feels like an area where AI agents 00:11:19.639 --> 00:11:22.419 can really supercharge things. They seem able 00:11:22.419 --> 00:11:26.259 to break through information barriers that, you 00:11:26.259 --> 00:11:28.279 know, a regular Google search just hits a wall 00:11:28.279 --> 00:11:30.600 against. This is the huge advantage, definitely. 00:11:30.960 --> 00:11:33.240 Agents can navigate sites that require logins, 00:11:33.460 --> 00:11:35.460 assuming you give them secure test account credentials, 00:11:35.500 --> 00:11:38.500 of course, think accessing platforms like Upwork 00:11:38.500 --> 00:11:41.159 or academic databases, maybe private forums. 00:11:41.200 --> 00:11:43.860 They can gather real marketplace data directly. 00:11:44.120 --> 00:11:46.679 Imagine automatically collecting the average 00:11:46.679 --> 00:11:48.820 hourly rate being asked for Lang chain development 00:11:48.820 --> 00:11:51.419 jobs on various freelancer sites. That's powerful 00:11:51.419 --> 00:11:54.419 data. And crucially, they really excel at cross 00:11:54.419 --> 00:11:56.419 -referencing information from multiple sources. 00:11:56.740 --> 00:11:58.679 They can compile these comprehensive reports, 00:11:59.100 --> 00:12:01.539 looking for patterns, maybe discrepancies, across 00:12:01.539 --> 00:12:04.200 websites, forums, social media, all synthesized 00:12:04.200 --> 00:12:06.250 together. And we've actually seen this used in 00:12:06.250 --> 00:12:09.370 practice. I read about a study where agents analyze 00:12:09.370 --> 00:12:12.129 the AI skills market on freelancer platforms. 00:12:12.490 --> 00:12:15.070 And they found that actual client needs were 00:12:15.070 --> 00:12:17.950 often much simpler, like basic API integrations, 00:12:18.370 --> 00:12:20.549 than what the really technical online communities 00:12:20.549 --> 00:12:23.009 were usually discussing. That insight is gold 00:12:23.009 --> 00:12:26.870 for business strategy. Exactly. So to optimize 00:12:26.870 --> 00:12:29.159 using agents for research, You have to define 00:12:29.159 --> 00:12:31.700 your objectives very, very specifically. Don't 00:12:31.700 --> 00:12:33.480 just say research the real estate market. It 00:12:33.480 --> 00:12:35.840 needs to be like, find the average selling price 00:12:35.840 --> 00:12:37.940 per square meter for two -bedroom apartments 00:12:37.940 --> 00:12:41.120 in Thu Duc City, Ho Chi Minh City, sourced only 00:12:41.120 --> 00:12:44.259 from batdongsan .com .vn, and categorize the 00:12:44.259 --> 00:12:47.110 results by specific wards. You need to identify 00:12:47.110 --> 00:12:49.789 the exact target platforms, set clear parameters 00:12:49.789 --> 00:12:51.629 for what data to extract and how to organize 00:12:51.629 --> 00:12:54.330 it. Maybe export to a CSV file with these named 00:12:54.330 --> 00:12:56.649 columns. Plan for human intervention points, 00:12:56.850 --> 00:12:59.149 too, for things like complex captchas or decisions 00:12:59.149 --> 00:13:01.269 needing judgment. And always have a way to verify 00:13:01.269 --> 00:13:03.789 the information. Maybe cross -checking key data 00:13:03.789 --> 00:13:05.990 points. So thinking about those data silos we 00:13:05.990 --> 00:13:07.870 often end up in, how can agents help break those 00:13:07.870 --> 00:13:10.629 down? They can log into multiple sources and 00:13:10.629 --> 00:13:13.029 synthesize data from across the web. Okay, moving 00:13:13.029 --> 00:13:15.960 from gathering data to presenting it. Agents 00:13:15.960 --> 00:13:17.879 are also streamlining the creation of things 00:13:17.879 --> 00:13:21.039 like presentations and documents. No more staring 00:13:21.039 --> 00:13:23.700 at that intimidating blank slide deck, perhaps? 00:13:23.919 --> 00:13:26.639 Hopefully less of that. Yeah, automated deck 00:13:26.639 --> 00:13:29.039 generation is becoming a thing. It usually starts 00:13:29.039 --> 00:13:31.399 with topic research. You give it a topic, maybe 00:13:31.399 --> 00:13:33.559 the impact of quantum computing on cybersecurity. 