WEBVTT 00:00:00.000 --> 00:00:02.379 What happens when we stop building machines piece 00:00:02.379 --> 00:00:06.200 by piece and simply tell the machine what to 00:00:06.200 --> 00:00:09.259 build? Beat. That is the big question. Right 00:00:09.259 --> 00:00:12.320 now, in 2026, we are watching the death of drag 00:00:12.320 --> 00:00:14.839 and drop automation. We really are. It is happening 00:00:14.839 --> 00:00:17.420 incredibly fast. We are looking at a guide today. 00:00:17.579 --> 00:00:20.859 It's called The Future of Automation Beyond N8n 00:00:20.859 --> 00:00:24.199 with Claude Code. It paints a pretty wild picture. 00:00:24.460 --> 00:00:27.140 It does. It points to a reality where natural 00:00:27.140 --> 00:00:29.920 language doesn't just, you know. assist your 00:00:29.920 --> 00:00:33.020 workflows, it completely replaces the need to 00:00:33.020 --> 00:00:35.780 manually build them. Which is huge. So welcome 00:00:35.780 --> 00:00:39.039 to this deep dive. Today, our mission is to unpack 00:00:39.039 --> 00:00:41.719 this massive shift. We really need to get into 00:00:41.719 --> 00:00:43.899 the weeds on this. We are going to figure out 00:00:43.899 --> 00:00:46.520 how natural language is gutting traditional workflow 00:00:46.520 --> 00:00:48.640 building. We'll look at what this means for you, 00:00:48.679 --> 00:00:50.840 especially if you spent years learning these 00:00:50.840 --> 00:00:53.780 tools. And how to adapt your skills without getting 00:00:53.780 --> 00:00:56.170 left behind. Because it's an existential shift 00:00:56.170 --> 00:00:57.710 for a lot of builders out there. Oh, absolutely. 00:00:57.909 --> 00:00:59.670 The whole landscape is just morphing under our 00:00:59.670 --> 00:01:02.350 feet. So we need to establish the baseline first. 00:01:02.770 --> 00:01:05.890 We have to look at what actually changed in the 00:01:05.890 --> 00:01:08.310 literal interface of automation this year. Let's 00:01:08.310 --> 00:01:10.790 do it. For a long time, building automations 00:01:10.790 --> 00:01:13.870 meant, well, manually putting every single step 00:01:13.870 --> 00:01:16.650 together. Yeah, you were doing the manual labor 00:01:16.650 --> 00:01:19.310 of connecting the digital plumbing. Exactly. 00:01:19.370 --> 00:01:21.530 You were the one. Turning the wrench. Right, 00:01:21.609 --> 00:01:24.049 turning the wrench. If you used a visual tool 00:01:24.049 --> 00:01:27.629 like ANAN, you open up this massive blank canvas. 00:01:27.969 --> 00:01:30.670 Kind of intimidating, honestly. Very. You picked 00:01:30.670 --> 00:01:33.129 a specific trigger from a drop -down menu. You 00:01:33.129 --> 00:01:34.829 dragged and dropped different nodes onto the 00:01:34.829 --> 00:01:37.730 screen. You had to manually map the JSON data 00:01:37.730 --> 00:01:39.569 fields from one application to the next. Oh, 00:01:39.609 --> 00:01:42.549 the JSON mapping. It was painful. Then you clicked 00:01:42.549 --> 00:01:46.090 run. You tested every single branch. You fixed 00:01:46.090 --> 00:01:48.290 the inevitable errors. Always so many errors. 00:01:48.590 --> 00:01:50.750 And you repeated that loop until the workflow 00:01:50.750 --> 00:01:54.129 finally functioned. Beat. It's like stacking 00:01:54.129 --> 00:01:57.090 Lego blocks of data piece by piece, hoping the 00:01:57.090 --> 00:01:59.250 structure holds up. I love that analogy. That 00:01:59.250 --> 00:02:03.709 Lego analogy is spot on. And to be fair, that 00:02:03.709 --> 00:02:06.890 old method still works. No one is saying NEN 00:02:06.890 --> 00:02:09.590 is completely broken today. No, of course not. 00:02:09.750 --> 00:02:11.810 But it demands a tremendous amount of cognitive 00:02:11.810 --> 00:02:14.250 load from you. It really does. You have to juggle 00:02:14.250 --> 00:02:17.409 the high -level business logic, the specific 00:02:17.409 --> 00:02:20.509 quirks of different APIs, the exact data flow. 00:02:20.870 --> 00:02:22.969 And all the potential failure points. Exactly. 