WEBVTT 00:00:00.000 --> 00:00:03.180 Imagine your smartest search engine, right? But 00:00:03.180 --> 00:00:05.400 instead of just giving you an answer, it actually 00:00:05.400 --> 00:00:10.519 executed a complex multi -step plan all by itself. 00:00:10.880 --> 00:00:13.640 What if it could log into your email, find a 00:00:13.640 --> 00:00:15.960 specific document link, analyze it, and then 00:00:15.960 --> 00:00:18.239 pop a summary into your Notion while you were 00:00:18.239 --> 00:00:20.219 sleeping? Well, that shift is basically, we're 00:00:20.219 --> 00:00:22.399 moving past just, you know, passive chatbots. 00:00:22.460 --> 00:00:25.899 We're entering the era of the proactive digital 00:00:25.899 --> 00:00:28.960 workforce. Okay. And our focus today is perplexity 00:00:28.960 --> 00:00:31.850 comment. It's this autonomous AI agent system 00:00:31.850 --> 00:00:34.329 designed specifically for those sophisticated, 00:00:34.549 --> 00:00:37.229 integrated workflows that, let's be honest, eat 00:00:37.229 --> 00:00:39.609 up hours of our time right now. A tireless digital 00:00:39.609 --> 00:00:41.530 assistant, essentially. Exactly. Think of it 00:00:41.530 --> 00:00:43.789 that way. Welcome to the Deep Dive. So today 00:00:43.789 --> 00:00:45.770 we're unpacking the sources we found detailing 00:00:45.770 --> 00:00:48.490 perplexity comets, foundational capabilities. 00:00:48.950 --> 00:00:50.950 The mission here is to really look under the 00:00:50.950 --> 00:00:53.469 hood, understand not just that it works, but 00:00:53.469 --> 00:00:56.030 how this level of autonomy is actually possible. 00:00:56.329 --> 00:00:58.229 Yeah. And we've pulled together about seven... 00:00:58.649 --> 00:01:01.250 Use cases from the material that show some pretty 00:01:01.250 --> 00:01:03.810 massive productivity games. We'll kick things 00:01:03.810 --> 00:01:06.670 off by defining the core tech that makes this 00:01:06.670 --> 00:01:09.390 autonomy real. Something called agent chaining. 00:01:09.450 --> 00:01:11.250 Agent chaining. And then we'll dive into these 00:01:11.250 --> 00:01:13.650 high leverage examples, the ones that turn like 00:01:13.650 --> 00:01:17.030 tedious eight hour admin tasks into maybe 20 00:01:17.030 --> 00:01:19.609 minutes of just waiting. All right, let's get 00:01:19.609 --> 00:01:22.250 into it. This core innovation comment. It feels 00:01:22.250 --> 00:01:25.450 like a really significant leap beyond just asking 00:01:25.450 --> 00:01:27.750 a question and getting an answer back. It absolutely 00:01:27.750 --> 00:01:30.530 is. The sources emphasize that these agents aren't 00:01:30.530 --> 00:01:33.129 stuck in a chat window. They actually operate 00:01:33.129 --> 00:01:35.950 out there in your real applications. That's the 00:01:35.950 --> 00:01:38.430 crucial difference. And to do that, they need 00:01:38.430 --> 00:01:40.950 a whole suite of features for truly autonomous 00:01:40.950 --> 00:01:44.049 operation. And right at the top, agent chaining. 00:01:44.090 --> 00:01:45.969 That's the core engine, the real innovation here. 00:01:46.090 --> 00:01:48.760 So agent chaining. if i were to explain it simply 00:01:48.760 --> 00:01:51.659 it's like uh stacking specialized lego blocks 00:01:51.659 --> 00:01:53.799 you connect multiple agents each good at one 00:01:53.799 --> 00:01:56.719 thing in a sequence like a digital assembly line 00:01:56.719 --> 00:01:59.739 that's a great analogy one ai finds the resource 00:01:59.739 --> 00:02:02.700 passes its output to the next one which analyzes 00:02:02.700 --> 00:02:05.340 it maybe a third one formats the result okay 00:02:05.340 --> 00:02:08.219 and it automates these complex multi -step workflows 00:02:08.219 --> 00:02:11.199 without needing a