WEBVTT NOTE Sentence-level transcript of https://www.youtube.com/watch?v=0JZy5AuuQzc NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/0JZy5AuuQzc.html. A cue ends where the next begins, or 2 s after its last word. s1 00:00:01.880 --> 00:00:11.720 If I tell you the New York City tonight suddenly stops working, you tell me what could be possible reasons for no flights leaving JFK. s2 00:00:11.720 --> 00:00:16.320 The subway suddenly stops at the next station and stays there. s3 00:00:16.320 --> 00:00:19.720 None of your tap pay is working. s4 00:00:19.720 --> 00:00:25.200 The elevators you get in, they don't know which floor they are. s5 00:00:25.200 --> 00:00:27.560 The phones are not ringing. s6 00:00:27.560 --> 00:00:33.320 The Wall Street and the entire world comes to a standstill. s7 00:00:33.320 --> 00:00:35.920 What could be the reason? s8 00:00:37.960 --> 00:00:39.960 Software. s9 00:00:39.960 --> 00:00:45.680 Unless otherwise, there is a software issue that affects the entire globe. s10 00:00:45.680 --> 00:00:49.200 There is nothing that can stop this modern world. s11 00:00:49.200 --> 00:00:58.000 This is how much the city of New York and our entire globe is actually driven by unseen software. s12 00:00:58.000 --> 00:01:01.840 Not the AI kind of software that we mostly talk about. s13 00:01:01.840 --> 00:01:12.280 It's the traditional, deterministic, rule-based, reliable software that has brought the modern world to us. s14 00:01:12.280 --> 00:01:17.800 It is handcrafted for over 70 years by humans. s15 00:01:17.800 --> 00:01:19.360 I'm Hari. s16 00:01:19.360 --> 00:01:25.080 I'm the product person who has built software for almost like two decades now. s17 00:01:25.080 --> 00:01:28.280 And I have this one genuine theory. s18 00:01:29.200 --> 00:01:35.320 If at all there is a new industry in the human history, it must be software industry. s19 00:01:35.320 --> 00:01:45.680 Because you take navigation, agriculture, construction, and even astronomy, we had this in different forms and shapes earlier. s20 00:01:45.680 --> 00:01:50.240 Now that it may be more sophisticated, but they're nothing new to us. s21 00:01:50.240 --> 00:02:05.880 Today from getting to work to watching this video, from nuclear reactor to all the way to space station that drives the 125 trillion dollar economy the one common factor is software. s22 00:02:05.880 --> 00:02:12.679 So, it would be catastrophic if for some reason software stops, just like the Y2K. s23 00:02:12.959 --> 00:02:19.080 And our modern world was a lot to this kind of a software. s24 00:02:19.080 --> 00:02:26.080 The deterministic kind, where you trust it with your pacemaker and when you fly on your 747. s25 00:02:27.080 --> 00:02:35.360 So, if you look into the product here on the left side, you have the engineers. s26 00:02:36.840 --> 00:02:47.160 The engineers are hired and they are the source of judgment for every product out there with their unique taste. s27 00:02:47.160 --> 00:02:56.400 And I'll go to an extent and call their belief systems and value systems are reflective in the software that they shipped. s28 00:02:56.720 --> 00:03:02.519 The tool chain owned deterministic rules all the way. s29 00:03:02.600 --> 00:03:14.519 The software that runs the world and the tool chains next to it in a cross feedback loop where the assemblers built C, C built Unix, Unix built everything and then s30 00:03:14.519 --> 00:03:20.560 we built Kafka, Zookeeper, all the way up to Kubernetes that runs the world today. s31 00:03:20.920 --> 00:03:25.800 The only judgment in this picture is human judgment. s32 00:03:25.800 --> 00:03:29.040 And we all know how varied we are. s33 00:03:29.040 --> 00:03:35.720 And that variance is where every product edge originated. s34 00:03:36.600 --> 00:03:43.720 The block here is a high-level representation of the entire software industry and the world economy. s35 00:03:43.720 --> 00:03:44.680 Who has the IP? s36 