WEBVTT 1 00:00:00.241 --> 00:00:01.789 Ross Mike, welcome back to the pod. 2 00:00:01.829 --> 00:00:03.920 By the end of this episode, what are people going to learn? 3 00:00:04.340 --> 00:00:08.558 I hope I'm going to share some wisdom on how you can use the agents better. 4 00:00:08.618 --> 00:00:09.100 There's a lot 5 00:00:09.160 --> 00:00:11.048 of information going on right now. 6 00:00:11.651 --> 00:00:13.940 I disagree with most of it, and that's what we're going 7 00:00:13.940 --> 00:00:14.463 to talk about. 8 00:00:14.503 --> 00:00:17.879 So at the end, whether you're building something, using an agent for some 9 00:00:18.100 --> 00:00:20.409 sort of work, you have the best output possible. 10 00:00:20.630 --> 00:00:23.180 And is this going to be a technical dive or, 11 00:00:23.622 --> 00:00:25.149 you know, non-technical person can... 12 00:00:25.592 --> 00:00:26.476 Anyone can watch this. 13 00:00:26.496 --> 00:00:27.220 There's going to be a lot of 14 00:00:27.260 --> 00:00:33.020 diagrams that's all you're gonna make it clear to understand the concepts right easy okay basics 15 00:00:33.160 --> 00:00:46.920 let's go so the first thing that i want to announce previous episodes we probably disagree 16 00:00:47.020 --> 00:00:52.700 with this point but now what's true is the models are good the models are exceptionally good opus 17 00:00:52.780 --> 00:00:53.965 4.6 is amazing. 18 00:00:54.005 --> 00:00:55.390 GPT 5.4 is amazing. 19 00:00:55.731 --> 00:00:58.240 I know there's like two sets of camp where, especially when it 20 00:00:58.260 --> 00:01:01.433 comes to programming, people are like, oh, Opus is the better UI designer. 21 00:01:01.735 --> 00:01:02.679 GPT 5.4 is a better 22 00:01:03.080 --> 00:01:03.542 backend. 23 00:01:04.285 --> 00:01:08.240 Generally speaking, we've reached a point, we're not at AGI yet, where we reached a point 24 00:01:08.280 --> 00:01:11.711 where the models are good, but context still matters. 25 00:01:12.353 --> 00:01:14.600 And you have the power to steer the 26 00:01:14.680 --> 00:01:17.953 models in a direction where you can get quality or you can get slop. 27 00:01:18.615 --> 00:01:19.820 And that's what I really want 28 00:01:19.840 --> 00:01:20.463 to talk about. 29 00:01:21.106 --> 00:01:24.040 But before we get into all that, and feel free to cut me off because this topic 30 00:01:24.160 --> 00:01:27.749 excites me, we need to learn how context works. 31 00:01:28.471 --> 00:01:32.420 And context is the model assembling information 32 00:01:32.761 --> 00:01:34.348 that it needs to execute an action. 33 00:01:34.890 --> 00:01:37.400 And the way the context is assembled, let's say in a coding 34 00:01:37.500 --> 00:01:42.920 agent, but really in any sort of agent, is there's this general system prompt, usually by the model 35 00:01:43.040 --> 00:01:43.462 provider. 36 00:01:43.482 --> 00:01:46.073 So for example, cloud code leaked recently. 37 00:01:46.575 --> 00:01:47.740 And one of the cool things that, 38 00:01:48.443 --> 00:01:51.094 especially as a developer, I got to do is I got to read the system prompt. 39 00:01:51.456 --> 00:01:52.360 So they have this general 40 00:01:52.501 --> 00:01:56.175 system prompt that guides the model on how to act, what to do, what not to do. 41 00:01:56.637 --> 00:01:57.360 The system prompt is 42 00:01:57.400 --> 00:01:58.002 very important. 43 00:01:58.564 --> 00:02:02.658 And then you have a lot of people have agent.md files or cloud.md files. 44 00:02:02.999 --> 00:02:03.180 Now I'm 45 00:02:03.180 --> 00:02:06.691 just going to say off rip, 95% of people don't need this. 46 00:02:07.413 --> 00:02:09.500 The reason being is, again, you have 47 00:02:09.580 --> 00:02:12.430 to assume that the models are already good, right? 48 00:02:12.711 --> 00:02:15.320 Now, imagine I told you, Greg, every time we're 49 00:02:15.340 --> 00:02:17.971 about to shoot a podcast, Greg, you need a microphone. 50 00:02:19.256 --> 00:02:20.120 You know you need a microphone, 51 00:02:20.220 --> 00:02:20.340 right? 52 00:02:20.361 --> 00:02:21.946 You've done this plenty of times, right? 53 00:02:21.987 --> 00:02:25.439 So if I'm building, like, let's say a website with 54 00:02:26.543 --> 00:02:30.214 Cloud Code, and I'm telling Cloud Code, this code base uses React. 55 00:02:30.856 --> 00:02:32.120 I don't need to, because it has 56 00:02:32.240 --> 00:02:33.384 the code base in context. 57 00:02:33.445 --> 00:02:34.729 It can check the code, right? 58 00:02:35.091 --> 00:02:37.600 So there is this disparity where a lot of 59 00:02:37.700 --> 00:02:44.100 people are putting a lot of onus on the harness and the context building, and I'm low-key starting 60 00:02:44.181 --> 00:02:45.179 to strip things off. 61 00:02:45.580 --> 00:02:47.240 Like I'm going super, super minimal 62 00:02:47.683 --> 00:02:50.280 because again, not to sound like an anthropic 63 00:02:50.321 --> 00:02:51.160 or open AI shill, 64 00:02:51.240 --> 00:02:52.700 unfortunately I have not been acquired. 65 00:02:53.160 --> 00:02:54.219 None of them are paying me, 66 00:02:55.247 --> 00:02:56.440 but the models are really, really good. 67 00:02:57.242 --> 00:02:58.535 Wait, so 95% of the time, 68 00:02:59.240 --> 00:03:01.980 I don't even need to bother with an agent MD file? 69 00:03:02.141 --> 00:03:02.639 You don't. 70 00:03:02.800 --> 00:03:05.780 Like unless this is some sort of proprietary information. 71 00:03:05.880 --> 00:03:08.560 Yeah, what is the 5% of time I should care about it? 72 00:03:08.720 --> 00:03:10.520 Proprietary information that like 73 00:03:10.641 --> 00:03:12.100 may be specific to your company 74 00:03:12.282 --> 00:03:14.179 or some methodology that is specific to you 75 00:03:14.663 --> 00:03:17.580 that has to be referenced in every single conversation. 76 00:03:17.620 --> 00:03:19.860 Because the annoying part with an agent.md file 77 00:03:20.304 --> 00:03:22.260 is every time you go back and forth with the agent, 78 00:03:22.743 --> 00:03:24.640 it's added in the context, right? 79 00:03:25.141 --> 00:03:26.280 The cool thing about skills, 80 00:03:26.340 --> 00:03:28.020 and I'm gonna talk about skills in a second, 81 00:03:28.441 --> 00:03:29.720 the way skills are designed, 82 00:03:30.300 --> 00:03:33.120 the skills are used in a way 83 00:03:33.161 --> 00:03:34.700 that's called progressive disclosure. 84 00:03:34.880 --> 00:03:36.580 Meaning when you have a skill file, 85 00:03:36.843 --> 00:03:38.520 the entire thing is an additive context. 86 00:03:38.980 --> 00:03:40.520 It's just the title and the description. 87 00:03:41.160 --> 00:03:44.560 So the agent has the title and description in the context. 88 00:03:45.020 --> 00:03:48.240 And when you, let's say you have a Notion report skill, right? 89 00:03:48.341 --> 00:03:51.320 And you tell your agent, hey, I want you to create a Notion report. 90 00:03:51.921 --> 00:03:55.320 It's then going to check its context and be like, oh, I have this skill. 91 00:03:55.781 --> 00:03:57.700 Let me check out the entire document. 92 00:03:58.221 --> 00:03:59.900 So it's not in the context. 93 00:04:00.160 --> 00:04:02.720 What's in the context is the name and the description. 