00:02:23.129 --> 00:02:24.629 You have to hold all of that in your head at 00:02:24.629 --> 00:02:28.050 the exact same time. It's exhausting. Yeah. But 00:02:28.050 --> 00:02:31.610 that is the massive shift here. The literal interface 00:02:31.610 --> 00:02:34.699 of automation. has totally transformed. We went 00:02:34.699 --> 00:02:37.060 from a visual flow chart to a terminal window. 00:02:37.280 --> 00:02:39.319 Right. With cloud code, you're starting with 00:02:39.319 --> 00:02:41.439 the outcome first. Yeah. Natural language is 00:02:41.439 --> 00:02:43.800 the new user interface. It's wild to think about. 00:02:43.939 --> 00:02:45.900 You just explain where the data comes from, what 00:02:45.900 --> 00:02:47.400 needs to happen to it, and where the results 00:02:47.400 --> 00:02:51.000 should go. Yeah. Then the agent acts as the intermediary. 00:02:51.020 --> 00:02:53.400 It handles writing the actual underlying scripts 00:02:53.400 --> 00:02:55.439 to make that happen. It takes over the translation 00:02:55.439 --> 00:02:57.800 layer. You speak English or, you know, whatever 00:02:57.800 --> 00:03:00.319 language you prefer. Right. And the agent translates 00:03:00.319 --> 00:03:03.539 your intent into the precise code required to 00:03:03.539 --> 00:03:07.099 execute the task. But wait, I have to ask. Sure. 00:03:07.199 --> 00:03:09.379 If I'm just dictating the final house to the 00:03:09.379 --> 00:03:12.680 AI, what stops it from building the foundation 00:03:12.680 --> 00:03:15.740 out of cardboard? That is a great point. Does 00:03:15.740 --> 00:03:19.750 relying on the agent to build the how? Completely 00:03:19.750 --> 00:03:22.729 remove the need for human skill. Not at all. 00:03:22.830 --> 00:03:24.969 That is a huge misconception out there. Okay. 00:03:25.110 --> 00:03:26.930 The skill hasn't disappeared. It just shifted 00:03:26.930 --> 00:03:29.990 up the stack. You moved from the manual setup 00:03:29.990 --> 00:03:32.169 of nodes to giving much better instructions. 00:03:32.330 --> 00:03:35.110 Makes sense. The AI will absolutely try to build 00:03:35.110 --> 00:03:36.930 a cardboard foundation if you aren't specific. 00:03:37.169 --> 00:03:39.729 It will definitely try. Your skill is now spotting 00:03:39.729 --> 00:03:42.710 weak logic, anticipating edge cases, and refining 00:03:42.710 --> 00:03:45.389 the outputs. Because that first automated version 00:03:45.389 --> 00:03:48.090 inevitably falls short. Almost always. So we 00:03:48.090 --> 00:03:50.050 trade manual building for high -level system 00:03:50.050 --> 00:03:52.530 design and logic checking. Exactly. You are the 00:03:52.530 --> 00:03:55.229 general contractor now, not the bricklayer. To 00:03:55.229 --> 00:03:58.509 sex silence. And that brings us to why this shift 00:03:58.509 --> 00:04:00.949 matters more than most people realize. It's a 00:04:00.949 --> 00:04:03.990 huge deal. A lot of tech trends sound revolutionary, 00:04:04.110 --> 00:04:06.229 but they just fizzle out. Like the metaverse. 00:04:06.449 --> 00:04:08.930 Right, like the metaverse. But this one is different. 00:04:09.069 --> 00:04:12.129 The market forces behind it are incredibly aggressive. 00:04:12.680 --> 00:04:14.520 The source notes are really staggering prediction 00:04:14.520 --> 00:04:18.279 here. Yeah, let's hear it. By 2027, 50 % of enterprises 00:04:18.279 --> 00:04:21.079 are expected to adopt these AI -built systems. 00:04:21.360 --> 00:04:25.259 50%. That is a massive, highly accelerated adoption 00:04:25.259 --> 00:04:27.720 rate for large companies. Because the market 00:04:27.720 --> 00:04:32.220 demands raw speed, businesses want faster systems. 00:04:32.480 --> 00:04:34.560 They want drastically lower build time. Exactly. 