WEBVTT 00:00:00.000 --> 00:00:02.620 Welcome to the Deep Dive. Let's start with a 00:00:02.620 --> 00:00:05.019 claim that honestly sounds almost unbelievable, 00:00:05.280 --> 00:00:07.580 but it seems to be reshaping how work gets done. 00:00:08.320 --> 00:00:10.939 Imagine building a really complex AI automation, 00:00:11.240 --> 00:00:13.179 say something with, I don't know, 25 different 00:00:13.179 --> 00:00:15.160 steps, different tools talking to each other. 00:00:15.699 --> 00:00:17.660 Historically, that was, well, that was an expert's 00:00:17.660 --> 00:00:20.699 job, measured in hours, maybe even days. And 00:00:20.699 --> 00:00:23.420 now, yeah, now we're hearing that the entire 00:00:23.420 --> 00:00:25.519 structure, the whole thing mapped out, connected, 00:00:25.660 --> 00:00:28.579 partly set up, can be generated in about 90 seconds. 00:00:28.640 --> 00:00:30.600 That's the leap we're diving into today. It's 00:00:30.600 --> 00:00:32.759 kind of staggering. We are digging into sources 00:00:32.759 --> 00:00:35.780 all about Nintent's new AI workflow builder. 00:00:36.020 --> 00:00:37.899 And this isn't just, you know. A minor update, 00:00:38.079 --> 00:00:40.799 the claim is a 10x shift in how fast you can 00:00:40.799 --> 00:00:43.039 build things, opening up this whole agent era 00:00:43.039 --> 00:00:45.759 to pretty much everyone. Right. Our mission here 00:00:45.759 --> 00:00:48.340 is to untack how this tool basically tears down 00:00:48.340 --> 00:00:50.600 the old barriers. You know, automation used to 00:00:50.600 --> 00:00:53.320 be for coders, for specialists. The idea now 00:00:53.320 --> 00:00:57.280 is it's for, well, for you, for anyone. So we'll 00:00:57.280 --> 00:00:59.100 look at that speed increase, the different ways 00:00:59.100 --> 00:01:01.539 this AI builder actually works, and walk through 00:01:01.539 --> 00:01:04.379 three examples, simple to seriously complex. 00:01:04.840 --> 00:01:07.280 And crucially, what this means for the economics 00:01:07.280 --> 00:01:11.239 of your time and your work. Okay, let's get into 00:01:11.239 --> 00:01:14.439 it. Thinking back to how automation used to be 00:01:14.439 --> 00:01:16.900 built, it wasn't just hard, it felt like there 00:01:16.900 --> 00:01:18.950 was so much friction. Like the system wanted 00:01:18.950 --> 00:01:21.109 you to be an expert. Oh, absolutely. Even for 00:01:21.109 --> 00:01:23.750 someone like me who lives in these tools, a simple 00:01:23.750 --> 00:01:27.109 three -step thing, 10, 15 minutes, easy. If you 00:01:27.109 --> 00:01:28.810 were just starting out, forget it. Half an hour 00:01:28.810 --> 00:01:30.870 minimum and probably feeling pretty defeated 00:01:30.870 --> 00:01:32.650 afterwards. Yeah, defeated is a good word. The 00:01:32.650 --> 00:01:34.750 friction points were just everywhere. You had 00:01:34.750 --> 00:01:37.129 to manually look through hundreds of these nodes, 00:01:37.269 --> 00:01:39.310 these little tools, just to pick the right one. 00:01:40.269 --> 00:01:42.209 Setting up credentials. Yeah. Especially for 00:01:42.209 --> 00:01:44.129 Google stuff. That was like this legendary headache. 00:01:44.269 --> 00:01:46.510 It honestly felt like you needed a PhD sometimes 00:01:46.510 --> 00:01:49.090 just to connect two apps. Right. And the time 00:01:49.090 --> 00:01:52.370 spent debugging. Oh man, that often took longer 00:01:52.370 --> 00:01:54.849 than actually building it. But that whole mess 00:01:54.849 --> 00:01:57.569 just collapses with this AI co -pilot approach. 