94 00:04:03.202 --> 00:04:05.940 But that's enough for the agent to be like, oh, this is a skill I need. 95 00:04:06.243 --> 00:04:08.220 Let me go use it, which is fantastic. 96 00:04:08.502 --> 00:04:09.560 I'm a skills maxi. 97 00:04:09.841 --> 00:04:13.540 And I'm going to show later in the episode, like how you craft the perfect skills. 98 00:04:14.041 --> 00:04:19.480 But with agent.md and cloud.md files, it's context being added at every turn, right? 99 00:04:19.781 --> 00:04:23.219 So let's say you have like a thousand line file, cloud.md. 100 00:04:24.043 --> 00:04:25.276 And let's say that's like 7,000 tokens. 101 00:04:26.302 --> 00:04:28.900 You're spending 7,000 tokens on every run. 102 00:04:29.261 --> 00:04:30.840 Now, do you need to? 103 00:04:31.404 --> 00:04:32.180 Most likely not. 104 00:04:32.220 --> 00:04:33.480 It probably should be a skill. 105 00:04:33.580 --> 00:04:42.480 But if you have some sort of company proprietary information or like there's something specific that you do that the model needs to know at every single turn, then you use it. 106 00:04:42.900 --> 00:04:45.560 The thing is, 95% of people don't have that. 107 00:04:46.200 --> 00:04:46.337 Right. 108 00:04:46.501 --> 00:04:48.939 So I'm not a fan unless that's the case. 109 00:04:50.404 --> 00:04:52.660 So and the reason being is we're wasting tokens. 110 00:04:52.851 --> 00:04:53.020 Right. 111 00:04:53.181 --> 00:04:55.520 It's in every single turn. 112 00:04:56.243 --> 00:04:57.880 But this is where the beauty of skills come. 113 00:04:58.805 --> 00:04:59.900 I'll show my screen here. 114 00:05:00.963 --> 00:05:06.140 the your skill again this is not like word for word how it looks but a skill basically looks 115 00:05:06.180 --> 00:05:14.200 like this there is a name there is a description and then underneath is a bunch of information 116 00:05:14.401 --> 00:05:21.200 i'm going to put a bunch of info what when you create a skill.md file what gets added into the 117 00:05:21.280 --> 00:05:27.280 context is actually just the name and the description right the bunch of info doesn't 118 00:05:27.300 --> 00:05:27.762 get added. 119 00:05:28.223 --> 00:05:33.500 So imagine you have two sentences versus an agent.md that has like a thousand lines 120 00:05:34.022 --> 00:05:35.367 that get added into the context. 121 00:05:35.508 --> 00:05:38.620 We're talking thousands of tokens compared to a couple hundred. 122 00:05:39.381 --> 00:05:44.155 And the agent only gets the bunch of info when it realizes it needs this skill. 123 00:05:44.756 --> 00:05:46.020 So if I have, 124 00:05:46.140 --> 00:05:49.660 let's say, a certain way of generating a report, a certain way of structuring my code, 125 00:05:50.421 --> 00:05:56.200 why would I put that in the agent.md file when I can have the agent call on it progressively when 126 00:05:56.260 --> 00:05:57.504 it needs it, right? 127 00:05:57.564 --> 00:06:02.780 So this is why skills are honestly like I'm a shill, I'm a maxi, but people 128 00:06:02.900 --> 00:06:03.342 do it wrong. 129 00:06:03.643 --> 00:06:06.474 And I'm going to share the right way on how did we create skills. 130 00:06:07.137 --> 00:06:07.900 So, so far we have 131 00:06:07.920 --> 00:06:12.178 the system prompt, the agent.md, the skills, and then we have the tools, right? 132 00:06:12.198 --> 00:06:12.640 So if you're using 133 00:06:12.720 --> 00:06:17.700 cloud code, there's already built-in tools, a read tool, a write tool, like there's many tools that 134 00:06:17.720 --> 00:06:18.082 it uses. 135 00:06:18.122 --> 00:06:22.420 This has to be added into the context because the model, the model doesn't call the 136 00:06:22.480 --> 00:06:27.220 tools it like it's the agent harness around it that allows it to call the tools and then in this 137 00:06:27.320 --> 00:06:32.320 case we also have our code base right like whatever if we're building a web app a mobile app i know 138 00:06:32.400 --> 00:06:36.140 most people here won't care for the specific framework and honestly we're getting to a point 139 00:06:36.180 --> 00:06:41.300 if you're not technical you really shouldn't um and then we have the user conversation so this is 140 00:06:41.380 --> 00:06:47.720 what the complete context window is filled with right and this can total up to let's say like 141 00:06:48.222 --> 00:06:50.550 at the beginning, this could be like 20,000 tokens. 142 00:06:51.293 --> 00:06:53.340 And as the conversation continues to grow, 143 00:06:53.841 --> 00:06:57.171 you might reach your limit of 25, 250,000 tokens. 144 00:06:57.532 --> 00:07:00.160 And that's when you see both cloud code and open 145 00:07:00.240 --> 00:07:02.067 AI codex, they'll compact, right? 146 00:07:02.670 --> 00:07:03.955 So beautiful so far, right? 147 00:07:04.236 --> 00:07:05.360 This is how context works. 148 00:07:06.441 --> 00:07:09.120 Why skills are important and how you should generate skills. 149 00:07:10.722 --> 00:07:12.251 Let's say I have a specific workflow. 150 00:07:12.552 --> 00:07:13.880 For example, for my YouTube channel, 151 00:07:14.161 --> 00:07:19.480 you know we're at a point right now uh greg where we get sponsors now crazy when i first joined 152 00:07:19.821 --> 00:07:24.180 nuts when i first came to the pot not a thing we get sponsors just your mom sponsoring the channel 153 00:07:24.200 --> 00:07:29.300 yeah yeah it was just her showing love feeding me uh but now we get sponsors i get a lot of emails 154 00:07:29.802 --> 00:07:34.800 and some are good some are bad and it's a lot of time i'm sure you're aware to comb through and to 155 00:07:34.940 --> 00:07:41.140 check so i have an open claw agent that has its own email right i don't have it i don't haven't 156 00:07:41.180 --> 00:07:45.439 given it access to my email because there's like attack vectors and I've been hacked before. 157 00:07:45.459 --> 00:07:45.620 So I'm 158 00:07:45.660 --> 00:07:48.672 very careful with these things, but it has its own email. 159 00:07:49.114 --> 00:07:50.720 And every time I get an email from like a 160 00:07:50.780 --> 00:07:54.130 sponsor, I forward that email to the agent. 161 00:07:54.551 --> 00:07:57.440 Now, the first time I told my open cloud agent, I'm 162 00:07:57.460 --> 00:08:00.954 going to forward you emails, check every 15 minutes when you have an email. 163 00:08:01.637 --> 00:08:02.500 And when you check the 164 00:08:02.500 --> 00:08:04.951 email, do research on a sponsor and tell me if they're worth it. 165 00:08:05.855 --> 00:08:06.960 That's all I told the agent. 166 00:08:07.883 --> 00:08:13.960 every sponsor email i sent it it was like legit legit legit perfect perfect perfect there was no 167 00:08:14.401 --> 00:08:19.160 like there was no rejection there was no this is bad or these guys are a scam or this product's not 168 00:08:19.280 --> 00:08:25.120 good like there was no deep research being done by it so then i realized huh okay the model needs 169 00:08:25.401 --> 00:08:31.540 a step-by-step guide this is when i create a skill but here's the problem a lot of people 170 00:08:31.981 --> 00:08:38.819 will, I'll just write it down here, will identify, identify they have a workflow, right? 