00:04:34.579 --> 00:04:37.459 They want less manual, busy work standing between 00:04:37.459 --> 00:04:40.379 a new idea and its actual execution. Which makes 00:04:40.379 --> 00:04:42.879 sense. That is why Claude Co - automation fits 00:04:42.879 --> 00:04:45.920 so perfectly into the enterprise puzzle. It matches 00:04:45.920 --> 00:04:48.000 where the market is already desperately heading. 00:04:48.139 --> 00:04:51.060 The value shift here is profound. Let's trace 00:04:51.060 --> 00:04:53.740 it back a bit. A few years ago, the entire focus 00:04:53.740 --> 00:04:56.040 was just on chatbots. We just wanted to talk 00:04:56.040 --> 00:04:58.360 to the AI. Just typing in a prompt box. Right. 00:04:58.439 --> 00:05:02.199 Then we moved to injecting AI inside existing 00:05:02.199 --> 00:05:04.699 workflow tools. Like putting a smart filter in 00:05:04.699 --> 00:05:07.339 the middle of a process? Yeah. People build visual 00:05:07.339 --> 00:05:10.730 systems to summarize long content. or to automatically 00:05:10.730 --> 00:05:13.410 route customer support tickets based on sentiment. 00:05:13.550 --> 00:05:16.170 Yeah, the perceived value back then was just 00:05:16.170 --> 00:05:18.649 connecting an AI model to an existing workflow. 00:05:18.990 --> 00:05:21.129 Exactly. You were still building the pipes manually, 00:05:21.209 --> 00:05:23.670 but you put a smart filter in the middle. But 00:05:23.670 --> 00:05:26.670 now the value is using AI to create the entire 00:05:26.670 --> 00:05:30.079 workflow itself. The AI. is the pipe builder 00:05:30.079 --> 00:05:32.439 i can imagine traditional automation builders 00:05:32.439 --> 00:05:34.819 feeling a real sense of discomfort right now 00:05:34.819 --> 00:05:36.920 oh definitely maybe even a bit of existential 00:05:36.920 --> 00:05:39.540 dread i've heard it from a lot of them you spend 00:05:39.540 --> 00:05:42.379 three years learning all the little details and 00:05:42.379 --> 00:05:45.519 quirks of a tool like n8n all those specific 00:05:45.519 --> 00:05:49.199 api nodes and suddenly that specific highly paid 00:05:49.199 --> 00:05:52.339 busy work is becoming outdated it's a genuinely 00:05:52.339 --> 00:05:54.839 tough pill to swallow you feel like your hard 00:05:54.839 --> 00:05:57.579 -earned knowledge is depreciating overnight but 00:05:57.579 --> 00:06:00.399 is it No, that time wasn't wasted at all. The 00:06:00.399 --> 00:06:03.519 fundamental job is just changing. Your professional 00:06:03.519 --> 00:06:05.839 value used to come from knowing how to manually 00:06:05.839 --> 00:06:08.139 build the machine. Now it comes from knowing 00:06:08.139 --> 00:06:10.500 which machine actually needs to be built. That's 00:06:10.500 --> 00:06:12.819 a good way to put it. You need to know how it 00:06:12.819 --> 00:06:16.800 should behave in the real world and how to improve 00:06:16.800 --> 00:06:19.839 it when business needs change. But with this 00:06:19.839 --> 00:06:22.339 shift happening so fast, I have to ask. Go for 00:06:22.339 --> 00:06:26.839 it. Is knowing N80 inside and out actually a 00:06:26.839 --> 00:06:30.480 liability now? Does it lock you into old ways 00:06:30.480 --> 00:06:33.500 of thinking? It's not a liability to know the 00:06:33.500 --> 00:06:36.800 tool deeply, but getting stuck in that specific 00:06:36.800 --> 00:06:39.860 interface is a trap. You have to adapt. You absolutely 00:06:39.860 --> 00:06:42.560 have to evolve. Your job title might be the same, 00:06:42.680 --> 00:06:45.180 but your function is evolving. You are going 00:06:45.180 --> 00:06:47.199 from being a mechanic to being an architect. 00:06:47.660 --> 00:06:49.819 A mechanic turns the wrench. And an architect 00:06:49.819 --> 00:06:52.300 designs the flow of the entire building. Your 00:06:52.300 --> 00:06:54.839 core value is now understanding business outcomes, 00:06:55.060 --> 00:06:57.720 not just connecting tools. Perfectly said. You 00:06:57.720 --> 00:07:00.240 stay valuable when you deeply understand workflow 00:07:00.240 --> 00:07:03.199 logic. Because tools change. Tools and interfaces 00:07:03.199 --> 00:07:06.339 will always change. Systems thinking does not 00:07:06.339 --> 00:07:08.839 change. Okay, let's pull this out of the clouds. 