00:01:57.969 --> 00:02:01.890 Simple stuff. One, two minutes. complex five 00:02:01.890 --> 00:02:05.230 to ten and the extreme case that one shot build 00:02:05.230 --> 00:02:07.310 yeah under two minutes so the whole interaction 00:02:07.310 --> 00:02:09.650 changes you just you describe what you want in 00:02:09.650 --> 00:02:12.189 plain english Pretty much. You give it the prompt, 00:02:12.310 --> 00:02:15.490 the AI figures out the right nodes, connects 00:02:15.490 --> 00:02:18.310 them logically, and here's the kicker, it intelligently 00:02:18.310 --> 00:02:20.569 grabs your saved credentials if you have them. 00:02:20.710 --> 00:02:22.990 And it even puts in some basic settings. It's 00:02:22.990 --> 00:02:25.009 basically building the skeleton for you. That 00:02:25.009 --> 00:02:26.650 doesn't sound like just making things faster. 00:02:26.750 --> 00:02:28.530 That sounds like completely removing the wall 00:02:28.530 --> 00:02:30.889 that kept most people out. It really is a paradigm 00:02:30.889 --> 00:02:33.430 shift. It is. So if the setup gets that simple, 00:02:33.550 --> 00:02:36.050 what's left for the user to actually do? The 00:02:36.050 --> 00:02:38.789 user gets to focus on the strategy. the why and 00:02:38.789 --> 00:02:40.629 then checking the final details maybe that last 00:02:40.629 --> 00:02:43.169 10 percent okay so it's important to understand 00:02:43.169 --> 00:02:46.430 this ai workflow builder isn't just like a chatbot 00:02:46.430 --> 00:02:49.270 giving advice it's described as an active agent 00:02:49.270 --> 00:02:52.430 it has access to all the nen documentation knows 00:02:52.430 --> 00:02:55.889 about the 500 plus tools or nodes and the best 00:02:55.889 --> 00:02:58.509 ways to use them right nodes think of them like 00:02:58.509 --> 00:03:01.509 specialized lego blocks for your automation each 00:03:01.509 --> 00:03:04.750 one does a specific job gets data processes it 00:03:04.750 --> 00:03:07.120 sends it somewhere else And the sources highlight 00:03:07.120 --> 00:03:10.800 two really distinct modes for this AI. Knowing 00:03:10.800 --> 00:03:13.599 the difference seems key. Let's start with ask 00:03:13.599 --> 00:03:16.280 mode. Ask mode is your helper. It's like having 00:03:16.280 --> 00:03:18.419 a consultant sitting next to you. It can explain 00:03:18.419 --> 00:03:20.759 how a node works, suggest ways to build something, 00:03:20.919 --> 00:03:23.020 answer your questions. But, and this is crucial, 00:03:23.120 --> 00:03:24.819 it won't actually change anything in your workflow. 00:03:24.939 --> 00:03:27.060 It's read -only, basically. Good for planning. 00:03:27.199 --> 00:03:29.060 Okay, planner mode. Then there's build mode. 00:03:29.199 --> 00:03:31.300 That sounds like where the action happens. That's 00:03:31.300 --> 00:03:33.729 the agent doing the work. It adds the nodes, 00:03:33.909 --> 00:03:36.409 connects the wires, fills in the settings, even 00:03:36.409 --> 00:03:38.590 handles credentials, all based on your prompt. 00:03:38.870 --> 00:03:41.229 This is how you get those super fast builds. 00:03:41.770 --> 00:03:45.189 But there's a catch. A pretty significant one 00:03:45.189 --> 00:03:48.889 mentioned. A 1 ,000 character limit for the prompt. 00:03:49.129 --> 00:03:52.900 1 ,000 characters. Hmm. That does feel restrictive. 00:03:53.060 --> 00:03:55.479 Doesn't that immediately rule out a lot of complex, 00:03:55.740 --> 00:03:58.740 real -world business processes that need detailed 00:03:58.740 --> 00:04:00.860 instructions? Yeah, that's the challenge they 00:04:00.860 --> 00:04:03.599 flagged. Detailed automations often need way 00:04:03.599 --> 00:04:06.099 more than 11 ,000 characters to describe accurately, 00:04:06.419 --> 00:04:08.680 forcing you to be super concise or figure out 00:04:08.680 --> 00:04:11.180 another way. So why is that 1 ,000 -character 00:04:11.180 --> 00:04:13.319 limit such a big deal for more advanced users? 