171 00:08:38.840 --> 00:08:39.020 You have 172 00:08:39.040 --> 00:08:42.130 some sort of workflow and then they'll jump to create the skill right away. 173 00:08:43.054 --> 00:08:44.940 This is the, let me 174 00:08:45.080 --> 00:08:45.723 click hide here. 175 00:08:46.144 --> 00:08:47.569 This is the worst thing you can do. 176 00:08:47.770 --> 00:08:50.480 I'm just going to draw arrows to signify that 177 00:08:50.580 --> 00:08:51.121 this is bad. 178 00:08:51.743 --> 00:08:52.385 You don't do these. 179 00:08:52.806 --> 00:08:58.200 And the reason why you don't do this is imagine you hire an 180 00:08:58.240 --> 00:09:04.500 employee or you're mentoring somebody, correct me if I'm wrong, you're probably going to tell them 181 00:09:05.102 --> 00:09:05.643 what to do. 182 00:09:06.647 --> 00:09:09.456 And if they ask you questions on how to do it, you'll help them. 183 00:09:09.496 --> 00:09:10.740 You would ideally like 184 00:09:10.780 --> 00:09:11.463 them to fail. 185 00:09:11.945 --> 00:09:14.214 And then you want to then tell them, no, this is how you do it. 186 00:09:14.234 --> 00:09:15.580 Like there needs to be 187 00:09:15.620 --> 00:09:17.026 some sort of experiential learning. 188 00:09:17.588 --> 00:09:20.258 The way I've been creating skills, Greg, and I have like 100% 189 00:09:20.840 --> 00:09:26.440 hit right now when I tell my agent to do something specific, is I actually walk with it step by step 190 00:09:27.082 --> 00:09:28.025 on doing the workflow. 191 00:09:28.466 --> 00:09:33.120 So in the case of my YouTube analysis, I told the agent, okay, 192 00:09:33.240 --> 00:09:34.125 I just sent you an email. 193 00:09:34.286 --> 00:09:35.111 Tell me about the company. 194 00:09:35.513 --> 00:09:36.740 Company's this, this, that, and that. 195 00:09:37.120 --> 00:09:37.381 Okay. 196 00:09:37.422 --> 00:09:41.340 Check their Twitter, check their YouTube, check their trust pilot, check if they've raised 197 00:09:41.440 --> 00:09:41.822 any money. 198 00:09:42.344 --> 00:09:46.920 If two of these are, have not, if two of these don't exist or not in good standing, 199 00:09:47.461 --> 00:09:48.406 automatic rejection. 200 00:09:48.808 --> 00:09:50.416 It checked and it was like, you're absolutely right. 201 00:09:50.436 --> 00:09:51.120 I was using Opus. 202 00:09:52.283 --> 00:09:57.700 these this is not a good company and then it would just we would we have a spreadsheet in 203 00:09:57.860 --> 00:10:01.840 google sheets it'd be like no contact it's so frustrating too right because you're like 204 00:10:02.361 --> 00:10:08.740 you give it a task and it seems like so binary like right or wrong and then when you tell it 205 00:10:09.402 --> 00:10:13.360 hey like why didn't you look at the trust pilot why didn't you see if they've raised money 206 00:10:13.862 --> 00:10:18.920 you're absolutely right yeah absolutely it's like what and and the thing is the reason why this is 207 00:10:18.980 --> 00:10:22.811 the cases, the models actually don't think. 208 00:10:23.332 --> 00:10:24.957 They're predictors of tokens, right? 209 00:10:25.318 --> 00:10:26.060 So when you 210 00:10:26.140 --> 00:10:30.360 give it English, when I give it English, it maps it on this vector graph, and then it looks for the 211 00:10:30.460 --> 00:10:33.093 closest resemblance, and it says this is the response, right? 212 00:10:33.475 --> 00:10:34.400 So when you say, what is the 213 00:10:34.440 --> 00:10:35.304 capital of France? 214 00:10:35.746 --> 00:10:38.760 It maps it again on this graph, and it says, oh, Paris is pretty close, right? 215 00:10:38.800 --> 00:10:39.443 Then it gives you Paris. 216 00:10:39.764 --> 00:10:43.017 It has no, it doesn't think, it doesn't understand. 217 00:10:43.097 --> 00:10:43.700 It feels like it 218 00:10:43.800 --> 00:10:44.262 understands. 219 00:10:44.303 --> 00:10:45.287 It feels like it thinks. 220 00:10:45.709 --> 00:10:47.317 Heck, it even feels like it has emotion. 221 00:10:47.638 --> 00:10:48.000 That's because 222 00:10:48.000 --> 00:10:51.817 it's been trained on so much data, but it actually does not know how to think. 223 00:10:51.917 --> 00:10:52.560 And this is where 224 00:10:52.981 --> 00:10:56.995 a lot of people will be frustrated with like, why is it not understanding me? 225 00:10:57.376 --> 00:10:58.380 You have to walk with 226 00:10:58.440 --> 00:10:58.541 it. 227 00:10:58.922 --> 00:11:00.851 So I told it, okay, this is how you research. 228 00:11:00.871 --> 00:11:02.459 And it's like, okay, it researches. 229 00:11:02.499 --> 00:11:02.740 And guess 230 00:11:02.780 --> 00:11:02.901 what? 231 00:11:02.921 --> 00:11:03.927 This is part of the context. 232 00:11:04.390 --> 00:11:06.020 And we're like, okay, now that you're done researching, 233 00:11:06.361 --> 00:11:08.672 when it's a good company, these are the qualities you look for. 234 00:11:08.953 --> 00:11:10.300 And then when it's really good, 235 00:11:10.700 --> 00:11:11.443 send me an email. 236 00:11:12.266 --> 00:11:15.720 And then once we had a successful run and we did it again and again, 237 00:11:16.262 --> 00:11:18.088 then I converted it to a skill. 238 00:11:18.911 --> 00:11:21.460 The reason being is a lot of people create the skills themselves, 239 00:11:21.660 --> 00:11:25.560 or I mean, they'll use the AI to create the skill, but it doesn't have the context on what 240 00:11:25.560 --> 00:11:27.590 a successful run looks like, right? 241 00:11:27.610 --> 00:11:29.520 Because most of the times, especially if you're using OpenClaw, 242 00:11:29.941 --> 00:11:31.951 it's probably going to fail at the API call. 243 00:11:31.991 --> 00:11:33.879 It's probably going to call the data wrong. 244 00:11:33.900 --> 00:11:34.000 Like 245 00:11:34.000 --> 00:11:35.813 there's so many places it's going to get wrong. 246 00:11:35.854 --> 00:11:36.740 And I see a lot of people saying, 247 00:11:37.122 --> 00:11:38.026 it's just so frustrating. 248 00:11:38.047 --> 00:11:39.112 This is terrible technology. 249 00:11:39.133 --> 00:11:39.917 Why doesn't it work? 250 00:11:39.937 --> 00:11:40.500 It's because 251 00:11:40.500 --> 00:11:42.127 you don't understand how an agent works, right? 252 00:11:42.508 --> 00:11:45.360 It will mimic you perfectly, but you've given it 253 00:11:45.400 --> 00:11:46.664 nothing to mimic, right? 254 00:11:47.004 --> 00:11:50.474 So I will do the workflow myself. 255 00:11:50.494 --> 00:11:52.740 So the updated version is identify the 256 00:11:52.760 --> 00:11:58.234 workflow, go back and forth and teach it. 257 00:11:58.454 --> 00:12:00.660 So like I'm doing it, like I'll be like, okay, first do 258 00:12:00.720 --> 00:12:01.202 the research. 259 00:12:01.765 --> 00:12:02.468 Here's the result. 260 00:12:03.110 --> 00:12:04.496 And I'm like, what do you think about this? 261 00:12:04.898 --> 00:12:05.380 Oh, these guys are 262 00:12:05.420 --> 00:12:05.681 terrible. 263 00:12:05.701 --> 00:12:06.505 You're absolutely right. 264 00:12:06.666 --> 00:12:06.907 Okay. 265 00:12:07.329 --> 00:12:09.840 What do you, you should go to the Google sheet and mark 266 00:12:09.900 --> 00:12:10.984 this as bad company. 