00:07:09.220 --> 00:07:11.220 We need to ground this high -level philosophy 00:07:11.220 --> 00:07:13.779 into something real. Let's do it. Let's look 00:07:13.779 --> 00:07:16.779 at a concrete, practical comparison of how these 00:07:16.779 --> 00:07:18.839 two methods actually work under the hood. The 00:07:18.839 --> 00:07:21.459 source breaks this down really well. It contrasts 00:07:21.459 --> 00:07:24.620 the traditional N8n step -by -step method. with 00:07:24.620 --> 00:07:27.100 Claude Code's five steps. This is where the rubber 00:07:27.100 --> 00:07:29.779 meets the road. In NANN, you always start with 00:07:29.779 --> 00:07:32.500 the individual workflow steps. You drag in a 00:07:32.500 --> 00:07:35.160 webhook to act as a trigger. You connect an HTTP 00:07:35.160 --> 00:07:38.459 request node. You set up a specific API call, 00:07:38.600 --> 00:07:41.540 which usually means digging through documentation. 00:07:42.040 --> 00:07:45.220 Hours of reading docs. You manually map the data 00:07:45.220 --> 00:07:47.879 fields. You add conditional filters. And you 00:07:47.879 --> 00:07:50.939 test each and every path. It's incredibly powerful, 00:07:51.120 --> 00:07:54.560 but it takes serious, focused time. Contrast 00:07:54.560 --> 00:07:57.399 that with cloud code. Here, you start entirely 00:07:57.399 --> 00:07:59.600 from the end result. The five steps are very 00:07:59.600 --> 00:08:01.420 distinct. Why are they? Step one, you define 00:08:01.420 --> 00:08:04.519 the outcome. Step two, you provide the data source. 00:08:04.779 --> 00:08:08.120 Step three, you explain the process. Step four, 00:08:08.259 --> 00:08:10.860 you designate the destination. And step five, 00:08:11.000 --> 00:08:13.279 you test it. Let's use the specific YouTube example 00:08:13.279 --> 00:08:15.600 from the source material. It's a classic automation 00:08:15.600 --> 00:08:19.139 use case. It really is. The goal is simple. You 00:08:19.139 --> 00:08:21.319 want to check a specific YouTube channel every 00:08:21.319 --> 00:08:24.300 eight hours. You want to find any new videos, 00:08:24.420 --> 00:08:26.740 grab the transcript, summarize that transcript, 00:08:26.879 --> 00:08:28.459 and send it to your own Slack channel or email. 00:08:28.660 --> 00:08:30.680 Pretty standard request. If I were building that 00:08:30.680 --> 00:08:32.759 the old way, that's at least four or five complex 00:08:32.759 --> 00:08:36.539 nodes. Plus, dealing with YouTube's messy API 00:08:36.539 --> 00:08:39.340 pagination. Which is always a headache. But in 00:08:39.340 --> 00:08:41.559 the new paradigm, step one is just writing that 00:08:41.559 --> 00:08:44.100 outcome out clearly in plain English. Step two, 00:08:44.200 --> 00:08:47.529 you add the required context. You just give the 00:08:47.529 --> 00:08:49.669 system the raw YouTube channel link. And tell 00:08:49.669 --> 00:08:52.049 it to send the final result to your specific 00:08:52.049 --> 00:08:54.710 Slack workspace. The clearer you are, the more 00:08:54.710 --> 00:08:57.110 robust the code it writes will be. Then step 00:08:57.110 --> 00:08:59.929 three is where the heavy lifting happens. You 00:08:59.929 --> 00:09:02.830 explain the process and Cloud Code actually generates 00:09:02.830 --> 00:09:05.690 the workflow script. It figures out how to authenticate 00:09:05.690 --> 00:09:08.210 and check for new videos. It handles the processing 00:09:08.210 --> 00:09:10.409 of the transcript data. But hold on, I need to 00:09:10.409 --> 00:09:12.649 push back on this magic happening in step three. 