00:04:13.599 --> 00:04:15.520 Real -world automations often need descriptions 00:04:15.520 --> 00:04:18.199 much longer than that constraint allows. All 00:04:18.199 --> 00:04:19.779 right, let's look at the first case study, Agent 00:04:19.779 --> 00:04:23.019 1. Pretty straightforward. Monitor an email inbox, 00:04:23.300 --> 00:04:26.220 use AI to figure out a response, and draft that 00:04:26.220 --> 00:04:28.279 reply. Yeah, the kind of thing that used to be 00:04:28.279 --> 00:04:31.439 10, maybe 15 minutes of clicking around. In the 00:04:31.439 --> 00:04:33.500 test described, they gave it a single prompt. 00:04:33.680 --> 00:04:36.560 The AI picked the right Gmail trigger, something 00:04:36.560 --> 00:04:39.079 called an AI agent node, and the Gmail send node, 00:04:39.220 --> 00:04:41.899 hooked them up, popped in the credentials, took 00:04:41.899 --> 00:04:44.740 maybe 30, 40 seconds. Hold on. The AI agent node. 00:04:44.819 --> 00:04:47.439 Can you clarify what that specific Lego block 00:04:47.439 --> 00:04:50.500 does? Good catch. Yeah. So if a node is a tool, 00:04:50.660 --> 00:04:53.519 the AI agent node is basically a bridge. It lets 00:04:53.519 --> 00:04:55.480 your workflow talk directly to a large language 00:04:55.480 --> 00:04:58.560 model. Think chat GPT, Claude, models like that. 00:04:58.660 --> 00:05:01.060 So you can use it for tasks like summarizing 00:05:01.060 --> 00:05:03.860 text, writing drafts, analyzing sentiment, that 00:05:03.860 --> 00:05:05.300 kind of thing, right within your automation. 00:05:05.600 --> 00:05:07.899 Got it. Makes sense. So 40 seconds for the build. 00:05:08.259 --> 00:05:10.639 But the sources said there were still minor tweaks 00:05:10.639 --> 00:05:14.019 needed, like fixing how data flowed into one 00:05:14.019 --> 00:05:16.600 field. Even with that, the total time was apparently 00:05:16.600 --> 00:05:18.899 only around three minutes. Which is still a huge 00:05:18.899 --> 00:05:20.980 jump in productivity. And that speed really hinges 00:05:20.980 --> 00:05:22.600 on getting rid of those old friction points. 00:05:22.860 --> 00:05:26.360 The biggest one they mention. Fixing Google authentication. 00:05:26.939 --> 00:05:29.120 Oh, I remember hearing about that. The old way 00:05:29.120 --> 00:05:32.040 sounded like torture. Going into the Google Cloud 00:05:32.040 --> 00:05:35.120 console, creating Oval stuff, client ID secrets. 00:05:35.379 --> 00:05:37.699 The sources literally called it the number one 00:05:37.699 --> 00:05:41.060 reason people just gave up. Exactly. Now, it's 00:05:41.060 --> 00:05:43.589 just sign in with Google. One click. Apparently, 00:05:43.649 --> 00:05:45.589 they had to go through a pretty expensive Google 00:05:45.589 --> 00:05:48.009 audit to make that happen. It really shows they're 00:05:48.009 --> 00:05:50.529 serious about making it accessible. And alongside 00:05:50.529 --> 00:05:52.930 that, the credential management is smarter, too. 00:05:53.069 --> 00:05:55.629 The AI figures out which saved account to use. 00:05:55.790 --> 00:05:58.550 Yeah, it suggests the right one based on the 00:05:58.550 --> 00:06:01.189 node. Little things, but they add up. So how 00:06:01.189 --> 00:06:03.850 big a deal is fixing that Google authentication 00:06:03.850 --> 00:06:06.470 for getting more people to use this? It removes 00:06:06.470 --> 00:06:09.339 the single biggest hurdle. Massively democratizing 00:06:09.339 --> 00:06:12.480 access for non -technical users. Okay, let's 00:06:12.480 --> 00:06:15.600 ramp up the