00:13:34.159 --> 00:13:36.940 And the agent goes off and scans relevant sources. 00:13:37.529 --> 00:13:40.149 RCIF papers, tech journals, whatever you specify. 00:13:40.750 --> 00:13:43.289 Then it handles the content synthesis part. It 00:13:43.289 --> 00:13:45.570 combines the text, pulls out key data points, 00:13:46.190 --> 00:13:48.350 structures them into coherent slides, generating 00:13:48.350 --> 00:13:51.330 titles, bullet points, even draft speaker notes. 00:13:51.870 --> 00:13:53.909 It can even try to incorporate relevant images 00:13:53.909 --> 00:13:56.169 or basic charts for some visual enhancement. 00:13:56.549 --> 00:13:59.629 And it tries to ensure a logical structure, intro, 00:13:59.909 --> 00:14:02.950 body, conclusion. That sounds incredibly useful 00:14:02.950 --> 00:14:05.169 for just getting that first draft down, which 00:14:05.169 --> 00:14:07.029 is often the hardest part. So it's probably not 00:14:07.029 --> 00:14:08.629 going to look like something a top graphic designer 00:14:08.629 --> 00:14:11.269 made, but the speed and comprehensiveness must 00:14:11.269 --> 00:14:14.250 be a huge advantage. Precisely. While the design 00:14:14.250 --> 00:14:16.970 might be basic, it excels at speed you can get 00:14:16.970 --> 00:14:20.090 that first draft in minutes, not hours. It offers 00:14:20.090 --> 00:14:22.629 comprehensiveness too, often pulling in perspectives 00:14:22.629 --> 00:14:25.070 or data points you might have overlooked. And 00:14:25.070 --> 00:14:27.230 it ensures consistent formatting throughout. 00:14:27.360 --> 00:14:30.279 which is nice. The best approach, really, is 00:14:30.279 --> 00:14:32.879 to use this AI -generated content as a solid 00:14:32.879 --> 00:14:35.559 foundation, a starting point. Then you come in 00:14:35.559 --> 00:14:37.600 with your human insights, refine the narrative, 00:14:37.799 --> 00:14:41.019 and use proper design tools like Canva or PowerPoint 00:14:41.019 --> 00:14:44.360 to make it look polished. So for presentations 00:14:44.360 --> 00:14:46.440 and docs, it's really about getting speed over 00:14:46.440 --> 00:14:49.240 initial perfection, right? Exactly. Agents give 00:14:49.240 --> 00:14:51.559 you a comprehensive head start, saving valuable 00:14:51.559 --> 00:14:53.899 human time for refinement. This is all happening 00:14:53.899 --> 00:14:56.039 now, but the next frontier seems even closer 00:14:56.039 --> 00:14:58.279 to home, literally. We're talking about agent 00:14:58.279 --> 00:15:00.879 browser technology, the idea of agents running 00:15:00.879 --> 00:15:02.919 directly within your own web browser, not just 00:15:02.919 --> 00:15:04.860 on some remote virtual machine in the cloud. 00:15:05.159 --> 00:15:06.559 Yeah, this is really interesting. Companies like 00:15:06.559 --> 00:15:08.700 Perplexity are pushing hard in this direction. 00:15:09.440 --> 00:15:11.659 The capabilities are pretty impressive because 00:15:11.659 --> 00:15:14.909 they run locally. These agents can perform automations 00:15:14.909 --> 00:15:18.029 much faster with lower latency. They can securely 00:15:18.029 --> 00:15:20.389 access the services you're already logged into, 00:15:20.669 --> 00:15:23.549 your email, your social media accounts, your 00:15:23.549 --> 00:15:26.190 work tools, which allows for much more seamless 00:15:26.190 --> 00:15:28.889 integration into your personal workflow. You 00:15:28.889 --> 00:15:31.129 could potentially perform bulk operations much 00:15:31.129 --> 00:15:33.889 more easily, like cleaning up thousands of spam 00:15:33.889 --> 00:15:36.909 emails, or mass unfollowing inactive accounts, 00:15:37.350 --> 00:15:40.009 or curating content feeds based on your logged 00:15:40.009 --> 00:15:42.580 in history. And they could provide real -time 00:15:42.580 --> 00:15:44.960 assistance, offering contextual suggestions or 00:15:44.960 --> 00:15:47.659 actions as you're browsing different sites. Okay, 00:15:47.720 --> 00:15:50.500 so what might this look like in terms of practical 00:15:50.500 --> 00:15:53.299 daily use cases? Well, you could imagine setting 00:15:53.299 --> 00:15:56.080 up automations like, every morning, check my 00:15:56.080 --> 00:15:58.399 inbox, delete emails from known spam domains, 00:15:58.980 --> 00:16:00.740 automatically unsubscribe from newsletters I 00:16:00.740 --> 00:16:02.820 haven't opened in three months, and flag emails 00:16:02.820 --> 00:16:05.840 from my boss as important. or for content curation. 