00:09:12.830 --> 00:09:15.210 Okay, push back. How does the agent actually 00:09:15.210 --> 00:09:19.519 know YouTube? specific API rate limits? Or the 00:09:19.519 --> 00:09:22.379 exact endpoint to hit? Good question. Does it 00:09:22.379 --> 00:09:25.259 just guess based on its training data? Or does 00:09:25.259 --> 00:09:27.639 it research the documentation live? It's a mix 00:09:27.639 --> 00:09:29.580 of both. And that is where it gets fascinating. 00:09:29.840 --> 00:09:32.559 Oh, so. It relies on its massive pre -training 00:09:32.559 --> 00:09:35.960 to know standard API structures. But if it's 00:09:35.960 --> 00:09:39.299 a newer tool or it hits an error, clod code can 00:09:39.299 --> 00:09:42.659 act like a terminal agent. Really? Yeah. It can 00:09:42.659 --> 00:09:45.620 actually execute curl commands to test endpoints 00:09:45.620 --> 00:09:49.240 live. It reads the error message and rewrites 00:09:49.240 --> 00:09:51.659 its own script to fix the authentication issue. 00:09:51.860 --> 00:09:54.240 It essentially debugs its own initial assumptions. 00:09:54.480 --> 00:09:57.620 Exactly. That is wild. But what about the logic 00:09:57.620 --> 00:10:00.299 gaps? What do you mean? For instance, how does 00:10:00.299 --> 00:10:03.679 the agent handle edge cases, like avoiding summarizing 00:10:03.679 --> 00:10:06.000 the exact same video twice when it runs eight 00:10:06.000 --> 00:10:08.600 hours later? That's the real test of an AI builder 00:10:08.600 --> 00:10:11.159 right there. The agent usually figures out the 00:10:11.159 --> 00:10:13.639 initial deduplication logic on its own. Like 00:10:13.639 --> 00:10:15.679 setting up a database. Yeah, instead of you having 00:10:15.679 --> 00:10:18.340 to wire up a complex database node, Claude might 00:10:18.340 --> 00:10:20.740 just spin up a lightweight SQLite database right 00:10:20.740 --> 00:10:23.100 there on your local machine to log previous video 00:10:23.100 --> 00:10:25.860 IDs. But here is the catch. There's always a 00:10:25.860 --> 00:10:29.360 catch. the human must carefully review and test 00:10:29.360 --> 00:10:32.980 that specific logic during step four. You can 00:10:32.980 --> 00:10:36.120 never blindly trust that it handled the edge 00:10:36.120 --> 00:10:38.519 cases correctly. Never. The agent writes the 00:10:38.519 --> 00:10:41.620 deduplication logic, but you must verify it works. 00:10:42.119 --> 00:10:44.559 Always. Agents are brilliant, but they still 00:10:44.559 --> 00:10:47.159 make bizarre mistakes. That is exactly why Step 00:10:47.159 --> 00:10:50.360 5 is iterating and improving. You guide the system. 00:10:50.539 --> 00:10:52.600 You look at the output and say, hey, this summary 00:10:52.600 --> 00:10:55.399 is too long, make it bullet points. Or, you missed 00:10:55.399 --> 00:10:58.059 a video, check your timestamp logic. You iterate 00:10:58.059 --> 00:11:00.620 together. Exactly. We are going to pause for 00:11:00.620 --> 00:11:02.820 a quick break. When we come back, we need to 00:11:02.820 --> 00:11:04.899 talk about exactly where this new method breaks 00:11:04.899 --> 00:11:07.360 down. Because building faster definitely means 00:11:07.360 --> 00:11:10.799 we can break things faster. Beat. This Deep Dive 00:11:10.799 --> 00:11:13.820 is brought to you by CloudScale Solutions. Are 00:11:13.820 --> 00:11:16.100 you struggling to manage your growing data infrastructure? 00:11:16.519 --> 00:11:19.080 CloudScale offers seamless, secure, and scalable 00:11:19.080 --> 00:11:21.179 cloud hosting tailored for modern enterprises. 