complexity. Agent 2, advanced calendar 00:06:15.600 --> 00:06:18.100 analysis. This thing needed to run daily, look 00:06:18.100 --> 00:06:20.339 at calendar events, figure out productivity metrics, 00:06:20.560 --> 00:06:23.939 create a custom HTML report. And then email it. 00:06:24.000 --> 00:06:26.199 Yeah, that's definitely more involved. Manually, 00:06:26.240 --> 00:06:28.720 you're easily looking at 30 to 60 minutes, maybe 00:06:28.720 --> 00:06:31.279 more if you hit snags. And right away, you hit 00:06:31.279 --> 00:06:33.279 that thousand character wall. The goal is just 00:06:33.279 --> 00:06:35.699 too detailed. So they couldn't just type it all 00:06:35.699 --> 00:06:37.839 in. They had to use this hybrid strategy. Explain 00:06:37.839 --> 00:06:40.500 that. Right. The clever workaround was using 00:06:40.500 --> 00:06:42.639 another AI. First, they mentioned Claude Sonnet 00:06:42.639 --> 00:06:45.759 4 .5 to break down the big complex goal into 00:06:45.759 --> 00:06:48.199 tiny step -by -step instructions like step one, 00:06:48.360 --> 00:06:51.379 add a calendar trigger. Step two, add a code 00:06:51.379 --> 00:06:53.839 node. then they fed those small instructions 00:06:53.839 --> 00:06:57.860 one by one into the n8n ai builder okay that 00:06:57.860 --> 00:07:00.160 is clever using one ai to manage the other it 00:07:00.160 --> 00:07:02.800 is and honestly here's a bit of a confession 00:07:02.800 --> 00:07:06.000 even doing this stuff all the time i still wrestle 00:07:06.000 --> 00:07:08.660 with prompt drift myself you give an ai a really 00:07:08.660 --> 00:07:11.600 long complex prompt sometimes it just loses the 00:07:11.600 --> 00:07:14.040 plot you know makes weird mistakes halfway through 00:07:14.040 --> 00:07:17.139 that cascade into Well, into chaos. Sounds like 00:07:17.139 --> 00:07:19.120 a nightmare to debug. It's what we call debugging 00:07:19.120 --> 00:07:21.819 hell in the notes. Seriously. Trying to find 00:07:21.819 --> 00:07:25.000 the one error in a huge 25 node workflow that 00:07:25.000 --> 00:07:28.040 the AI built all at once. No thanks. This node 00:07:28.040 --> 00:07:30.360 by node method, even if it feels slower initially, 00:07:30.560 --> 00:07:33.019 it prevents that massive headache. It's kind 00:07:33.019 --> 00:07:34.680 of admitting that sometimes breaking the problem 00:07:34.680 --> 00:07:37.579 down yourself or using a specialized AI for planning 00:07:37.579 --> 00:07:39.879 is just better. So building it piece by piece, 00:07:39.939 --> 00:07:42.300 node by node, it keeps the AI focused on one 00:07:42.300 --> 00:07:44.860 thing at a time and it lets the human check each 00:07:44.860 --> 00:07:47.279 step. Exactly. Quality control at each stage. 00:07:47.399 --> 00:07:49.259 The results they reported were pretty amazing. 00:07:49.399 --> 00:07:51.980 For this quite complex agent, the total build 00:07:51.980 --> 00:07:53.899 time was about 15 minutes, but that was while 00:07:53.899 --> 00:07:55.779 they were explaining it on video. So realistically, 00:07:56.180 --> 00:07:58.779 probably five to 10 minutes normally. Compare 00:07:58.779 --> 00:08:01.899 that to 45, maybe 90 minutes the old way. It 00:08:01.899 --> 00:08:04.819 really shows the power of combining the NAI for 00:08:04.819 --> 00:08:07.360 the structure and maybe a different AI for the 00:08:07.360 --> 00:08:09.660 really nuanced bits like code generation. So 00:08:09.660 --> 00:08:12.579 besides using an external AI for planning, what's 00:08:12.579 --> 00:08:14.459 the main advantage of building node by node? 00:08:14.540 --> 00:08:16.920 It isolates potential configuration issues to 00:08:16.920 --> 00:08:18.899 