human to step in between stages 00:02:11.199 --> 00:02:15.090 but for that to work These agents need cross 00:02:15.090 --> 00:02:17.210 -platform access. Right. You need to get into 00:02:17.210 --> 00:02:20.330 the apps. Exactly. Secure integration with your 00:02:20.330 --> 00:02:23.229 actual ecosystem, Gmail, Slack, Notion, whatever 00:02:23.229 --> 00:02:26.650 you use. They operate inside those tools. So 00:02:26.650 --> 00:02:29.270 that access is key. If agent chaining is the 00:02:29.270 --> 00:02:32.909 assembly line, how critical is that ability to 00:02:32.909 --> 00:02:36.520 securely get into, say, your email for these 00:02:36.520 --> 00:02:38.479 workflows to actually succeed. Oh, it's absolutely 00:02:38.479 --> 00:02:41.159 fundamental. Accessing those apps transforms 00:02:41.159 --> 00:02:43.539 the AI from something that just gives you information 00:02:43.539 --> 00:02:46.080 to something that performs actions for you. Access 00:02:46.080 --> 00:02:48.740 transforms the AI from search tool to digital 00:02:48.740 --> 00:02:51.110 worker. Got it. Yeah. It's the difference between 00:02:51.110 --> 00:02:53.490 asking where the store is and asking the AI to, 00:02:53.590 --> 00:02:55.370 you know, order your groceries from the store. 00:02:55.449 --> 00:02:57.189 And the way they handle context seems pretty 00:02:57.189 --> 00:02:59.210 smart, too. The sources talk about this at Symbol 00:02:59.210 --> 00:03:01.930 Magic. Oh, yeah. The at symbol. Using that, you 00:03:01.930 --> 00:03:04.250 can tell an agent to look at info from like an 00:03:04.250 --> 00:03:06.870 open browser tab or a document you uploaded or 00:03:06.870 --> 00:03:09.389 even just a previous chat thread. It avoids all 00:03:09.389 --> 00:03:11.449 that copying and pasting. Right. It makes the 00:03:11.449 --> 00:03:13.770 context so much richer, more immediate, makes 00:03:13.770 --> 00:03:15.870 the whole interaction feel more efficient, more. 00:03:17.159 --> 00:03:20.379 human in a way. Plus, there's the power of scheduled 00:03:20.379 --> 00:03:23.139 automation. You build a workflow once, save it, 00:03:23.199 --> 00:03:25.639 and then just set it to run daily or weekly. 00:03:25.759 --> 00:03:28.759 Imagine waking up Monday morning to a full competitive 00:03:28.759 --> 00:03:31.379 analysis report just sitting in your inbox. Set 00:03:31.379 --> 00:03:34.539 it and forget it AI style. That combination, 00:03:34.979 --> 00:03:37.560 the internal orchestration with agent chaining 00:03:37.560 --> 00:03:40.300 and the external access to apps. That really 00:03:40.300 --> 00:03:42.560 defines this new level of autonomy, doesn't it? 00:03:42.620 --> 00:03:44.639 It really does. Okay, let's make this concrete. 00:03:44.780 --> 00:03:46.560 Let's look at one of those time sinks, the email 00:03:46.560 --> 00:03:49.280 -to -link workflow. Manually, just trying to 00:03:49.280 --> 00:03:51.219 find a customer summary someone emailed you last 00:03:51.219 --> 00:03:54.319 week, buried in some link. That's a whole sequence 00:03:54.319 --> 00:03:57.699 of searching, clicking, reading, summarizing. 00:03:58.180 --> 00:04:02.580 Yeah, total cognitive load. So Comet automates 00:04:02.580 --> 00:04:05.069 that whole chain. Complex, right? Find the right 00:04:05.069 --> 00:04:08.409 email, extract maybe a hidden URL, navigate to 00:04:08.409 --> 00:04:10.689 that page, actually analyze the content, then 00:04:10.689 --> 00:04:14.710 summarize it usefully. Five distinct steps. Minimum. 