00:11:21.639 --> 00:11:23.840 Their intuitive dashboard lets you monitor performance 00:11:23.840 --> 00:11:26.720 in real time. Plus, their dedicated support team 00:11:26.720 --> 00:11:30.419 is available 24 -7. Stop worrying about server 00:11:30.419 --> 00:11:33.240 uptime. Start focusing on your business. Visit 00:11:33.240 --> 00:11:36.340 CloudScaleSolutions .com today. Learn more and 00:11:36.340 --> 00:11:39.590 secure your digital future. And we're back. Taking 00:11:39.590 --> 00:11:41.470 the human out of the manual building process 00:11:41.470 --> 00:11:44.450 introduces a scary variable. It really does. 00:11:44.690 --> 00:11:47.690 If I am just dictating the final house, what 00:11:47.690 --> 00:11:51.110 happens if the AI decides to cut corners? We 00:11:51.110 --> 00:11:53.590 need to transition to the friction points. Cloud 00:11:53.590 --> 00:11:56.850 code automation is undeniably powerful, but the 00:11:56.850 --> 00:11:59.889 source highlights major pitfalls. There are four 00:11:59.889 --> 00:12:02.110 main pitfalls that will absolutely ruin your 00:12:02.110 --> 00:12:04.230 day if you aren't careful. The first is unclear 00:12:04.230 --> 00:12:07.090 prompts. It is the classic garbage in, garbage 00:12:07.090 --> 00:12:09.559 out problem, just scaled up. If your initial 00:12:09.559 --> 00:12:12.340 instructions are vague, your resulting workflow 00:12:12.340 --> 00:12:15.519 will be fragile and vague. You have to be ruthlessly 00:12:15.519 --> 00:12:17.980 specific about the exact triggers. The exact 00:12:17.980 --> 00:12:20.080 shape of the data. And the precise formatting 00:12:20.080 --> 00:12:22.639 of the outputs. The second massive problem is 00:12:22.639 --> 00:12:25.399 hallucinations. We see this constantly in the 00:12:25.399 --> 00:12:28.299 broader AI space, but it's really dangerous here. 00:12:28.480 --> 00:12:31.039 Let me define that really quick. It's AI generating 00:12:31.039 --> 00:12:33.940 false information or code that looks completely 00:12:33.940 --> 00:12:37.750 real. Nailed it! That's exactly it. And it's 00:12:37.750 --> 00:12:41.610 insidious. The Python script or the API call 00:12:41.610 --> 00:12:44.509 the cloud code generates might look absolutely 00:12:44.509 --> 00:12:47.210 perfect to the human eye. It uses the right syntax, 00:12:47.289 --> 00:12:49.730 the right variable names. But it simply doesn't 00:12:49.730 --> 00:12:51.590 work when you actually run it because the endpoint 00:12:51.590 --> 00:12:54.389 it made up doesn't exist. That is why you must 00:12:54.389 --> 00:12:57.769 test the workflow with real, messy, real -world 00:12:57.769 --> 00:13:01.379 data. You cannot just trust clean looking terminal 00:13:01.379 --> 00:13:04.440 output. The third pitfall is context drift. This 00:13:04.440 --> 00:13:06.960 happens constantly in longer building sessions. 00:13:07.240 --> 00:13:09.539 The agent simply forgets earlier rules you established. 00:13:10.080 --> 00:13:12.580 or it randomly changes a piece of logic that 00:13:12.580 --> 00:13:15.220 you both already fixed 10 minutes ago. It's so 00:13:15.220 --> 00:13:17.700 annoying. I'll make a vulnerable admission here. 00:13:17.899 --> 00:13:20.580 I still wrestle with context drift myself when 00:13:20.580 --> 00:13:22.539 the agent forgets my earlier rules. Oh, I've 00:13:22.539 --> 00:13:24.340 been there. It happens to literally everyone. 00:13:24.620 --> 00:13:26.500 I will spend an hour refining a script with an 00:13:26.500 --> 00:13:28.740 agent, and suddenly it forgets my earlier formatting 00:13:28.740 --> 00:13:31.840 rules and breaks the whole pipeline. It is incredibly 00:13:31.840 --> 00:13:34.039 frustrating. It really is. It is a limitation 00:13:34.039 --> 00:13:36.179 of how these models work. As the conversation 00:13:36.179 --> 00:13:38.740 gets longer, the context window fills up. The 00:13:38.740 --> 00:13:41.299 AI's attention mechanism gets diluted and it 00:13:41.299 --> 00:13:43.740 starts dropping older constraints. So let's fix. 00:13:43.940 --> 00:13:45.960 The best way to fight it is keeping your individual 00:13:45.960 --> 00:13:49.039 tasks very small. You have to build one tiny 00:13:49.039 --> 00:13:51.580 workflow component at a time. And if the project 00:13:51.580 --> 00:13:54.299 gets too big? You need to stop. Ask the agent 00:13:54.299 --> 00:13:56.960 to summarize all past decisions into a text file. 00:13:57.639 --> 00:14:00.639 Then start a fresh session with that summary 00:14:00.639 --> 00:14:03.460 as the system prompt. That is a great tactical 00:14:03.460 --> 00:14:05.580 tip. Thanks. The fourth problem they outline 00:14:05.580 --> 00:14:08.340 is poor scoping. Sometimes the agent gets overly 00:14:08.340 --> 00:14:11.360 ambitious. It builds way too much complex architecture 00:14:11.360 --> 00:14:14.059 for a simple task. Like setting up a massive 00:14:14.059 --> 00:14:16.340 database when a text file would do. Exactly. 