267 00:12:11.586 --> 00:12:12.108 I've done that. 268 00:12:12.529 --> 00:12:15.419 Once I've had that back and forth, then I tell the AI, 269 00:12:17.562 --> 00:12:22.012 review what you did and then create the skill. 270 00:12:22.493 --> 00:12:25.500 So now it has actual context with how it worked 271 00:12:26.183 --> 00:12:27.731 and it's going to create the skill beautifully. 272 00:12:27.751 --> 00:12:28.656 I don't handwrite skills. 273 00:12:28.676 --> 00:12:29.440 I don't think you need to. 274 00:12:29.900 --> 00:12:30.924 You can use AI to do it. 275 00:12:30.944 --> 00:12:36.100 They even have a skill to create skills, skill exception, but you should 276 00:12:36.140 --> 00:12:38.451 have the context of what a successful run looks like. 277 00:12:38.993 --> 00:12:40.560 And this is why, by the way, Greg, I don't 278 00:12:40.600 --> 00:12:41.303 install skills. 279 00:12:42.428 --> 00:12:45.400 Like I've seen people like, oh, this notion skill, this social media skill, 280 00:12:45.460 --> 00:12:45.681 whatever. 281 00:12:46.144 --> 00:12:46.707 I'll review it. 282 00:12:46.828 --> 00:12:47.451 I'll check it out. 283 00:12:47.713 --> 00:12:49.040 I'll even give it to my AI and be like, oh, 284 00:12:49.040 --> 00:12:50.265 what are some things we can learn from this? 285 00:12:50.747 --> 00:12:53.880 But I don't download skills because your agent needs 286 00:12:53.940 --> 00:12:57.874 the context of a successful run, which you then turn to skills, right? 287 00:12:58.396 --> 00:12:59.500 And this is the big thing 288 00:12:59.520 --> 00:12:59.741 I see. 289 00:12:59.761 --> 00:13:03.055 You see skills marketplaces, you see download this and that. 290 00:13:03.336 --> 00:13:04.340 First of all, it's an 291 00:13:04.781 --> 00:13:06.005 easy way to attack somebody. 292 00:13:06.346 --> 00:13:10.720 So I would be very, very careful with downloading some random person 293 00:13:10.860 --> 00:13:11.141 skills. 294 00:13:11.784 --> 00:13:14.594 But second of all, again, it's all about context, right? 295 00:13:14.935 --> 00:13:16.400 It's all about, and you know, 296 00:13:16.620 --> 00:13:18.728 OpenClaw has a memory layer and all these types of things. 297 00:13:19.089 --> 00:13:21.860 You want it to do the right thing. 298 00:13:21.880 --> 00:13:21.940 And 299 00:13:21.960 --> 00:13:25.234 the only way it can do the right thing is if you give it the proper context. 300 00:13:25.375 --> 00:13:26.560 And to me, the best 301 00:13:26.580 --> 00:13:31.115 way to create a skill is to work with it in your specific workflow. 302 00:13:31.596 --> 00:13:32.700 Once you have a successful run, 303 00:13:33.001 --> 00:13:37.700 tell it okay review what you just did this is the skill you need to create i'll pause here i mean it 304 00:13:37.760 --> 00:13:44.920 makes sense right because if you hired an employee you would do the same thing yeah you wouldn't you 305 00:13:44.920 --> 00:13:49.640 wouldn't just be like okay go do this thing good luck yeah uh and by the way this is how you're 306 00:13:49.640 --> 00:13:56.240 going to go do things forever you would map out a workflow you would identify what right and wrong 307 00:13:56.562 --> 00:13:59.440 is you would do it iteratively. 308 00:14:00.403 --> 00:14:02.600 And then once you've gotten to that point, 309 00:14:02.680 --> 00:14:03.418 you would codify it. 310 00:14:03.801 --> 00:14:03.946 100%. 311 00:14:04.664 --> 00:14:05.960 And I think like, that's the thing. 312 00:14:06.040 --> 00:14:08.820 Like we should treat models and these agents 313 00:14:09.485 --> 00:14:11.220 like very new employees 314 00:14:12.044 --> 00:14:14.720 versus like these black magic boxes 315 00:14:14.942 --> 00:14:16.660 that like know everything, right? 316 00:14:17.068 --> 00:14:17.640 They know everything 317 00:14:17.741 --> 00:14:19.280 because they've been trained on a lot of data, 318 00:14:19.763 --> 00:14:22.280 but they don't know your workflow, your steps, right? 319 00:14:22.401 --> 00:14:25.420 So I see a lot of people who have, you know, 320 00:14:25.941 --> 00:14:34.060 15 like right off the rip they'll set up open claw and um 15 sub-agents 30 skills yet you haven't 321 00:14:34.100 --> 00:14:37.920 even set up your own workflows right and these things are cool right off the bat and and there's 322 00:14:37.960 --> 00:14:44.560 a perfect time to use sub-agents i use sub-agents a lot but the way you build like i call it scaling 323 00:14:44.680 --> 00:14:49.820 for productivity not scaling for what looks cool right like i've seen like for example paper clip 324 00:14:49.880 --> 00:14:55.260 paper clip looks awesome cool i used it i loved it right but i think people would be more 325 00:14:55.340 --> 00:15:01.860 productive if they built up from scratch their own version meaning like okay you have your own like 326 00:15:02.703 --> 00:15:07.560 um you know like editor right content creator so you're you're asking people to do the work 327 00:15:08.161 --> 00:15:15.040 100 100 because the thing is it's like look i'm in the position where like people using like these 328 00:15:15.140 --> 00:15:19.580 beefed up things make a lot more sense for me and the reason being is like i can build a product 329 00:15:19.680 --> 00:15:23.940 like that like i know what your audience wants i know what my audience wants like you know heck 330 00:15:23.960 --> 00:15:25.920 Like I can spin up agents and build this thing, right? 331 00:15:26.201 --> 00:15:32.260 But if I'm going to be completely honest, if you want to scale for productivity, it starts with one agent and you building up the skills. 332 00:15:32.780 --> 00:15:38.460 And then, okay, now you've built up some skills and now you add a sub agent and your one agent manages multiple agents, right? 333 00:15:38.500 --> 00:15:43.980 Like imagine this, like imagine I start a company and off rip, I have 10 employees. 334 00:15:44.882 --> 00:15:46.860 Never managed a team in my life. 335 00:15:47.020 --> 00:15:48.980 Heck, I don't even have a really big family. 336 00:15:49.143 --> 00:15:50.120 So like I'm alone. 337 00:15:50.322 --> 00:15:50.760 You know what I mean? 338 00:15:50.840 --> 00:15:57.280 so it's like you have to sort of yeah it's not sexy um and i apologize if this is not the cool 339 00:15:57.340 --> 00:16:01.040 thing people wanted to hear but you sort of have to put in the work and build it up and i and i 340 00:16:01.261 --> 00:16:07.380 personally believe you're building skills like your personal human skills not skill.md files 341 00:16:07.882 --> 00:16:12.240 that when the models get better with the agents get better you will be more valuable because at 342 00:16:12.260 --> 00:16:17.840 the end of the day as long as there's no new paradigm for models lms just predict tokens 343 00:16:18.683 --> 00:16:23.920 they don't understand or know the way you and i do right and this is why although like yeah the 344 00:16:24.020 --> 00:16:28.980 job scene and all this stuff is scary i genuinely believe anyone who knows how these tools work and 345 00:16:29.020 --> 00:16:34.020 like knows how to build agents and like craft skills and like knows how to make them productive 346 00:16:34.682 --> 00:16:39.800 we're in for a good run so you're saying that if you know how to