00:03:44.680 --> 00:03:46.800 Where is the continuous loop? s37 00:03:46.800 --> 00:03:54.600 And how that loop where every brick on the tool chain helps us scale new heights. s38 00:03:54.600 --> 00:03:59.400 Whether the software is deterministic or probabilistic, that's a question. s39 00:03:59.400 --> 00:04:01.120 And hold this picture. s40 00:04:01.120 --> 00:04:04.040 Let's watch one card change. s41 00:04:04.040 --> 00:04:18.760 Which is, yes, in the age of AI, the engineers today are assisted or driven or sometimes they claim it's completely autonomous by models that I would call rented. s42 00:04:19.440 --> 00:04:28.840 The good thing is that it creates a level playing field with everyone having equal access to the bleeding edge models and harnesses. s43 00:04:28.840 --> 00:04:44.320 Teams who cannot afford earlier to the talent pools of Google and Meta who handcrafted this modern software can now afford intelligence through APIs where it claims to have solved s44 00:04:44.320 --> 00:04:45.640 the software problem. s45 00:04:45.640 --> 00:04:48.520 Let's see whether it really solved or what's happening. s46 00:04:48.520 --> 00:05:02.919 If you actually pay attention and look under the hood everything around the model, the hardness, the agents, the tools, the MCP are still the same old deterministic software. s47 00:05:02.919 --> 00:05:09.080 People who are building this would definitely know that the these are not AI products. s48 00:05:09.080 --> 00:05:15.840 These are the same old deterministic software built through maybe the models. s49 00:05:15.840 --> 00:05:22.960 And what did not change and stay put is what's the outcome? s50 00:05:22.960 --> 00:05:31.320 At the end of the world day, the world requires software that does the same thing reliably every time. s51 00:05:31.320 --> 00:05:41.400 So, the final artifact is a deterministic software that drives the world economy as we know it. s52 00:05:41.400 --> 00:05:45.040 If that is the case, then what is the role of AI here? s53 00:05:45.040 --> 00:05:56.400 Its role is to one to accelerate the development of deterministic software at the lowest cost possible. s54 00:05:57.000 --> 00:06:07.920 And write deterministic software that never existed before because of budget not an interesting problem, hard problems. s55 00:06:07.920 --> 00:06:13.480 So, those are the two use cases I see going to play out in the next 5 to 10 years. s56 00:06:13.480 --> 00:06:31.520 So, the bottom line is we will be using a probabilistic gear to crank out a reliable deterministic software by augmenting or somebody will go to an extent say replacing s57 00:06:31.520 --> 00:06:33.280 human judgment. s58 00:06:33.280 --> 00:06:34.720 Is it possible? s59 00:06:34.720 --> 00:06:36.040 Where are we today? s60 00:06:36.040 --> 00:06:39.480 We will see in this talk. s61 00:06:41.240 --> 00:06:43.560 Who knows this person? s62 00:06:43.560 --> 00:06:45.520 I'm a big fanboy of him. s63 00:06:45.520 --> 00:06:48.520 He's a superhuman and his friends. s64 00:06:51.040 --> 00:07:03.040 Okay, now I see you will connect dots because you know the couple of men's there and he's part of the elite sub-10 club. s65 00:07:03.040 --> 00:07:07.880 Two of them are relatively well known, but all three are superhumans. s66 00:07:07.880 --> 00:07:10.240 Aren't they? s67 00:07:11.760 --> 00:07:24.240 So, in 2015 nine souls competed and the difference between them is only 27 hundredths of a second. s68 00:07:24.240 --> 00:07:26.800 And Gatlin lost the gold by 100th of a second. s69 00:07:26.800 --> 00:07:27.600 Can you believe it? s70 00:07:27.600 --> 00:07:30.520 It's not comprehensible to human mind. s71 00:07:30.520 --> 00:07:38.000 The scientific measurements gave that, okay, that is how Usain Bolt won over Gatlin that night. s72 00:07:38.000 --> 00:07:41.600 On another night the order could be completely different. s73 00:07:41.600 --> 00:07:46.800 I still run those 100 meters and I take 