00:14:16.779 --> 00:14:19.519 Other times it builds too little and misses obvious 00:14:19.519 --> 00:14:22.220 error handling. So how do we prevent the system 00:14:22.220 --> 00:14:24.240 from overcomplicating the build from the start? 00:14:24.440 --> 00:14:26.200 It comes back to being a stripped architect. 00:14:26.340 --> 00:14:28.700 You have to keep the initial tasks extremely 00:14:28.700 --> 00:14:30.759 small. You have to set hard boundaries early 00:14:30.759 --> 00:14:33.639 in the prompt. Explicitly tell it, keep this 00:14:33.639 --> 00:14:37.909 workflow under 50 lines of code. Or only use 00:14:37.909 --> 00:14:40.610 these three specific tools. You have to rein 00:14:40.610 --> 00:14:43.610 in its tendency to over -engineer. Build one 00:14:43.610 --> 00:14:46.809 single verifiable workflow at a time. Keep the 00:14:46.809 --> 00:14:49.470 scope tight and set strict boundaries right from 00:14:49.470 --> 00:14:52.450 the start. Yes. You are the director. You provide 00:14:52.450 --> 00:14:54.710 the guardrails or the system will run off the 00:14:54.710 --> 00:14:57.929 tracks entirely. Two secs silence. Let's bring 00:14:57.929 --> 00:14:59.870 this all together. I know some listeners might 00:14:59.870 --> 00:15:02.129 feel a bit overwhelmed by this paradigm shift. 00:15:02.480 --> 00:15:04.639 It's a lot to process. But the source makes a 00:15:04.639 --> 00:15:07.700 deeply reassuring point. If you took the time 00:15:07.700 --> 00:15:10.340 to learn tools like NA and or make over the last 00:15:10.340 --> 00:15:13.700 few years, your foundational knowledge is actually 00:15:13.700 --> 00:15:16.960 a massive superpower in this new era. It truly 00:15:16.960 --> 00:15:19.590 is a superpower. Those traditional visual tools 00:15:19.590 --> 00:15:21.970 forced you to learn how automation actually works 00:15:21.970 --> 00:15:23.889 under the hood. You inherently understand what 00:15:23.889 --> 00:15:26.169 a webhook trigger is. You understand the fragility 00:15:26.169 --> 00:15:28.950 of data flow from one API to another. You understand 00:15:28.950 --> 00:15:31.769 edge cases and why robust error handling is non 00:15:31.769 --> 00:15:34.289 -negotiable. You know exactly how a massive system 00:15:34.289 --> 00:15:37.289 breaks when one tiny step fails. You know that 00:15:37.289 --> 00:15:40.490 APIs frequently time out, or that JSON payloads 00:15:40.490 --> 00:15:42.830 unexpectedly change their formatting. That experience 00:15:42.830 --> 00:15:45.929 gives you a crucial debugging mindset. When Cloud 00:15:45.929 --> 00:15:48.730 Code writes a script that fails, You aren't helpless. 00:15:49.009 --> 00:15:52.230 Right. You know why it failed. Which means you 00:15:52.230 --> 00:15:54.529 know exactly how to instruct the agent to fix 00:15:54.529 --> 00:15:57.289 it. That knowledge transfers directly. Exactly. 00:15:57.590 --> 00:15:59.990 And when you combine that deep debugging knowledge 00:15:59.990 --> 00:16:02.649 with the speed of an AI agent, the potential 00:16:02.649 --> 00:16:06.870 is staggering. Whoa, imagine scaling to build 00:16:06.870 --> 00:16:09.590 100 complex pipelines in an afternoon just by 00:16:09.590 --> 00:16:12.230 guiding the system. It's incredible to even think 00:16:12.230 --> 00:16:14.330 about. But you can only achieve that kind of 00:16:14.330 --> 00:16:17.490 scale safely if you have that paranoid, rigorous 00:16:17.490 --> 00:16:20.330 debugging mindset built in. The new workflow 00:16:20.330 --> 00:16:22.850 is fundamentally about thinking in systems. You 00:16:22.850 --> 00:16:25.009 spend your time reviewing the logic rather than 00:16:25.009 --> 00:16:26.909 just looking at the output. You spend your energy 00:16:26.909 --> 00:16:29.110 testing the weird edge cases. You constantly 00:16:29.110 --> 00:16:31.629 improve the underlying structure over time. So 00:16:31.629 --> 00:16:33.149 for the person listening right now, what is the 00:16:33.149 --> 00:16:35.350 ultimate first step a listener should take today? 