single components, making troubleshooting far 00:08:18.899 --> 00:08:22.000 easier and faster. Hashtag tag hashtag mid -roll 00:08:22.000 --> 00:08:24.660 sponsor read. Sponsor read placeholder. Which 00:08:24.660 --> 00:08:27.680 brings us to the final test. Agent 3. This one 00:08:27.680 --> 00:08:30.680 sounds wild. A complex, long to short form content 00:08:30.680 --> 00:08:33.500 factory. Fetch a video, transcribe it, analyze 00:08:33.500 --> 00:08:35.799 it for viral bits, chop it into multiple short 00:08:35.799 --> 00:08:38.799 clips, make thumbnails, A -B test titles, publish 00:08:38.799 --> 00:08:41.279 everything. We're talking 25, maybe more nodes 00:08:41.279 --> 00:08:43.679 needed here. Yeah, historically. That's not hours. 00:08:43.740 --> 00:08:45.759 That's potentially days of frustrating work, 00:08:45.840 --> 00:08:48.960 even for an expert. So for this test, they basically 00:08:48.960 --> 00:08:50.720 threw a Hail Mary. They took the description, 00:08:50.940 --> 00:08:53.340 chopped it down ruthlessly to fit exactly 1 ,000 00:08:53.340 --> 00:08:55.740 characters, and fed the whole monster prompt 00:08:55.740 --> 00:08:58.799 to the AI builder at once. The ultimate one -shot 00:08:58.799 --> 00:09:01.600 test. Oh, wow. Okay, pushing the limits. So what 00:09:01.600 --> 00:09:03.580 happened? Did it just crash and burn? This is 00:09:03.580 --> 00:09:06.679 the part that, whoa, seriously, imagine scaling 00:09:06.679 --> 00:09:09.840 this kind of capability. The NAN -AI built the 00:09:09.840 --> 00:09:13.019 entire massive 25 -node workflow, the structure, 00:09:13.179 --> 00:09:16.110 the triggers, the logic, the connections. In 00:09:16.110 --> 00:09:18.549 about 90 seconds. That's just an incredible display 00:09:18.549 --> 00:09:20.570 of technical leverage right there. 90 seconds 00:09:20.570 --> 00:09:23.389 for a 25 -step structure. That is genuinely hard 00:09:23.389 --> 00:09:25.669 to get your head around. But let's be clear on 00:09:25.669 --> 00:09:27.429 what it didn't do. The sensitive stuff, right? 00:09:27.490 --> 00:09:30.549 Like actually getting API keys or logging into 00:09:30.549 --> 00:09:32.750 external accounts. Correct. It couldn't do that, 00:09:32.830 --> 00:09:34.870 which makes sense security -wise, but it built 00:09:34.870 --> 00:09:38.610 maybe 80, 90 % of the whole thing. And it even 00:09:38.610 --> 00:09:40.889 gave specific instructions on what manual steps 00:09:40.889 --> 00:09:42.830 were left. Like, okay, now you need to go here 00:09:42.830 --> 00:09:46.049 and authenticate this account. It perfectly highlights 00:09:46.049 --> 00:09:48.929 the emerging model. Humans provide the strategy, 00:09:49.090 --> 00:09:51.629 the taste, the critical security bits like authentication. 00:09:52.129 --> 00:09:55.529 The AI provides the raw speed and does the heavy 00:09:55.529 --> 00:09:58.070 lifting of execution. And that speed, that ease, 00:09:58.389 --> 00:10:02.360 it does more than just save time. It changes 00:10:02.360 --> 00:10:05.019 the economics, doesn't it? Specifically, the 00:10:05.019 --> 00:10:07.460 threshold for when automation even makes sense. 00:10:07.700 --> 00:10:09.860 Oh, fundamentally. Think about the old way you'd 00:10:09.860 --> 00:10:11.700 decide whether to automate something. You'd think, 00:10:11.779 --> 00:10:14.100 okay, this annoying task wastes maybe two minutes 00:10:14.100 --> 00:10:16.200 a week. But building the automation will take 00:10:16.200 --> 00:10:18.500 me two hours. Not worth it. So you just kept 00:10:18.500 --> 00:10:20.659 doing the annoying task manually. Yeah. Forever. 