10 13 seconds, not 10. s74 00:07:46.800 --> 00:07:49.160 How about you? s75 00:07:49.240 --> 00:07:57.640 That's the difference between us and these extraordinary athletes competing at the edge of human capability. s76 00:07:57.640 --> 00:08:03.280 But as a fan boy, I will argue my favorites like I just did earlier. s77 00:08:03.280 --> 00:08:07.600 But I also speak to professionals in this game. s78 00:08:07.600 --> 00:08:18.400 And in their opinion, the gap is so tiny that the previous best or who holds the world record does not matter when they get onto the field. s79 00:08:18.400 --> 00:08:22.160 On a given day, anybody can beat anyone. s80 00:08:23.360 --> 00:08:28.160 I'm going to apply the same yardstick on models to drive a couple of points. s81 00:08:28.160 --> 00:08:39.039 Six labs the recent study came out from Stanford a few dozen Elo points here and there. s82 00:08:39.039 --> 00:08:42.960 And the leaderboard is changing with every release cadence. s83 00:08:42.960 --> 00:08:49.960 And the popular question this week which model are you using and how smarter did you feel by using that model? s84 00:08:49.960 --> 00:08:53.839 Isn't that the great dinner table argument these days? s85 00:08:54.240 --> 00:09:00.320 But I can tell you one thing, it is not necessarily a good boardroom argument. s86 00:09:00.320 --> 00:09:00.680 Why? s87 00:09:00.680 --> 00:09:05.720 Because often the emotional side of us, like the fan boy moment, peaks into meeting rooms. s88 00:09:05.720 --> 00:09:07.760 After all, we're all humans. s89 00:09:07.760 --> 00:09:10.720 I'm not saying don't compare models. s90 00:09:10.720 --> 00:09:20.960 All I'm saying is don't make a small moving frontier gap into your product roadmap as an advantage. s91 00:09:20.960 --> 00:09:25.320 Let's dive a little bit deeper here because there are some numbers to unpack. s92 00:09:25.320 --> 00:09:33.480 Anthropic 1503 Mistral 1460 87 points here and there. s93 00:09:33.480 --> 00:09:39.920 It does not mean Anthropic is 87 points smarter than Mistral. s94 00:09:39.920 --> 00:09:45.120 Actually, the scores, the way we should read them, is relative rating derived from model comparison. s95 00:09:45.120 --> 00:09:54.600 It actually tells us how these models rank relative to one another in that evaluation environment. s96 00:09:54.800 --> 00:09:58.360 It's not the absolute intelligence scale. s97 00:09:58.880 --> 00:10:00.880 But here is the elephant in the room. s98 00:10:00.880 --> 00:10:13.240 In all these conversations across multiple teams, I have seen multiple organizations, while we think we are comparing models, what I've observed is many times, they're not even comparing models. s99 00:10:13.240 --> 00:10:18.280 We mistakenly compare the harness or the product for models. s100 00:10:18.280 --> 00:10:23.200 The Claude code versus Cursor or Codex isn't a model comparison. s101 00:10:23.200 --> 00:10:26.840 It's closer to comparing Jamaica versus USA. s102 00:10:26.840 --> 00:10:31.000 You're comparing an athlete and a training system around the athlete. s103 00:10:31.000 --> 00:10:33.680 So, the model is the athlete or the engine. s104 00:10:33.680 --> 00:10:36.440 It provides the underlying capability. s105 00:10:36.440 --> 00:10:41.560 The harness determines how this engine is used. s106 00:10:41.560 --> 00:10:51.440 The tools, context, instructions, orchestration, memory, workflows, the MCP tool callings, that's the harness around it. s107 00:10:51.440 --> 00:10:56.520 The product determines why anyone should care because that's the end product. s108 00:10:56.520 --> 00:11:04.560 Today, you and I can rent the engine and very much all the harnesses that come with engine. s109 00:11:04.560 --> 00:11:08.720 Some come tightly coupled, some are loosely coupled. s110 00:11:09.320 --> 00:11:17.600 So, if everybody has the same source, like we saw in the previous picture, where is the differentiation? s111 00:11:18.680 --> 00:11:32.480 