00:16:06.139 --> 00:16:08.820 Every day at 8 a .m. find the top five trending 00:16:08.820 --> 00:16:11.639 YouTube videos about sustainable urban development 00:16:11.639 --> 00:16:13.820 and put the links and summaries in a note for 00:16:13.820 --> 00:16:17.220 me, even social media management. Take this blog 00:16:17.220 --> 00:16:19.600 post draft, schedule it to post on my LinkedIn 00:16:19.600 --> 00:16:22.600 and Twitter tomorrow at 9 a .m. sharp, and monitor 00:16:22.600 --> 00:16:25.340 mentions of my brand name, flagging any potentially 00:16:25.340 --> 00:16:28.919 negative comments from my review here. The potential 00:16:28.919 --> 00:16:31.759 for streamlining our daily digital grind is pretty 00:16:31.759 --> 00:16:34.279 huge. So what's the main advantage then of having 00:16:34.279 --> 00:16:36.899 these agents run locally in your browser? Greater 00:16:36.899 --> 00:16:38.899 speed and seamless integration with your personal 00:16:38.899 --> 00:16:41.480 logged -in accounts. Sponsor. Okay, as these 00:16:41.480 --> 00:16:43.759 agent capabilities keep growing, and they are 00:16:43.759 --> 00:16:46.519 growing fast, responsible automation and managing 00:16:46.519 --> 00:16:49.240 the risks involved become just critically important. 00:16:49.600 --> 00:16:52.279 It feels a bit like giving a very powerful, very 00:16:52.279 --> 00:16:54.279 fast tool to someone who hasn't used it before. 00:16:54.440 --> 00:16:57.789 You absolutely need guardrails. Absolutely. You 00:16:57.789 --> 00:16:59.429 have to be careful financially, for instance. 00:16:59.769 --> 00:17:01.990 You must always be the final approver for any 00:17:01.990 --> 00:17:05.309 transaction. Never, ever give an agent unrestricted 00:17:05.309 --> 00:17:07.470 access to financial accounts or payment methods. 00:17:08.009 --> 00:17:10.710 For things like investment decisions, sure, agents 00:17:10.710 --> 00:17:13.410 can analyze data, but they completely lack human 00:17:13.410 --> 00:17:16.519 context, judgment, and risk tolerance. Don't 00:17:16.519 --> 00:17:18.880 let them make those calls. Be really cautious 00:17:18.880 --> 00:17:20.900 about data security when instructing them to 00:17:20.900 --> 00:17:23.500 access sensitive platforms. And like we said 00:17:23.500 --> 00:17:26.380 before, using dedicated test accounts for experimenting 00:17:26.380 --> 00:17:29.220 with agents is just smart practice to limit any 00:17:29.220 --> 00:17:31.700 potential damage if things go wrong. Those are 00:17:31.700 --> 00:17:34.200 crucial precautions. And thinking about prompt 00:17:34.200 --> 00:17:35.680 engineering, which we've touched on a few times, 00:17:35.799 --> 00:17:37.539 it's not just about getting good results. It's 00:17:37.539 --> 00:17:39.599 fundamental to safety and control too, isn't 00:17:39.599 --> 00:17:42.950 it? It's absolutely central to control. You need 00:17:42.950 --> 00:17:45.769 refined control through your prompts. Be extremely 00:17:45.769 --> 00:17:49.140 specific. Leave no room for ambiguity. Set clear 00:17:49.140 --> 00:17:52.380 boundaries, explicit stopping conditions, quantity 00:17:52.380 --> 00:17:55.960 limits, negative constraints. Don't do X. Use 00:17:55.960 --> 00:17:58.039 examples and references whenever possible to 00:17:58.039 --> 00:18:00.059 show the agent exactly the kind of output you 00:18:00.059 --> 00:18:03.380 want. And definitely start by testing with low 00:18:03.380 --> 00:18:07.200 stakes, easily