00:10:20.820 --> 00:10:23.000 Exactly. But the new calculation is completely 00:10:23.000 --> 00:10:26.179 different. It's, okay, this task wastes two minutes, 00:10:26.200 --> 00:10:28.379 maybe even just two minutes a month. But building 00:10:28.379 --> 00:10:30.850 the automation will take me... Two minutes. Yeah, 00:10:30.889 --> 00:10:33.590 absolutely worth it. The economic barrier to 00:10:33.590 --> 00:10:35.750 automating just plummeted. It basically went 00:10:35.750 --> 00:10:37.950 to zero. It's now worthwhile to automate even 00:10:37.950 --> 00:10:40.730 the tiniest little time -wasting inefficiencies. 00:10:41.070 --> 00:10:44.190 That shift in cost -benefit seems huge. It creates 00:10:44.190 --> 00:10:46.509 a potential gap, doesn't it? Definitely. So what's 00:10:46.509 --> 00:10:48.629 the competitive implication for professionals 00:10:48.629 --> 00:10:51.690 who decide not to embrace this? They will struggle 00:10:51.690 --> 00:10:54.610 to compete with those who deploy an army of fast, 00:10:54.850 --> 00:10:58.659 always -on AI agents daily. Hashtags, tag, tag, 00:10:58.679 --> 00:11:01.379 big idea recap. So if we boil it all down, the 00:11:01.379 --> 00:11:03.240 core idea is pretty straightforward. Building 00:11:03.240 --> 00:11:05.940 these AI agents, these automations, just got 00:11:05.940 --> 00:11:08.159 something like 10 times easier. And that's massively 00:11:08.159 --> 00:11:10.679 speeding up the arrival of this agent era we 00:11:10.679 --> 00:11:12.940 keep hearing about. It puts real power into the 00:11:12.940 --> 00:11:16.019 hands of, you know, solo entrepreneurs, non -technical 00:11:16.019 --> 00:11:18.039 folks, people who were previously locked out. 00:11:18.200 --> 00:11:21.580 Yeah, the winning formula seems clear. It's human 00:11:21.580 --> 00:11:25.779 expertise. The strategy, the judgment, the final 00:11:25.779 --> 00:11:28.460 checks combined with AI execution for that incredible 00:11:28.460 --> 00:11:31.700 speed and scale, that combination just multiplies 00:11:31.700 --> 00:11:34.720 value. Technology is the ultimate lever, really. 00:11:35.200 --> 00:11:37.879 Hashtag tag, tag, tag, outro. So if you're listing 00:11:37.879 --> 00:11:40.220 and thinking, OK, I want this leverage, the practical 00:11:40.220 --> 00:11:42.500 advice from the sources seems to be don't overthink 00:11:42.500 --> 00:11:44.539 it. Just start. The biggest hurdle is inertia. 00:11:45.070 --> 00:11:47.009 Yeah, focus on building those first one or two 00:11:47.009 --> 00:11:48.850 agents. Just push through that initial learning 00:11:48.850 --> 00:11:51.190 curve. Once you do that, you've kind of conquered 00:11:51.190 --> 00:11:53.470 90 % of the friction. Keep it simple at first. 00:11:53.649 --> 00:11:55.590 Use that node -by -node method if things get 00:11:55.590 --> 00:11:58.210 complex. And always, always remember, your job 00:11:58.210 --> 00:12:01.149 is that crucial last 10%. Reviewing, refining, 00:12:01.210 --> 00:12:03.370 making sure the AI actually did what you intended. 00:12:03.610 --> 00:12:05.409 Okay, let's leave you with a final thought. Pull 00:12:05.409 --> 00:12:07.190 from the conclusion of the material we looked 00:12:07.190 --> 00:12:10.070 at. The gap between people who embrace building 00:12:10.070 --> 00:12:12.690 and deploying these kinds of fast AI agents and 00:12:12.690 --> 00:12:15.240 those who don't. that gap is likely going to 00:12:15.240 --> 00:12:18.620 widen fast, possibly becoming impossible to bridge. 00:12:19.139 --> 00:12:21.039 So the question isn't if you should automate 00:12:21.039 --> 00:12:23.039 that two -minute inefficiency, but how quickly 00:12:23.039 --> 00:12:25.259 you're going to decide to do it out to your music.