The one that used to come, which made these different companies thrive in different areas, the source or the first principles of that is from the diversity and the variance of their engineer minds. s112 00:11:32.560 --> 00:11:44.400 Given the model is rented, increasingly, what we are seeing is that once the capability is available to everyone, then what are we staring at? s113 00:11:45.880 --> 00:11:52.640 The model and the hardness do not care who is typing and from which part of the world. s114 00:11:52.720 --> 00:11:59.600 Or whether you are a E6, E7, or a E1 engineer. s115 00:11:59.680 --> 00:12:03.840 Find at the same problem, same tools. s116 00:12:03.840 --> 00:12:07.080 It gives the next person what it gave you. s117 00:12:07.080 --> 00:12:10.400 And I want to call this answer as a median. s118 00:12:10.400 --> 00:12:15.320 And this median keeps changing with every release cadence, as we saw in the diagram there. s119 00:12:15.320 --> 00:12:24.240 The previous problems solved get into these LLMs, and then incrementally, that's what these releases come one after another. s120 00:12:24.240 --> 00:12:27.120 Just like software cadence. s121 00:12:27.240 --> 00:12:34.680 I'm super simplifying here, but in a nutshell, that's what keeps the median moving up front, which is a good thing. s122 00:12:35.160 --> 00:12:40.480 So, at the same time, it's also a kind of a trap. s123 00:12:40.480 --> 00:12:51.240 Because it is good enough to ship, but it is hard to tell apart that what the next person is going to be shipped. s124 00:12:51.520 --> 00:12:59.120 Because the next person or the competitor could have also shipped something similar or exactly the same. s125 00:12:59.120 --> 00:13:02.760 This could not have happened before 2023. s126 00:13:03.920 --> 00:13:05.480 The median product was expensive. s127 00:13:05.480 --> 00:13:11.240 You needed a team, time, stack. s128 00:13:11.240 --> 00:13:18.240 Today, that solved problem is free. s129 00:13:18.240 --> 00:13:23.360 And the problem that is free is free for everyone. s130 00:13:23.360 --> 00:13:27.440 And you go out to the customer like how everybody goes out. s131 00:13:27.440 --> 00:13:38.280 And then this problem of so many products which are similar flooding the market then the distribution becomes the moat, not your product. s132 00:13:38.360 --> 00:13:43.520 So, what is that moat that you can create in this new environment? s133 00:13:43.520 --> 00:13:57.880 I'm going to quickly go through this GitHub's own number to reiterate the point that this model unharness are producing enormous amount of new software. s134 00:13:57.880 --> 00:14:03.120 And to me these are really good software. s135 00:14:03.800 --> 00:14:06.240 But the outcomes are similar. s136 00:14:06.240 --> 00:14:08.440 This is not crowd of copycats. s137 00:14:08.440 --> 00:14:09.800 Definitely not. s138 00:14:09.800 --> 00:14:17.240 It's a crowd of people who have arrived at the same answer because the source or the cookie cutter is the same. s139 00:14:17.720 --> 00:14:24.200 And the word that our industry choose for what they built is AI slop. s140 00:14:24.240 --> 00:14:26.000 The software AI slop. s141 00:14:26.000 --> 00:14:30.839 I want to take that word away from you and give you a new meaning to it. s142 00:14:30.839 --> 00:14:32.600 Let me be fair first. s143 00:14:32.600 --> 00:14:42.080 If you have been developing for a long time and you stumbled upon this AI assisted development it felt like magic, isn't it? s144 00:14:42.520 --> 00:14:45.280 Because we were taking a black box, pointing something to it. s145 00:14:45.280 --> 00:14:49.839 You would have definitely felt at some point in time, "Oh, it did really well." s146 00:14:49.839 --> 00:14:51.600 Two weeks ago I shipped an iOS app. s147 00:14:51.600 --> 00:14:59.160 I've been a developer, but I've never dreamt of developing an app because remember we had lanes, embedded developer, app developer. s148 