00:16:35.740 --> 00:16:38.500 First, do not skip learning the automation basics. 00:16:38.840 --> 00:16:42.360 Right. Understand webhooks and APIs. Then, start 00:16:42.360 --> 00:16:45.139 with one very small, low -stakes cloud code task. 00:16:45.379 --> 00:16:47.799 Maybe automate a simple daily email. From there, 00:16:48.000 --> 00:16:50.840 rigorously practice testing and refining your 00:16:50.840 --> 00:16:53.220 natural language instructions. Do not try to 00:16:53.220 --> 00:16:55.320 build a massive, complex enterprise system on 00:16:55.320 --> 00:16:58.379 day one. Learn how the agent thinks first. Master 00:16:58.379 --> 00:17:01.519 the basics, start small with AI, and practice 00:17:01.519 --> 00:17:04.619 rigorous testing. Precisely. That is the only 00:17:04.619 --> 00:17:07.839 reliable path. That is how you move your automations 00:17:07.839 --> 00:17:09.859 from just saying it works to saying it works 00:17:09.859 --> 00:17:13.140 reliably every single time. Let's recap the big 00:17:13.140 --> 00:17:15.619 idea we've explored today. Automation has truly 00:17:15.619 --> 00:17:18.920 grown up. It really has. The major shift in 2026 00:17:18.920 --> 00:17:21.220 is moving aggressively away from knowing how 00:17:21.220 --> 00:17:24.299 to use a specific, highly visual tool. The new 00:17:24.299 --> 00:17:26.960 era of automation is entirely about how to design 00:17:26.960 --> 00:17:29.880 a resilient system. and how to clearly communicate 00:17:29.880 --> 00:17:32.660 a business outcome. Your strategic thinking matters 00:17:32.660 --> 00:17:34.819 significantly more than the specific interface 00:17:34.819 --> 00:17:37.079 you happen to be typing into. The interfaces 00:17:37.079 --> 00:17:39.039 are going to keep changing faster and faster. 00:17:39.240 --> 00:17:41.960 Your ability to think in abstract systems is 00:17:41.960 --> 00:17:43.960 the only thing that keeps you relevant. So my 00:17:43.960 --> 00:17:46.579 challenge to you is this. Start thinking in outcomes 00:17:46.579 --> 00:17:49.140 rather than individual steps. Yeah, ditch the 00:17:49.140 --> 00:17:51.619 blank canvas. The next time you need to automate 00:17:51.619 --> 00:17:54.640 a tedious task, do not jump straight into a visual 00:17:54.640 --> 00:17:57.759 tool and start connecting nodes. Sit back. and 00:17:57.759 --> 00:17:59.900 write down exactly what you want to achieve first 00:17:59.900 --> 00:18:04.160 in plain text. But this also leaves us with a 00:18:04.160 --> 00:18:06.539 much deeper question to consider as we watch 00:18:06.539 --> 00:18:08.900 this technology accelerate. Always more questions. 00:18:09.200 --> 00:18:11.640 If natural language allows us to build incredibly 00:18:11.640 --> 00:18:14.759 complex systems instantly today, what happens 00:18:14.759 --> 00:18:17.119 when these agents stop waiting for our prompts? 00:18:17.380 --> 00:18:20.019 That's a slightly terrifying thought. What happens 00:18:20.019 --> 00:18:22.660 to the architect when the machine starts proactively 00:18:22.660 --> 00:18:25.440 suggesting and building its own workflow optimizations 00:18:25.440 --> 00:18:28.259 before we even realize a process is broken? That 00:18:28.259 --> 00:18:30.380 is the next frontier, and it's coming faster 00:18:30.380 --> 00:18:32.380 than we think. Keep questioning. the process. 00:18:32.619 --> 00:18:33.180 Talk soon.