do this you won't join the 347 00:16:39.840 --> 00:16:46.640 permanent underclass the permanent underclass so is the permanent underclass basically like i've 348 00:16:46.680 --> 00:16:53.080 seen this these this this on twitter a lot is that basically ai has replaced you so now you're just 349 00:16:53.801 --> 00:17:02.280 from what i understand it's once agi comes all these white collar workers are going to lose 350 00:17:02.300 --> 00:17:09.760 their jobs and if you don't know how to build skills use ai people say you're joining the 351 00:17:09.820 --> 00:17:16.160 permanent underclass that's that's the term it's permanent too that's scary so i just have a little 352 00:17:16.240 --> 00:17:22.100 bit of time left yeah by the way like it's ridiculous to call it a permanent underclass 353 00:17:22.381 --> 00:17:27.780 yeah because that's terrifying i can understand underclass but permanent it's like like you say 354 00:17:27.800 --> 00:17:35.258 there's no hope like no yeah um i mean we are in like knowledge that took 20 people 20 years to 355 00:17:36.060 --> 00:17:40.660 acquire is not like 20 bucks a month right so there is like a huge shift right people who are 356 00:17:40.700 --> 00:17:46.980 non-technical or i think i saw yesterday like some guy had like 100 million dollars um and 357 00:17:47.341 --> 00:17:54.780 he vibe coded the whole app i think it was him 1.8 billion billion yeah so you know what i mean 358 00:17:54.820 --> 00:18:00.900 like it is the there is a shift right and i think this idea of like i love how you were like billion 359 00:18:01.041 --> 00:18:05.580 you were about to just leave this podcast and just be like no you know what it is i just realized man 360 00:18:05.660 --> 00:18:09.760 i overthink things like i just need to drop the thing release the thing and there's like wisdom 361 00:18:09.900 --> 00:18:14.800 in that like there needs to be this level of delusion which i don't have like i'm trying to 362 00:18:14.860 --> 00:18:18.500 work on where you're like this is just gonna work out we're just gonna launch the product 363 00:18:18.881 --> 00:18:24.939 it's gonna succeed and if it doesn't on to the next one because 1.8 billion yeah dude like b 364 00:18:25.601 --> 00:18:35.520 b usd yeah we're not talking monopoly because it was canadian uh it's it's um we're not talking 365 00:18:35.640 --> 00:18:44.600 carny coins we're talking we're talking real benjamin yeah yeah that makes sense that makes 366 00:18:44.640 --> 00:18:49.940 sense but yeah like i i hope this like understanding of like again i personally don't think you don't 367 00:18:49.960 --> 00:18:56.760 need an agent.md file unless you have something proprietary um skills are valuable build your own 368 00:18:56.880 --> 00:19:00.759 though build build your own like you know like when you asked your mom when you were a kid oh 369 00:19:01.040 --> 00:19:05.080 can we have mcdonald's and she's like we have food at home we have food at home build your own 370 00:19:05.200 --> 00:19:05.561 skills. 371 00:19:06.183 --> 00:19:13.460 For coding perspective, from coding wise, a lot of the companies, model companies have 372 00:19:13.540 --> 00:19:18.718 realized that the agents are really good at writing code, particularly TypeScript. 373 00:19:19.199 --> 00:19:19.420 And this 374 00:19:19.500 --> 00:19:25.880 is why there's been like, you see this advancement with like CloudCore work and like even OpenClaw. 375 00:19:26.301 --> 00:19:28.597 Really what they're doing under the hood is they're writing code, right? 376 00:19:28.618 --> 00:19:28.960 They're writing 377 00:19:29.040 --> 00:19:36.180 code calling apis and all this stuff so when it comes to building a project um you actually don't 378 00:19:36.280 --> 00:19:41.840 need skills or like you don't need an agent md file specific to the text that you use like i 379 00:19:41.880 --> 00:19:47.820 remember we used to i'm using react and you know convicts or i'm using next.js and superbase i'm 380 00:19:47.840 --> 00:19:52.680 using this and i'm using that and you put that in the agent md file you have like all these lines 381 00:19:53.322 --> 00:19:59.480 for the most part unless again you have a specific specific workflow unnecessary and the reason being 382 00:19:59.560 --> 00:20:05.940 is code itself has become context now so the more the more important thing is starting with a solid 383 00:20:06.040 --> 00:20:10.280 foundation templates used to be big back in the day people made lots of money with templates i 384 00:20:10.400 --> 00:20:15.940 believe templates are going to have a renaissance because if you have a solid like template right 385 00:20:16.020 --> 00:20:21.020 like whether it be like for a web app or mobile app because that becomes context for the agent 386 00:20:21.282 --> 00:20:23.800 It's going to build on top of that, right? 387 00:20:24.461 --> 00:20:27.560 And again, I didn't need some large agent.md file. 388 00:20:27.760 --> 00:20:29.760 I didn't need any large cloud MD file. 389 00:20:30.140 --> 00:20:35.020 What I needed was, again, minimal context usage and skills. 390 00:20:35.640 --> 00:20:39.000 So if there's anything anyone can learn from me 391 00:20:39.362 --> 00:20:41.920 is build your own skills, build your own skills. 392 00:20:42.302 --> 00:20:43.360 And there's this methodology. 393 00:20:43.863 --> 00:20:44.780 I don't know if I've shared this with you, 394 00:20:44.860 --> 00:20:46.140 recursively building skills. 395 00:20:46.540 --> 00:20:48.320 So let's say you've built your skill, right? 396 00:20:48.441 --> 00:20:51.360 I have, I'll draw a diagram because why not? 397 00:20:52.402 --> 00:20:54.260 Let's say I have a workflow. 398 00:20:55.220 --> 00:20:59.900 And after you like setting up my workflow with an agent, I've decided, you know what? 399 00:21:00.262 --> 00:21:03.460 I'm going to turn this into a skill, right? 400 00:21:03.560 --> 00:21:06.698 So this is my skill.md. 401 00:21:07.623 --> 00:21:08.820 Now, here's the thing. 402 00:21:09.301 --> 00:21:14.480 Even though you have the skill.md, the agent at some point is still going to mess up because 403 00:21:14.581 --> 00:21:16.900 there's probably gaps in the information it has in the skill. 404 00:21:17.621 --> 00:21:21.103 So when it messes up, I'm going to work with it again. 405 00:21:21.143 --> 00:21:21.964 How do I work with it? 406 00:21:22.424 --> 00:21:22.965 You messed up. 407 00:21:23.825 --> 00:21:25.026 Try calling the API again. 408 00:21:25.347 --> 00:21:26.287 Try doing this again. 409 00:21:26.447 --> 00:21:28.609 Or even ask it when it tells you, oh, I failed. 410 00:21:28.629 --> 00:21:29.550 I couldn't do this task. 411 00:21:30.090 --> 00:21:32.932 Believe it or not, when you tell the agent, why did you fail? 412 00:21:32.952 --> 00:21:35.394 When you ask it, like, what's the error that you got? 413 00:21:35.755 --> 00:21:37.076 It will tell you descriptively. 414 00:21:37.136 --> 00:21:38.277 Oh, I got a 505 error. 415 00:21:39.237 --> 00:21:40.418 You have insufficient credits. 416 00:21:40.718 --> 00:21:42.200 Like, oh, okay, so it's a credit issue. 