00:15:00.640 --> 00:15:02.720 AI today is making everyone a developer. s149 00:15:02.720 --> 00:15:05.040 To me that's a good thing and I mean it. s150 00:15:05.040 --> 00:15:07.880 The word slop is negative. s151 00:15:07.880 --> 00:15:10.400 And they use it without any empirical analysis. s152 00:15:10.400 --> 00:15:22.839 Here is why I feel that those 100,000 of repos and proliferation of AI assisted development is bringing new people into the field who has never developed. s153 00:15:23.080 --> 00:15:30.640 The success metrics for these new products cannot be looked at from the lens of those experts. s154 00:15:30.760 --> 00:15:38.160 I would like to define the success of those million AI apps as long as they meet the success of their creators. s155 00:15:38.160 --> 00:15:41.920 Because who defines what a successful product looks like? s156 00:15:41.920 --> 00:15:44.680 For us, it could be something that brings a revenue. s157 00:15:44.680 --> 00:15:51.480 For someone else, it could be something that they can show it next to their colleagues that hey, see I've built this wonderful website. s158 00:15:51.480 --> 00:15:53.400 That's a success, too. s159 00:15:53.400 --> 00:15:59.920 And I've seen millions of people actually done that with lovable and Replit, which is cool. s160 00:15:59.920 --> 00:16:02.680 They are not AI snobs. s161 00:16:05.920 --> 00:16:11.920 What do you get when you point a good problem at a coding agent? s162 00:16:12.360 --> 00:16:23.920 Most times than not, today it's useful and the finished artifact may be similar because, you know, the same cutter went into the same dough. s163 00:16:23.920 --> 00:16:29.720 But most software developed this route is falling into what I call the median. s164 00:16:29.720 --> 00:16:32.680 I am not telling you to stop developing or shipping it. s165 00:16:32.680 --> 00:16:36.000 It's perfectly fine for internal consumption. s166 00:16:36.000 --> 00:16:43.240 The one caveat here is that stop counting this as a differentiating factor. s167 00:16:46.400 --> 00:16:48.960 But this median is lacking one thing. s168 00:16:48.960 --> 00:16:53.480 We need to also know the positives as well as the disadvantages. s169 00:16:53.480 --> 00:16:55.640 It's not a new problem. s170 00:16:55.640 --> 00:17:05.800 Back in '90s, everybody was trying to solve the search problem and most of them had the basic crawler and an index. s171 00:17:05.800 --> 00:17:16.560 And whoever is going to solve that problem really well or who has that novelty element is going to rise up because that's a differentiating factor. s172 00:17:16.560 --> 00:17:23.920 So, today I say the median of most software that you can think of building is actually free. s173 00:17:23.920 --> 00:17:29.760 But, if you want to compete in the open market, you need that element of novelty. s174 00:17:32.520 --> 00:17:38.040 So, some credibility before I go further because I've spoke almost for a long time. s175 00:17:38.040 --> 00:17:40.720 I did not start building when the models got better. s176 00:17:40.720 --> 00:17:47.280 I have built more than a dozens of applications in public and contributed to Lama index another repos. s177 00:17:47.280 --> 00:17:53.960 The last 3 years have been quite exciting and that brought me to this topic and tiring at the same time. s178 00:17:53.960 --> 00:18:01.800 I want to show you this because the project I'm going to show you are not hobbyist first attempts. s179 00:18:01.800 --> 00:18:05.160 I tried to solve a real-world problem and I ended up with a median. s180 00:18:05.160 --> 00:18:07.920 I never expected that. s181 00:18:10.200 --> 00:18:12.160 So, I want to coin few terms here. s182 00:18:12.160 --> 00:18:14.560 The build first reflex. s183 00:18:14.560 --> 00:18:25.640 When we all got excited with this AI and things, we stopped doing something which is back in the days when building was costly and difficult, I used to search s184 00:18:25.640 --> 00:18:27.040 