reversible tasks. Research, maybe 00:18:07.200 --> 00:18:09.740 summarizing articles, drafting content before 00:18:09.740 --> 00:18:12.140 you even think about moving to high impact automations 00:18:12.140 --> 00:18:15.190 like client outreach or actual purchasing. and 00:18:15.190 --> 00:18:16.829 monitor continuously, especially when you're 00:18:16.829 --> 00:18:18.650 first setting things up. Be ready to jump in 00:18:18.650 --> 00:18:20.609 and stop it if needed. It also sounds like there 00:18:20.609 --> 00:18:22.509 are still definite technical limitations we need 00:18:22.509 --> 00:18:24.509 to be aware of, right? What can they not do effectively 00:18:24.509 --> 00:18:27.119 right now? Yeah. There are limits. Current agents 00:18:27.119 --> 00:18:29.880 generally can't use desktop software. They're 00:18:29.880 --> 00:18:32.160 mostly confined to browser -based applications 00:18:32.160 --> 00:18:35.059 and websites. Their output quality for really 00:18:35.059 --> 00:18:37.660 complex creative tasks, like nuanced writing 00:18:37.660 --> 00:18:39.920 or sophisticated design, is still pretty basic. 00:18:40.339 --> 00:18:42.779 They struggle significantly with nuanced decisions 00:18:42.779 --> 00:18:45.779 that require human judgment, empathy, or deep 00:18:45.779 --> 00:18:48.559 cultural understanding. And in very long, multi 00:18:48.559 --> 00:18:51.420 -step workflows, they can sometimes sort of forget 00:18:51.420 --> 00:18:53.839 their original objectives or lose the long -term 00:18:53.839 --> 00:18:56.380 context of the task. So putting it bluntly, what's 00:18:56.380 --> 00:18:59.099 the biggest risk if we're not careful with these 00:18:59.099 --> 00:19:01.819 agents? Losing financial control or seriously 00:19:01.819 --> 00:19:04.599 damaging your brand reputation. OK. The real 00:19:04.599 --> 00:19:06.859 power, then, it seems, starts to unlock when 00:19:06.859 --> 00:19:09.059 you begin combining these different agent capabilities 00:19:09.059 --> 00:19:11.180 together. Building your own automation systems 00:19:11.180 --> 00:19:13.500 for personal use or for business, it's about 00:19:13.500 --> 00:19:16.339 creating workflows. That's exactly the vision. 00:19:16.960 --> 00:19:20.160 To set up, say, your own daily automation system, 00:19:20.619 --> 00:19:22.680 you'd first identify those repetitive tasks you 00:19:22.680 --> 00:19:25.740 do all the time. Maybe checking competitor pricing, 00:19:26.039 --> 00:19:28.440 compiling social media reports, filtering your 00:19:28.440 --> 00:19:30.579 email inbox. Then you assess their automation 00:19:30.579 --> 00:19:32.980 potential. Are they mostly browser -based? Are 00:19:32.980 --> 00:19:35.640 they rule -driven enough for an agent? Then you 00:19:35.640 --> 00:19:37.759 design the workflow, which means writing those 00:19:37.759 --> 00:19:40.240 detailed step -by -step prompts we talked about. 00:19:40.500 --> 00:19:42.740 Then you test and refine. Start simple, maybe 00:19:42.740 --> 00:19:44.900 just one or two steps, and gradually increase 00:19:44.900 --> 00:19:47.200 the complexity as you get more confident. And 00:19:47.200 --> 00:19:50.410 always... Always monitor the performance, track 00:19:50.410 --> 00:19:52.650 the success rate, see where it messes up, and 00:19:52.650 --> 00:19:55.190 use that feedback to improve your prompts. So 00:19:55.190 --> 00:19:57.509 thinking bigger picture for agencies or small 00:19:57.509 --> 00:20:00.089 businesses, what could this really mean in terms 00:20:00.089 --> 00:20:02.410 of transforming how they operate? Oh, it could 00:20:02.410 --> 00:20:05.430 be huge. It could mean truly scalable service 00:20:05.430 --> 00:20:08.390 delivery. Imagine an SEO agency that can suddenly 00:20:08.390 --> 00:20:11.029 generate detailed technical audit reports for 00:20:11.029 --> 00:20:13.890 100 clients every month almost simultaneously 00:20:13.890 --> 00:20:16.750 without needing to triple their staff. It can 00:20:16.750 --> 00:20:19.130 enable a real shift towards outcome -based pricing. 