417 00:21:42.580 --> 00:21:42.740 Fine. 418 00:21:43.381 --> 00:21:52.200 so I would tell it that and then I would pass that failure back to the agent so let's say it did 419 00:21:52.280 --> 00:21:57.340 something wrong we identified the failure all I did was asking it I will give that failure back 420 00:21:57.400 --> 00:22:03.440 to the agent I'll be like you failed here this didn't work fix this it's going to fix it's going 421 00:22:03.460 --> 00:22:08.220 to write code it's going to do whatever it does once it fixes it and it's done it right now you 422 00:22:08.300 --> 00:22:12.432 tell it with the new fix, update the skill so this doesn't happen again. 423 00:22:13.395 --> 00:22:14.940 I have like for my YouTube 424 00:22:15.040 --> 00:22:16.786 channel, I have like a report generator. 425 00:22:17.268 --> 00:22:20.740 It calls Notion, Dubb Analytics, YouTube Analytics, Twitter 426 00:22:20.860 --> 00:22:24.070 Analytics, posts from like eight data sources. 427 00:22:24.572 --> 00:22:27.220 There's no way you're going to one prompt and 428 00:22:27.260 --> 00:22:28.204 the agent's going to do it. 429 00:22:28.727 --> 00:22:31.600 But every time I tell it to do that work, it takes like 10 minutes. 430 00:22:32.040 --> 00:22:33.445 It executes it flawlessly. 431 00:22:33.485 --> 00:22:33.726 Why? 432 00:22:34.087 --> 00:22:35.451 I went through five loops of this. 433 00:22:36.334 --> 00:22:38.220 Five iterations of recursively 434 00:22:38.641 --> 00:22:39.544 building this skill. 435 00:22:39.865 --> 00:22:41.129 And that skill is so good. 436 00:22:41.531 --> 00:22:44.440 I genuinely think if anyone's going to, if like 437 00:22:44.560 --> 00:22:48.900 skills marketplace is going to be a thing, there's going to be people who sell skills, like really 438 00:22:49.040 --> 00:22:55.740 well-defined, like step-by-step skills, because people are just creating them without having built 439 00:22:55.860 --> 00:22:57.927 out the workflow with the agent, right? 440 00:22:57.987 --> 00:23:01.820 So use the workflow by hand, like telling it each step. 441 00:23:02.520 --> 00:23:06.476 Once it's done it completely, create the skill.md file, continue to use it. 442 00:23:06.737 --> 00:23:07.520 It's going to mess up. 443 00:23:07.701 --> 00:23:12.100 When he messes up, you thank God you don't complain because a lot of people are like, oh, I messed up. 444 00:23:12.160 --> 00:23:12.640 I'm angry. 445 00:23:12.800 --> 00:23:15.360 No, this is a moment where you identify the error. 446 00:23:15.740 --> 00:23:16.820 Tell it this is the error. 447 00:23:16.960 --> 00:23:17.338 Fix it. 448 00:23:17.621 --> 00:23:18.520 It'll fix it itself. 449 00:23:19.001 --> 00:23:23.100 And then you tell it to update the skill file so that this doesn't happen again. 450 00:23:23.281 --> 00:23:26.740 So that's a little bit about shifting your expectation, right? 451 00:23:26.800 --> 00:23:30.880 Because people just assume it's going to work in the beginning. 452 00:23:31.342 --> 00:23:34.360 You're saying basically it's not going to work initially. 453 00:23:34.660 --> 00:23:36.839 There's going to be two, three, five, six hiccups. 454 00:23:38.223 --> 00:23:47.100 and over time it should be good so this is most people's expectations right yeah and the way 455 00:23:47.822 --> 00:23:54.180 i've personally experienced is it's like this so there's like this early area of investment that 456 00:23:54.220 --> 00:24:00.540 you have to make that sucks that nobody will tell you especially agent harness's company because 457 00:24:00.580 --> 00:24:04.880 they wouldn't raise as much money if they did but like this maybe i would give it two weeks 458 00:24:05.426 --> 00:24:06.480 because it took me two weeks. 459 00:24:06.720 --> 00:24:08.400 Like OpenClaw, when I first set up OpenClaw, 460 00:24:09.002 --> 00:24:11.480 I thought at the same time, I'm like, what is this garbage? 461 00:24:12.160 --> 00:24:13.740 Or like, it doesn't understand anything, it's confused. 462 00:24:13.860 --> 00:24:17.060 And then I realized like, oh, like, let me go lower level. 463 00:24:17.702 --> 00:24:18.800 The models and the agents, 464 00:24:19.160 --> 00:24:21.420 like they don't think like you and me, right? 465 00:24:21.863 --> 00:24:24.280 I could tell you, hey, Greg, 466 00:24:24.603 --> 00:24:26.820 we need a report on like, you know, 467 00:24:27.002 --> 00:24:28.120 the financials and notion. 468 00:24:29.282 --> 00:24:31.120 Because you probably were in the same business, 469 00:24:31.200 --> 00:24:32.700 we worked together, you would understand 470 00:24:32.801 --> 00:24:34.280 based on the context you have of the business, 471 00:24:34.662 --> 00:24:38.240 what that means but imagine a new guy joins like yeah i need to report on the financials 472 00:24:39.100 --> 00:24:42.139 so where do i even start it reminds me i wonder if we can put this clip in 473 00:24:42.882 --> 00:24:48.160 but in the office you watch the office i am not an office watcher unfortunately there's a clip 474 00:24:48.481 --> 00:24:57.100 that uh there's a new boss and the new boss goes to jim one of the main characters yeah 475 00:24:57.686 --> 00:24:59.000 And he asked for a rundown. 476 00:25:00.000 --> 00:25:03.960 So go the office, the office, the rundown. 477 00:25:05.945 --> 00:25:06.298 Oh, no. 478 00:25:07.580 --> 00:25:17.220 Basically, the whole episode is about Jim trying to ask around and be like, what is a rundown? 479 00:25:17.843 --> 00:25:18.580 Like, what is a rundown? 480 00:25:18.720 --> 00:25:20.840 He's like calling his dad, like, what is a rundown? 481 00:25:21.423 --> 00:25:21.940 You know what I mean? 482 00:25:21.980 --> 00:25:24.780 He's just, he didn't have the context. 483 00:25:24.887 --> 00:25:25.057 Yeah. 484 00:25:25.240 --> 00:25:26.079 He didn't have the context. 485 00:25:26.460 --> 00:25:26.560 Yeah. 486 00:25:27.261 --> 00:25:28.641 And it goes back to my initial point. 487 00:25:29.162 --> 00:25:33.925 The models are really, really good now, but the context matters more than anything. 488 00:25:33.965 --> 00:25:34.145 Right. 489 00:25:34.205 --> 00:25:38.969 So when you see like these large agent like companies and sub agents. 490 00:25:39.009 --> 00:25:46.935 And again, I'm not saying those don't work, but I'm saying probably won't work for you off rip because you haven't built it up to get to that point. 491 00:25:47.335 --> 00:25:47.575 Right. 492 00:25:47.635 --> 00:25:52.138 So let's say like for me, for example, I started with one agent. 493 00:25:52.759 --> 00:25:53.479 Let me draw this. 494 00:25:53.539 --> 00:25:54.420 I started with one agent. 495 00:25:54.906 --> 00:25:56.120 and this was like my main agent. 496 00:25:56.240 --> 00:25:57.120 This did everything, right? 497 00:25:57.180 --> 00:25:59.180 This checked my spreadsheet, 498 00:25:59.440 --> 00:26:00.900 this checked my sponsor's email 499 00:26:01.508 --> 00:26:02.480 and all these type of things. 500 00:26:02.901 --> 00:26:04.780 And once I had like predefined workflows, 501 00:26:05.460 --> 00:26:08.140 let's say for like working with sponsors, 502 00:26:08.703 --> 00:26:10.620 then I can actually have a sub-agent. 503 00:26:10.801 --> 00:26:12.100 What's the purpose of the sub-agent? 504 00:26:12.160 --> 00:26:13.980 The sub-agent does all the marketing stuff, right? 