more than I built. s185 00:18:27.040 --> 00:18:29.080 Today, it's flipped. s186 00:18:29.080 --> 00:18:39.120 You point a problem at a coding agent is more easier than going and searching and reading through those verbose whatever comes out and making a decision. s187 00:18:39.120 --> 00:18:41.160 We don't want to make decisions anymore. s188 00:18:41.160 --> 00:18:45.840 And this is getting the new normal in our craft. s189 00:18:45.840 --> 00:18:47.720 So, what do we do? s190 00:18:47.720 --> 00:18:51.520 Let's ask the coding agent to build a rack system. s191 00:18:52.160 --> 00:18:55.280 And I call this the build first reflex. s192 00:18:55.800 --> 00:18:59.560 Open a problem, open an agent. s193 00:18:59.560 --> 00:19:09.320 Yeah, you get the scaffoldings in the first 2 minutes and I'm sure if you've been used to this kind of development, you would do a demo in 2 hours, working demo. s194 00:19:09.320 --> 00:19:13.080 Where you give it to someone with rough edges, it works. s195 00:19:13.080 --> 00:19:16.240 And you could actually launch within 4 weeks. s196 00:19:16.240 --> 00:19:19.720 A real good product with support and everything. s197 00:19:19.920 --> 00:19:24.960 But only then you find out that, okay, I have built the 400th or 4,000th rag system. s198 00:19:24.960 --> 00:19:28.160 I took rag as an example here. s199 00:19:28.840 --> 00:19:32.800 So, I pointed the same problem and this is what we got. s200 00:19:32.800 --> 00:19:43.160 And I asked, "Hey, why did you build the nth rag system when more mature, feature-rich, open-source systems are available with thousands of stars and maturity? s201 00:19:43.160 --> 00:19:44.440 We could have just leveraged it. s202 00:19:44.440 --> 00:19:46.680 We should not have built it. s203 00:19:46.680 --> 00:19:49.880 3 years ago, we wouldn't have built it." s204 00:19:50.600 --> 00:19:55.400 So, it accepted that, "Hey, okay, I'm the culprit here. s205 00:19:55.400 --> 00:19:57.160 You said, 'Can you build?' s206 00:19:57.160 --> 00:19:58.440 I went ahead and built. s207 00:19:58.440 --> 00:20:02.080 If you had told me this, then I would have done something different." s208 00:20:02.080 --> 00:20:08.560 The median rag is what actually built, which is a solved problem. s209 00:20:08.680 --> 00:20:17.800 But if you're building it for the first time, it's going to be exciting for you because you're going to feel that I gave a PDF, I pointing it to a back-end LLM, s210 00:20:17.800 --> 00:20:19.440 and I'm asking the questions. s211 00:20:19.440 --> 00:20:28.480 The chunking and all those factors, which I understand or do not understand, it's going to solve it for me, the vector DB, and I got a solution. s212 00:20:28.480 --> 00:20:36.520 Small wins produce a lot of dopamine, and I fall for the same things as you are. s213 00:20:36.520 --> 00:20:38.760 And this is the verbose response. s214 00:20:38.760 --> 00:20:42.000 I defined that on the previous slide. s215 00:20:42.320 --> 00:20:48.520 Now, I'll stop talking about those trivial problems and show one of my applications. s216 00:20:48.520 --> 00:20:57.840 This one project is one of the representatives of the first 2 years, where I developed a lot of apps. s217 00:20:58.640 --> 00:21:03.360 Like you all know, my build-first reflex kicks in. s218 00:21:03.360 --> 00:21:07.760 I did this a lot of times to coin this term. s219 00:21:07.880 --> 00:21:12.880 I built it by sacrificing going out to play. s220 00:21:12.880 --> 00:21:17.240 I skipped many socials and movie nights. s221 00:21:17.240 --> 00:21:18.640 And I was building like crazy. s222 00:21:18.640 --> 00:21:23.480 It was like playing a game of chess where you are addicted to it. s223 00:21:23.480 --> 00:21:27.160 And I had every AI platform you could imagine. s224 00:21:27.160 --> 00:21:28.160 Max plan. s225 00:21:28.160 --> 00:21:30.640 Sometimes free. s226 00:21:32.400 --> 00:21:36.360 After all that effort, I finally soft launched. s227 