00:04:14.729 --> 00:04:17.410 But the user prompt is super simple. Just one 00:04:17.410 --> 00:04:20.110 high -level instruction. Something like, find 00:04:20.110 --> 00:04:22.389 Jane Doe's email about the customer draft and 00:04:22.389 --> 00:04:24.529 summarize the demographics in the link she sent. 00:04:24.730 --> 00:04:27.839 Exactly. Behind the scenes, Comet's orchestrating 00:04:27.839 --> 00:04:30.639 maybe five or more specialized agents. Email 00:04:30.639 --> 00:04:33.779 agent, link extractor, navigator, analyzer, summarizer, 00:04:33.879 --> 00:04:37.199 all in sequence. The user completely hands off 00:04:37.199 --> 00:04:39.540 after the prompt. So, okay, it's faster. But 00:04:39.540 --> 00:04:41.879 beyond speed, what's the biggest functional difference 00:04:41.879 --> 00:04:44.720 between me manually digging through emails versus 00:04:44.720 --> 00:04:46.899 an autonomous agent doing the whole sequence? 00:04:47.180 --> 00:04:48.699 Well, when you do it manually, you get interrupted, 00:04:48.779 --> 00:04:50.600 right? Click a link, see another email, suddenly 00:04:50.600 --> 00:04:52.060 you're down a rabbit hole. It happens all the 00:04:52.060 --> 00:04:54.629 time. The agent. It offers hands -off execution 00:04:54.629 --> 00:04:57.050 and guaranteed accuracy across that whole multi 00:04:57.050 --> 00:04:59.589 -step process. No distractions, fewer errors. 00:04:59.750 --> 00:05:02.509 The values in that reliable hands -off execution 00:05:02.509 --> 00:05:05.930 make sense. Okay, so to manage all this, this 00:05:05.930 --> 00:05:09.129 digital workforce, you need a control panel, 00:05:09.170 --> 00:05:11.529 right? Yeah, the command center. The sources 00:05:11.529 --> 00:05:13.829 detail the interface. You've got the main chat 00:05:13.829 --> 00:05:16.850 window, an assistant panel, like a co -pilot, 00:05:17.009 --> 00:05:19.449 and a show cuts panel. And for me, the most interesting 00:05:19.449 --> 00:05:21.589 part, maybe the most important for building trust, 00:05:21.810 --> 00:05:24.889 is the preview window. Oh, what's that? It gives 00:05:24.889 --> 00:05:27.810 you this real -time, transparent view of what 00:05:27.810 --> 00:05:29.870 the agent is actually doing. You see it navigating 00:05:29.870 --> 00:05:31.750 websites, clicking buttons, interacting with 00:05:31.750 --> 00:05:34.189 apps for you. You can literally watch it work. 00:05:34.620 --> 00:05:36.779 OK, that transparency feels critical, especially 00:05:36.779 --> 00:05:38.920 if you're trusting it with, you know, sensitive 00:05:38.920 --> 00:05:41.899 stuff. Exactly. Which brings us to custom shortcuts. 00:05:42.439 --> 00:05:45.379 Think of these as personalized, reusable agents 00:05:45.379 --> 00:05:47.920 you build yourself. You tailor them for specific 00:05:47.920 --> 00:05:50.920 recurring tasks you do all the time. So you define 00:05:50.920 --> 00:05:54.019 the name, the instructions, which AI model it 00:05:54.019 --> 00:05:56.860 uses, and importantly, the sources it can access, 00:05:57.079 --> 00:06:00.180 like only my Gmail and Notion or only public 00:06:00.180 --> 00:06:02.439 web search. Precisely. That control over sources 00:06:02.439 --> 00:06:04.899 is key. It's funny. I still wrestle with prompt 00:06:04.899 --> 00:06:07.180 drift myself sometimes, you know, especially 00:06:07.180 --> 00:06:09.579 if I'm relying on older context I save somewhere 00:06:09.579 --> 00:06:12.639 to generate replies. That feels like a potential 00:06:12.639 --> 00:06:15.540 risk here if it's accessing deep personal data. 00:06:15.660 --> 00:06:17.899 That's a really valid point, and it's a key consideration. 