505 00:26:14.721 --> 00:26:16.700 But I'm not creating the sub-agent 506 00:26:16.781 --> 00:26:17.840 for the sake of creating it. 507 00:26:18.122 --> 00:26:19.200 It's going to have skills, 508 00:26:19.320 --> 00:26:20.360 it's going to have context 509 00:26:20.823 --> 00:26:23.120 and it actually makes sense for me to have sub-agents, right? 510 00:26:23.422 --> 00:26:26.780 So I've built out my thing to like, now I have five sub-agents. 511 00:26:26.860 --> 00:26:32.740 I have one for marketing, one for business, one for personal, and that's it. 512 00:26:33.260 --> 00:26:40.840 And I'm willing to bet if I want open cloud or open cloud with anyone, my system is more productive because I didn't scale for what looks cool. 513 00:26:41.243 --> 00:26:42.280 I scale for productivity. 514 00:26:44.289 --> 00:26:44.720 That was a bar. 515 00:26:45.245 --> 00:26:46.100 That was a huge bar. 516 00:26:46.140 --> 00:26:46.840 We got to clip that. 517 00:26:47.181 --> 00:26:49.100 I was just thinking that clip, that's going to rip. 518 00:26:50.323 --> 00:26:51.360 Yeah, that was a bar. 519 00:26:51.864 --> 00:26:53.700 But what else do you want to leave people with? 520 00:26:53.840 --> 00:26:55.200 Or is this, this is the main point. 521 00:26:55.360 --> 00:26:55.480 Yeah. 522 00:26:55.540 --> 00:26:59.180 Like here's like the, we've got to put where the models are good. 523 00:26:59.522 --> 00:27:00.500 The models are really good. 524 00:27:01.001 --> 00:27:03.420 The context matters plus the harness, right? 525 00:27:03.560 --> 00:27:08.920 So for example, there was this benchmark, although I'm not a hundred percent supporting 526 00:27:09.040 --> 00:27:14.920 it, that there was a difference between the quality of output that cursor generated versus 527 00:27:15.263 --> 00:27:17.580 cloud code versus codex, right? 528 00:27:18.362 --> 00:27:23.840 Um, so what that tells me is that we've reached a point where the models are really, really good. 529 00:27:24.100 --> 00:27:25.349 They're probably going to get better. 530 00:27:25.389 --> 00:27:26.960 The next iteration is probably going to get better, 531 00:27:27.441 --> 00:27:33.700 but the harness and the tools that you surrounded, the context that you give it is going to matter 532 00:27:33.860 --> 00:27:34.242 even more. 533 00:27:34.923 --> 00:27:38.494 And just like in everything in life, less is more, right? 534 00:27:38.815 --> 00:27:40.600 Like building up step-by-step, 535 00:27:40.860 --> 00:27:45.176 making it productive for you first, before you add the shiny new thing. 536 00:27:45.317 --> 00:27:46.180 Like, cause I tried all 537 00:27:46.240 --> 00:27:50.120 these tools all the time like especially paperclip paperclip blew up and a lot of people have been 538 00:27:50.140 --> 00:27:55.140 talking about and it's fantastic but i'm willing to bet if people took two weeks to build up to 539 00:27:55.180 --> 00:27:59.820 the version because you can prompt open cloud to do all that stuff if they built up their own version 540 00:27:59.900 --> 00:28:04.640 of paperclip in two three weeks where like they're building things that they actually need their 541 00:28:04.700 --> 00:28:09.520 productivity level will skyrocket through the roof it's a hot take it's a hot take might get me in 542 00:28:09.580 --> 00:28:13.780 trouble no it won't get who's it gonna get you in trouble with maybe paperclip uses a billion 543 00:28:13.900 --> 00:28:21.600 dollars and they don't acquire my podcast i think uh listen you're you're out there you're trying 544 00:28:21.680 --> 00:28:26.280 things and you're just sharing what you're learning in real time so if you're just you're 545 00:28:26.280 --> 00:28:30.340 not things can change by the way yeah like two weeks from now it could be like no give the agent 546 00:28:30.541 --> 00:28:35.580 everything there's this new memory paper that google released and like now like it has the 547 00:28:35.600 --> 00:28:42.180 ability to index information and stuff but as it as it pertains to real life less is more simple 548 00:28:42.321 --> 00:28:47.380 is better right if you can't explain it in in a few sentences you probably don't really understand 549 00:28:47.480 --> 00:28:52.900 it right and i find that the models are trained on so much information especially when it comes 550 00:28:52.920 --> 00:28:58.560 to programming building and like and um what do you call like day-to-day work like financial work 551 00:28:58.761 --> 00:29:03.400 or like any sort of like you know checking contracts and stuff like the the model companies 552 00:29:03.480 --> 00:29:08.740 are focusing on that like on white collar work the models are really really good what matters 553 00:29:08.861 --> 00:29:11.050 more is the harness and the tools you provided. 554 00:29:11.392 --> 00:29:13.100 And the one thing that you and I have that the 555 00:29:13.180 --> 00:29:19.280 models don't have is my specific workflow, my specific taste, my specific strategy of doing 556 00:29:19.380 --> 00:29:19.621 things. 557 00:29:20.043 --> 00:29:22.491 And those can be codified in skills, right? 558 00:29:22.571 --> 00:29:25.040 This is why like skills make sense when 559 00:29:25.180 --> 00:29:25.863 you build them. 560 00:29:26.505 --> 00:29:30.900 Not if you download my skill, like I have this one skill, like again, don't 561 00:29:30.920 --> 00:29:31.703 download it. 562 00:29:31.884 --> 00:29:33.630 I'm telling you now, do not download it. 563 00:29:33.670 --> 00:29:34.252 Don't use it. 564 00:29:34.754 --> 00:29:36.260 I just put it so I can get some 565 00:29:36.541 --> 00:29:37.243 GitHub stars. 566 00:29:39.531 --> 00:29:42.300 I have this one skill and it's literally a code structure skill. 567 00:29:43.201 --> 00:29:45.387 And I'll put the mark down so people can see it. 568 00:29:46.590 --> 00:29:48.395 It's 116 lines. 569 00:29:48.576 --> 00:29:50.040 It's basically after AI has 570 00:29:50.100 --> 00:29:51.165 generated a bunch of code. 571 00:29:51.707 --> 00:29:53.294 I like it structured in a certain way. 572 00:29:53.334 --> 00:29:54.740 So it's easy for me to review it. 573 00:29:55.322 --> 00:29:59.240 And like I mentioned earlier with skills, the only thing that gets added into context is the 574 00:29:59.260 --> 00:30:00.044 name and description. 575 00:30:00.526 --> 00:30:02.515 So when I look at the name, it's code structure. 576 00:30:02.556 --> 00:30:03.480 When I look at the description 577 00:30:04.002 --> 00:30:08.500 use when multiple workflows duplicate the same operational logic when deciding that blah blah 578 00:30:08.500 --> 00:30:13.560 blah blah some nerd stuff so when i tell the agent i want to clean up the code structure 579 00:30:14.082 --> 00:30:19.380 it checks the skills it has it sees the name it reads the description it's like oh this makes 580 00:30:19.420 --> 00:30:24.680 sense then it progressively discloses meaning once it realizes it needs this skill then it adds the 581 00:30:24.760 --> 00:30:30.840 rest of this right versus if this was my agent.md file imagine every single time and we can actually 582 00:30:31.461 --> 00:30:41.580 check how many tokens this is let me check um what was it open ai token tokenizer if i go to this 583 00:30:42.963 --> 00:30:51.120 so this is 944 tokens so if this was an agent.md file every single time i have a chat i'm adding 584 00:30:51.321 --> 00:30:58.059 944 tokens tokens ain't cheap now no but if i just have the name and the description is just 53 585 00:30:58.521 --> 00:30:59.564 And it's not even cheap. 