00:20:19.730 --> 00:20:21.990 If you can automate much of the repetitive execution 00:20:21.990 --> 00:20:24.849 work, you can stop billing just for hours worked 00:20:24.849 --> 00:20:27.569 and start focusing on delivering specific results 00:20:27.569 --> 00:20:30.750 for clients. Rapid prototyping for new service 00:20:30.750 --> 00:20:33.119 ideas becomes much easier and cheaper. And it 00:20:33.119 --> 00:20:35.519 can improve quality control by ensuring consistent 00:20:35.519 --> 00:20:38.220 output for tasks like keyword research or market 00:20:38.220 --> 00:20:41.420 analysis. How could agents really transform a 00:20:41.420 --> 00:20:44.299 small business bottom line? They allow for rapid 00:20:44.299 --> 00:20:46.420 scaling and a stronger focus on outcomes, not 00:20:46.420 --> 00:20:49.819 just ours. Looking ahead, this whole agent revolution 00:20:49.819 --> 00:20:52.920 seems poised to profoundly reshape not just tasks, 00:20:53.059 --> 00:20:55.319 but the job market itself and even core business 00:20:55.319 --> 00:20:57.359 models. We're not just talking about new tools. 00:20:57.440 --> 00:20:59.240 It feels like we're talking about entirely new 00:20:59.240 --> 00:21:01.920 ways of working. I think that's right. We'll 00:21:01.920 --> 00:21:04.720 almost certainly see new job categories emerge. 00:21:04.980 --> 00:21:07.960 Things like AI agent supervisor, senior prompt 00:21:07.960 --> 00:21:10.839 engineer, maybe workflow automation specialist, 00:21:11.519 --> 00:21:14.079 titles we haven't even thought of yet. Business 00:21:14.079 --> 00:21:16.980 models will continue to shift, likely accelerating 00:21:16.980 --> 00:21:19.079 the move away from just selling software licenses 00:21:19.079 --> 00:21:21.720 towards selling access to outcome -based services 00:21:21.720 --> 00:21:24.720 powered by agents. Personal branding becomes 00:21:24.720 --> 00:21:27.579 even more important as your individual expertise 00:21:27.579 --> 00:21:30.420 becomes more discoverable and verifiable by agents 00:21:30.420 --> 00:21:33.660 searching for specific skills. And overall, businesses 00:21:33.660 --> 00:21:36.200 that effectively leverage agents could experience 00:21:36.200 --> 00:21:38.400 significantly accelerated growth by reducing 00:21:38.400 --> 00:21:40.700 personnel bottlenecks and operational costs. 00:21:41.299 --> 00:21:43.599 So for individuals listening, what are the crucial 00:21:43.599 --> 00:21:45.819 non -automatable skills we should be focusing 00:21:45.819 --> 00:21:48.000 on cultivating to thrive in this coming agent 00:21:48.000 --> 00:21:51.119 era? Okay, number one is deep domain expertise. 00:21:51.369 --> 00:21:54.009 Humans still need to provide the direction, the 00:21:54.009 --> 00:21:57.230 goals, the context, the AI executes. Semantic 00:21:57.230 --> 00:22:00.230 precision is vital, the ability to articulate 00:22:00.230 --> 00:22:02.890 complex requirements clearly and unambiguously 00:22:02.890 --> 00:22:06.089 to an AI agent. That's prompt engineering, but 00:22:06.089 --> 00:22:08.470 at a higher level. Strategic thinking becomes 00:22:08.470 --> 00:22:10.930 critical, deciding what is actually valuable 00:22:10.930 --> 00:22:13.589 to automate versus what absolutely needs that 00:22:13.589 --> 00:22:16.769 human touch or judgment. Adaptability, just being 00:22:16.769 --> 00:22:18.529 willing to constantly learn and experiment with 00:22:18.529 --> 00:22:21.349 these tools as they evolve almost weekly. And 00:22:21.349 --> 00:22:24.289 finally, risk assessment, evaluating when automation 00:22:24.289 --> 00:22:26.410 is appropriate and safe and when it's just too 00:22:26.410 --> 00:22:28.349 risky for a particular task. And you had to pick 00:22:28.349 --> 00:22:30.390 one. What's the single most important skill people 00:22:30.390 --> 00:22:32.789 need to cultivate to stay competitive? Strategic 00:22:32.789 --> 00:22:34.849 thinking combined with that precise