00:06:18.259 --> 00:06:21.240 The sources actually mention a bonus use case 00:06:21.240 --> 00:06:24.360 autofilling forms, like for podcast guests. You're 00:06:24.360 --> 00:06:26.519 right. The agent can fill out a complex form 00:06:26.519 --> 00:06:28.939 in under two minutes. Huge time saver. Yeah. 00:06:29.100 --> 00:06:32.189 But. And this is crucial. The material explicitly 00:06:32.189 --> 00:06:36.370 says you must review the agent's output. Because 00:06:36.370 --> 00:06:38.850 it might pull slightly outdated info, maybe an 00:06:38.850 --> 00:06:41.230 old bio from an email somewhere. Speed is great, 00:06:41.310 --> 00:06:44.370 but accuracy needs that human check. Okay, so 00:06:44.370 --> 00:06:47.050 given that context can shift slightly, how do 00:06:47.050 --> 00:06:49.529 these customizable shortcuts maintain consistency 00:06:49.529 --> 00:06:52.339 for those critical recurring tasks? Well, the 00:06:52.339 --> 00:06:54.579 shortcuts lock in the instructions. They ensure 00:06:54.579 --> 00:06:56.519 recurring tasks get performed the exact same 00:06:56.519 --> 00:06:59.420 way every time. You don't have to retype complex 00:06:59.420 --> 00:07:02.100 commands and risk variations. Shortcuts ensure 00:07:02.100 --> 00:07:04.660 consistency by standardizing the instructions. 00:07:04.959 --> 00:07:08.110 Got it, sponsor. All right, let's shift gears 00:07:08.110 --> 00:07:10.709 a bit from saving minutes to saving hours. We're 00:07:10.709 --> 00:07:12.949 getting into more strategic automation now. Let's 00:07:12.949 --> 00:07:14.550 take the YouTube channel performance analysis. 00:07:14.970 --> 00:07:17.410 Use case two. That sounds like a beast. Oh, it 00:07:17.410 --> 00:07:19.490 is. If you're a content creator or an analyst 00:07:19.490 --> 00:07:23.529 manually going through, say, 64 videos, categorizing 00:07:23.529 --> 00:07:25.829 topics, checking view counts, watch time, trying 00:07:25.829 --> 00:07:29.009 to spot trends, that's easily six to nine hours 00:07:29.009 --> 00:07:32.089 of really focused work. A necessary but, yeah, 00:07:32.149 --> 00:07:34.949 brutal admin deep dive. Okay, so how did the 00:07:34.949 --> 00:07:37.819 agent handle it? Single tromped. The agent, or 00:07:37.819 --> 00:07:40.319 rather a chain of agents, scans all the channel 00:07:40.319 --> 00:07:42.680 data, categorizes everything, identifies the 00:07:42.680 --> 00:07:44.879 top performers, the underperformers, and then 00:07:44.879 --> 00:07:47.100 here's the really strategic bit. It recommends 00:07:47.100 --> 00:07:50.199 five new trending topics to cover based on market 00:07:50.199 --> 00:07:53.480 data analysis. Whoa. The ROI is just staggering. 00:07:53.540 --> 00:07:55.920 The user's time drops to maybe 15, 20 minutes, 00:07:55.980 --> 00:07:58.139 and most of that is just passive waiting while 00:07:58.139 --> 00:08:00.480 the report gets compiled. Short pause. Seriously, 00:08:00.540 --> 00:08:02.899 imagine scaling that. Across dozens of competitor 00:08:02.899 --> 00:08:05.240 channels. Wow. Saving six to nine hours in 20 00:08:05.240 --> 00:08:07.680 minutes. An analyst could spend those saved hours 00:08:07.680 --> 00:08:10.699 actually creating or strategizing based on the 00:08:10.699 --> 00:08:12.939 insights, not just digging for them. That's the 00:08:12.939 --> 00:08:15.160 real strategic shift. It totally elevates the 00:08:15.160 --> 00:08:18.720 human role. Okay. Another powerful one. The news 00:08:18.720 --> 00:08:23.300 concierge agent. Use case three. Automating research 00:08:23.300 --> 00:08:26.519 for something niche like AI and personal finance 00:08:26.519 --> 00:08:29.930 news. Yeah, manually curating really relevant, 00:08:29.990 --> 00:08:31.930 high -quality stuff for a specific audience. 