586 00:30:59.604 --> 00:31:04.680 It's just like, you're not trying to hit the limit quicker than you need to hit the limit. 587 00:31:04.861 --> 00:31:07.559 Because the model will get dumb as the context window closes, right? 588 00:31:07.820 --> 00:31:24.560 So if you have like a context window and I can draw this out, if this is your context window and like the optimal is you're between like there's always like maybe like 10% already filled with all the system prompt and all that stuff. 589 00:31:24.640 --> 00:31:29.053 you want to be between like, you know, fresh to like 70%. 590 00:31:29.094 --> 00:31:30.919 Because the closer you get to 99, 591 00:31:31.761 --> 00:31:34.974 100%, like 99, 90, 80%, it starts to get dumb, right? 592 00:31:35.315 --> 00:31:36.500 And you could think of this like a human, 593 00:31:36.600 --> 00:31:41.439 like imagine you throw a bunch of information again and again and again and again. 594 00:31:41.459 --> 00:31:41.640 And this 595 00:31:41.680 --> 00:31:46.799 is why like, when I like was in school, like last minute studying never worked for me. 596 00:31:46.839 --> 00:31:47.120 Because like, 597 00:31:47.140 --> 00:31:48.466 I didn't pay attention the entire year. 598 00:31:48.848 --> 00:31:50.475 Now I have to learn about polynomials. 599 00:31:50.897 --> 00:31:51.419 And I have to 600 00:31:52.102 --> 00:31:56.880 do these graphs and there's this weird notation, it's impossible for me to catch up, right? 601 00:31:57.341 --> 00:31:58.708 And it's the same way with the agents. 602 00:31:58.748 --> 00:32:00.577 You want to keep your context window. 603 00:32:00.597 --> 00:32:01.180 You want to save your 604 00:32:01.220 --> 00:32:02.685 context window because it saves you money. 605 00:32:03.006 --> 00:32:06.940 But not only that, it makes a more performant agent. 606 00:32:07.161 --> 00:32:08.705 So less is more. 607 00:32:09.026 --> 00:32:09.748 Less is more. 608 00:32:09.828 --> 00:32:11.433 Rely more on the model strengths. 609 00:32:11.854 --> 00:32:13.640 And what the model needs is what's 610 00:32:14.142 --> 00:32:18.018 unique and special about you, your workflow, your business, not general knowledge. 611 00:32:18.098 --> 00:32:18.580 Don't tell the 612 00:32:18.640 --> 00:32:26.240 model use react it knows to use react don't tell the model um you know things that like should 613 00:32:26.401 --> 00:32:31.920 already be known uh for the like you know tasks like for example like let's say i'm doing a 614 00:32:31.960 --> 00:32:38.020 financial report and in the agents.md file i say um to denote money use a dollar sign 615 00:32:39.785 --> 00:32:45.460 it's going to use a dollar sign right now if you have a specific currency then you like oh use this 616 00:32:45.600 --> 00:32:51.520 currency this is the you know like for something that the agent won't do manually like won't know 617 00:32:51.600 --> 00:32:57.160 manually that's when you have like your agent.mds claw.mds but honestly these are a farce you don't 618 00:32:57.200 --> 00:33:04.680 need them um skills skills skills skills skills is what it's at thanks for keeping it real i 619 00:33:04.720 --> 00:33:09.720 appreciate you man that's all i'm gonna do no i appreciate it uh like always i'll include links 620 00:33:09.901 --> 00:33:15.640 where you can follow Ross Mike on YouTube and X and other places in the show notes in the 621 00:33:15.680 --> 00:33:16.302 description. 622 00:33:16.322 --> 00:33:17.265 So go follow him there. 623 00:33:17.987 --> 00:33:19.572 Always clearly breaking down things. 624 00:33:19.773 --> 00:33:22.120 We, uh, I have to 625 00:33:22.140 --> 00:33:22.642 be real with you. 626 00:33:22.662 --> 00:33:24.129 You weren't going to come on the show today. 627 00:33:24.651 --> 00:33:25.113 I wasn't. 628 00:33:25.234 --> 00:33:26.138 And I'll be honest. 629 00:33:26.258 --> 00:33:26.620 I told 630 00:33:26.760 --> 00:33:28.045 Greg and I'm just going to be frank. 631 00:33:28.065 --> 00:33:32.580 I'm like, I don't have that banger, you know, something new 632 00:33:32.680 --> 00:33:33.062 drop in. 633 00:33:33.082 --> 00:33:33.725 Let's review it. 634 00:33:33.826 --> 00:33:36.960 Cause if we going to be honest, they're not that many tools dropping 635 00:33:37.060 --> 00:33:42.480 nowadays like unfortunately the big dogs are running the show yeah um the clods and the 636 00:33:43.021 --> 00:33:49.400 the anthropics and the open ai especially when it comes to general purpose and and coding they sort 637 00:33:49.440 --> 00:33:54.820 of run the game so they're releasing updates and like all the stuff has already been covered so i 638 00:33:54.880 --> 00:33:58.580 was like greg i don't know if i have anything valuable and what did i say you're like the 639 00:33:58.640 --> 00:34:02.320 people you know you got to think about impact you got to think about what you know this could 640 00:34:02.380 --> 00:34:07.260 apply to someone's and you showed me like a testimony right like i sent a text to you yeah 641 00:34:07.440 --> 00:34:15.720 i'm gonna pull it up uh i sent a text to you of someone who saw a video that we did together 642 00:34:16.943 --> 00:34:22.438 and that video got him into coding now he's running a cake business and he's making 150 643 00:34:23.200 --> 00:34:31.320 000 a year and growing and he said the greg and ross mike episode in november last year is what 644 00:34:31.380 --> 00:34:37.360 got me into coding i've recommended to everyone asking how to start out and i just sent you that 645 00:34:37.440 --> 00:34:42.039 text i said it's not about the numbers it's not about you know because you said in the text 646 00:34:43.323 --> 00:34:49.660 you don't see it sometimes right i need everything we do to get to 200k views yeah yeah yeah and i'm 647 00:34:49.740 --> 00:34:55.340 just like i hope this gets 200k views or more so like and comment to juice those algorithms but 648 00:34:55.681 --> 00:35:01.260 if it gets 2000 and two people end up taking this information and changes their business, 649 00:35:01.400 --> 00:35:03.490 their productivity, how they think about things. 650 00:35:03.591 --> 00:35:05.380 And, you know, I think that's why, 651 00:35:06.242 --> 00:35:12.000 I think that's why you and myself have been put on this planet earth is to inspire people to get 652 00:35:12.040 --> 00:35:13.224 their creative juices flowing. 653 00:35:13.826 --> 00:35:17.800 And so I thank you for, for coming on and, and taking time out of 654 00:35:17.800 --> 00:35:18.041 your day. 655 00:35:18.061 --> 00:35:21.980 And I appreciate the motivation and yeah, I hope this helps somebody and I can't 656 00:35:22.103 --> 00:35:22.840 to be back with more. 657 00:35:23.280 --> 00:35:23.515 Absolutely. 658 00:35:23.880 --> 00:35:24.120 All right. 659 00:35:24.160 --> 00:35:24.719 Catch you later, dude.