communication 00:22:34.849 --> 00:22:37.769 skill for directing agents. So today we've really 00:22:37.769 --> 00:22:39.910 gone on a deep dive into this emerging world 00:22:39.910 --> 00:22:42.289 of AI agents, and it feels like a true paradigm 00:22:42.289 --> 00:22:44.250 shift. We're moving from just responsive chat 00:22:44.250 --> 00:22:47.490 to genuinely proactive action, giving AI these 00:22:47.490 --> 00:22:49.490 digital hands and eyes. Yeah, and this isn't 00:22:49.490 --> 00:22:51.769 just some far -off theory anymore. It's happening 00:22:51.769 --> 00:22:54.890 now. It's automating everything from, you know, 00:22:55.089 --> 00:22:57.609 online purchases and professional outreach to 00:22:57.609 --> 00:23:00.890 even complex content creation workflows. The 00:23:00.890 --> 00:23:03.309 ability to effectively define goals and direct 00:23:03.309 --> 00:23:07.039 these agents is becoming like a new kind of superpower. 00:23:07.230 --> 00:23:10.549 But, as we discussed, with that power comes significant 00:23:10.549 --> 00:23:13.430 responsibility. We learned how critically important 00:23:13.430 --> 00:23:16.529 it is to use specific prompts, to set clear boundaries, 00:23:16.910 --> 00:23:19.490 and to always maintain human oversight and control, 00:23:19.710 --> 00:23:21.730 especially when dealing with sensitive tasks 00:23:21.730 --> 00:23:25.690 like finances or brand reputation. Exactly. The 00:23:25.690 --> 00:23:28.470 future here isn't really about AI replacing humans 00:23:28.470 --> 00:23:30.789 entirely. It's much more about humans learning 00:23:30.789 --> 00:23:34.089 how to effectively team up with AI. Our domain 00:23:34.089 --> 00:23:36.750 expertise, our strategic thinking, our ability 00:23:36.589 --> 00:23:38.710 need to communicate precisely those skills are 00:23:38.710 --> 00:23:40.690 going to be more valuable than ever in directing 00:23:40.690 --> 00:23:43.430 these powerful tools. The era of AI agents isn't 00:23:43.430 --> 00:23:45.289 just on the horizon. It feels like it's already 00:23:45.289 --> 00:23:48.049 here, actively reshaping how we work, how we 00:23:48.049 --> 00:23:50.009 learn, how we interact with the digital world. 00:23:50.509 --> 00:23:52.069 The question really isn't whether you should 00:23:52.069 --> 00:23:54.309 start using them, but maybe how quickly you can 00:23:54.309 --> 00:23:56.670 master them to stay ahead of the curve. Indeed. 00:23:57.309 --> 00:23:59.720 You can think of it this way. The future of work, 00:23:59.940 --> 00:24:02.359 a lot of it, is being written right now in code 00:24:02.359 --> 00:24:05.539 and in prompts, and AI agents are kind of holding 00:24:05.539 --> 00:24:09.019 the pen. Our role as humans is to guide that 00:24:09.019 --> 00:24:11.700 hand effectively. So maybe a final thought for 00:24:11.700 --> 00:24:14.980 everyone listening. What's one specific repetitive 00:24:14.980 --> 00:24:18.059 task in your daily routine that an AI agent could 00:24:18.059 --> 00:24:20.779 potentially tackle for you, even just partially, 00:24:21.240 --> 00:24:23.200 starting this week? Think about it, maybe do 00:24:23.200 --> 00:24:25.180 a little research, and start experimenting in 00:24:25.180 --> 00:24:27.650 a low -risk way. We really hope this Deep Dive 00:24:27.650 --> 00:24:29.769 has given you a clearer picture of what AI agents 00:24:29.769 --> 00:24:32.390 are and maybe some actionable first steps for 00:24:32.390 --> 00:24:34.609 starting to integrate them into your world. The 00:24:34.609 --> 00:24:37.450 key is to stay curious, stay adaptable, and keep 00:24:37.450 --> 00:24:39.789 asking how these rapidly evolving technologies 00:24:39.789 --> 00:24:42.190 can best serve your specific goals. Thank you 00:24:42.190 --> 00:24:43.990 so much for joining us on the Deep Dive. Until 00:24:43.990 --> 00:24:45.970 next time, keep exploring, keep learning, and 00:24:45.970 --> 00:24:48.609 keep asking the big questions. Out to your own 00:24:48.609 --> 00:24:48.849 music.