00:08:32.090 --> 00:08:35.129 That takes hours every single week. So in this 00:08:35.129 --> 00:08:37.929 use case, the prompt is super precise. It clearly 00:08:37.929 --> 00:08:40.590 defines the AI's role. You are a world -class 00:08:40.590 --> 00:08:43.470 research assistant. And crucially, it uses that 00:08:43.470 --> 00:08:45.769 symbol again. It links to a previous article 00:08:45.769 --> 00:08:48.269 the user liked, setting a clear benchmark for 00:08:48.269 --> 00:08:50.029 tone, for quality. So you're not just telling 00:08:50.029 --> 00:08:52.009 it what to find, but how to judge quality and 00:08:52.009 --> 00:08:53.850 what kind of analytical lens to use. Exactly. 00:08:53.850 --> 00:08:56.679 And specifying the output format like... title 00:08:56.679 --> 00:08:59.559 three five line summary source link means the 00:08:59.559 --> 00:09:01.480 output is instantly usable for the newsletter 00:09:01.480 --> 00:09:03.399 and once you save that as a scheduled task yeah 00:09:03.399 --> 00:09:06.059 say every friday morning the active work for 00:09:06.059 --> 00:09:09.379 the creator drops to zero zero minutes per week 00:09:09.379 --> 00:09:13.019 but that level of precision it hinges on getting 00:09:13.019 --> 00:09:16.019 the prompt right how important is that clear 00:09:16.019 --> 00:09:18.860 role definition world -class research assistant 00:09:18.860 --> 00:09:21.659 for making these complex research agents really 00:09:22.080 --> 00:09:25.179 Nail it. Oh, role definition is absolutely essential. 00:09:25.419 --> 00:09:27.559 It guides the AI to apply the right expertise, 00:09:27.720 --> 00:09:30.279 the right analytical lens. It ensures the output 00:09:30.279 --> 00:09:32.139 aligns with the strategic goal, not just some 00:09:32.139 --> 00:09:35.580 generic search results. OK, final segment. Let's 00:09:35.580 --> 00:09:37.179 look at the most sophisticated stuff analysis 00:09:37.179 --> 00:09:39.080 that goes beyond research and delivers reports 00:09:39.080 --> 00:09:42.340 right into team tools. Use case for the LinkedIn 00:09:42.340 --> 00:09:44.940 content researcher to Slack report. Sounds like 00:09:44.940 --> 00:09:47.039 great competitive intel. Totally. The goal is 00:09:47.039 --> 00:09:49.580 clear. Analyze five specific competitor LinkedIn 00:09:49.580 --> 00:09:52.100 accounts. Yeah. Find the top 10 most engaged 00:09:52.100 --> 00:09:54.840 posts from the last week. Compile it and deliver 00:09:54.840 --> 00:09:56.620 it automatically to a specific Slack channel. 00:09:56.799 --> 00:09:58.720 So it's orchestrating LinkedIn scraping, then 00:09:58.720 --> 00:10:00.519 some pretty smart analysis, report building, 00:10:00.659 --> 00:10:03.340 and finally secure Slack delivery. Right. And 00:10:03.340 --> 00:10:05.580 it cuts down what could be a two to three hour 00:10:05.580 --> 00:10:08.580 manual reporting task to maybe 15, 20 minutes 00:10:08.580 --> 00:10:11.139 of just waiting. Yeah. Freeze up analysts for 00:10:11.139 --> 00:10:13.210 the important part. interpreting the strategy. 00:10:13.470 --> 00:10:16.230 Amazing. And even more advanced seems to be the 00:10:16.230 --> 00:10:19.750 partnerships manager agent, use case six. This 00:10:19.750 --> 00:10:22.470 scans Gmail for partnership emails. Yeah, scans 00:10:22.470 --> 00:10:25.250 Gmail, pulls out key details, sender, company, 00:10:25.409 --> 00:10:28.210 maybe the tool URL, and adds it all neatly into 00:10:28.210 --> 00:10:30.250 a Notion database. Okay, that's useful organization. 00:10:30.710 --> 00:10:33.269 But the sources say the real value is something 00:10:33.269 --> 00:10:35.990 more. Yes. The prompt actually instructs the 00:10:35.990 --> 00:10:39.039 agent to provide strategic recommendations. Right 00:10:39.039 --> 00:10:40.500 there in the notion, though, it's like based 00:10:40.500 --> 00:10:42.019 on its understanding of your business goals, 00:10:42.200 --> 00:10:44.620 is this partnership actually worth pursuing or 00:10:44.620 --> 00:10:47.799 not? So it acts like an AI gatekeeper, pre -analyzing, 00:10:47.799 --> 00:10:50.240 organizing, even scoring requests. Yeah, exactly. 00:10:50.679 --> 00:10:53.240 But when you get into strategic conclusions like 00:10:53.240 --> 00:10:55.460 that, recommending for or against a partnership, 00:10:55.740 --> 00:10:57.779 the source material strongly advises critical 00:10:57.779 --> 00:11:00.710 review. So what does trust but verify really 00:11:00.710 --> 00:11:03.149 mean in that specific context? When the AI is 00:11:03.149 --> 00:11:05.330 doing strategic scoring. Right. It means you 00:11:05.330 --> 00:11:07.830 always have to review the AI's reasoning, look 00:11:07.830 --> 00:11:10.250 at the data it used, check its conclusions before 00:11:10.250 --> 00:11:11.990 you make a big business decision off the back 00:11:11.990 --> 00:11:14.529 of it. You're confirming it aligns with human 00:11:14.529 --> 00:11:17.509 strategy, not just blindly accepting its score. 00:11:17.750 --> 00:11:20.330 Makes sense. And finally, they briefly mentioned 00:11:20.330 --> 00:11:24.830 use case seven, which felt almost meta. Using 00:11:24.830 --> 00:11:26.909 the Comet Assistant to help you build complex 00:11:26.909 --> 00:11:30.309 workflows inside another tool, like OpenAI's 00:11:30.309 --> 00:11:33.450 Agent Builder. Yeah, AI helping build AI structures, 00:11:33.649 --> 00:11:36.230 bridging that complexity gap for advanced users 00:11:36.230 --> 00:11:38.690 almost instantly. It shows the system operating 00:11:38.690 --> 00:11:40.570 at this really high level, right? Not just doing 00:11:40.570 --> 00:11:43.049 your admin, but helping you craft the next wave 00:11:43.049 --> 00:11:45.590 of automation tools yourself. Pretty cool. So, 00:11:45.629 --> 00:11:47.629 wrapping this up. What we've really seen today 00:11:47.629 --> 00:11:50.250 feels like the undeniable start of the autonomous 00:11:50.250 --> 00:11:52.549 agent era. AI isn't just a passive assistant 00:11:52.549 --> 00:11:54.850 anymore. It's becoming an active, independent 00:11:54.850 --> 00:11:57.909 workforce, handling entire complex workflows 00:11:57.909 --> 00:12:00.409 across all our professional tools. Yeah, the 00:12:00.409 --> 00:12:04.149 ROI is just, it's undeniable. Across all these 00:12:04.149 --> 00:12:07.169 examples, these multi -step, deeply integrated 00:12:07.169 --> 00:12:09.629 workflows, they conservatively save professionals 00:12:09.629 --> 00:12:13.059 10. maybe 15 hours a week. And the advantages 00:12:13.059 --> 00:12:15.879 boil down to three things, true autonomy, deep 00:12:15.879 --> 00:12:17.860 integration with the tools we already use, and 00:12:17.860 --> 00:12:20.820 effortless scheduled automation. So here's a 00:12:20.820 --> 00:12:22.919 thought to leave you with. If you can now automate 00:12:22.919 --> 00:12:25.159 all that routine analysis, all that recurring 00:12:25.159 --> 00:12:28.059 reporting, what's the highest value strategic 00:12:28.059 --> 00:12:30.179 task you could dedicate those extra 10 or 15 00:12:30.179 --> 00:12:33.070 hours to this week? That's the new potential 00:12:33.070 --> 00:12:35.110 this tech unlocks. Definitely something to think 00:12:35.110 --> 00:12:37.830 about. Consider those complex multi -step tasks 00:12:37.830 --> 00:12:40.450 that just bog down your schedule right now. Think 00:12:40.450 --> 00:12:42.490 about how you might break them down into a chain 00:12:42.490 --> 00:12:44.990 of agent workflows and really focus on defining 00:12:44.990 --> 00:12:47.389 clear output formats, whether that's a table, 00:12:47.529 --> 00:12:49.769 a structured Slack message, a pre -scored notion 00:12:49.769 --> 00:12:52.389 entry. That seems to be the key to getting immediate 00:12:52.389 --> 00:12:54.309 value from these new autonomous systems.