00:21:36.400 --> 00:21:39.040 I'm a bit of a perfectionist. s228 00:21:39.440 --> 00:21:49.040 So, until then I was bombastic because I could see my idea become something which is real in no time. s229 00:21:49.040 --> 00:21:55.640 Where as a product manager, it would have taken me a big team, architects. s230 00:21:55.640 --> 00:21:58.480 I was able to do it all alone. s231 00:21:59.800 --> 00:22:06.360 I was bombastic until I found out that I haven't calibrated to reality yet. s232 00:22:06.360 --> 00:22:14.280 The truth is my circle told me that they have seen similar calculators elsewhere. s233 00:22:15.560 --> 00:22:24.240 And before anyone thinks I stole from them or they from me, the surprising thing is that nobody copied from anybody. s234 00:22:24.240 --> 00:22:24.880 It's the same thing. s235 00:22:24.880 --> 00:22:26.840 It's a copycat work. s236 00:22:26.840 --> 00:22:34.880 When I analyzed some of those open and closed calculators, I found it same sometimes. s237 00:22:35.240 --> 00:22:37.640 Not even similar. s238 00:22:37.680 --> 00:22:43.440 And how could people across could arrive at similar solution? s239 00:22:43.440 --> 00:22:48.000 Even if you look at the LLM routing, everywhere is converging. s240 00:22:48.000 --> 00:22:51.840 Like those models there, even solutions are converging. s241 00:22:52.480 --> 00:22:56.280 And competing for a tiny space. s242 00:22:56.520 --> 00:23:03.760 And it's like having same developers in your team and your competitors also have the same developers. s243 00:23:05.760 --> 00:23:15.200 So, you can build intuition only through building and I'm sure you're able to those builders would agree with me. s244 00:23:15.200 --> 00:23:20.520 We now have something to recalibrate and find a successful path. s245 00:23:20.520 --> 00:23:31.720 So, after 2 years where I spent almost all time I was tired cuz I spent multiple Any exciting idea I had, I went ahead and built with a build first reflex. s246 00:23:31.720 --> 00:23:41.280 And almost every time I ended up with a crowded lane where there are already two or three people or I would be the first and I would take a month off and come back there would be 60 s247 00:23:41.280 --> 00:23:43.160 similar products. s248 00:23:43.160 --> 00:23:49.880 And whoever has a better distribution is obviously going to uh win that battle. s249 00:23:51.400 --> 00:23:54.800 So, I would like to coin another new term. s250 00:23:54.800 --> 00:23:57.000 Okay, the build first then what? s251 00:23:57.000 --> 00:24:03.200 In the AI software development life cycle, a novelty budget. s252 00:24:03.200 --> 00:24:13.680 So, if you can get at least one unit of true novelty like how that page rank was for Google, you can cross this median at least for a good amount of time. s253 00:24:13.680 --> 00:24:22.680 You pick a problem and you want to be the best solution for the problem in the world, then you have a good chance of beating the crowd. s254 00:24:22.680 --> 00:24:33.160 For the first 2 years I spent, you know, my days and nights on architecture, agent design, prompt chains, skills, and somebody would share, "Hey, this is my Claude new workflow. s255 00:24:33.160 --> 00:24:35.120 That's my Claude new workflow." s256 00:24:35.120 --> 00:24:37.200 I did all those things. s257 00:24:37.200 --> 00:24:45.280 And most of the times than not, trust me, you would end up with similar outputs if not same. s258 00:24:45.280 --> 00:25:00.280 So, I had to rework my calculator that I showed you and this time I spent almost all of the time on how to differentiate and almost none on flaunting my Claude code or code skills which I used to do earlier. s259 00:25:00.280 --> 00:25:02.040 Because that felt like magic. s260 00:25:02.040 --> 00:25:05.960 The median from model isn't opinionated. s261 00:25:05.960 --> 00:25:07.960 It is converged best practices. s262 00:25:07.960 --> 00:25:13.000 So I picked a lane and I treated it like a budget. s263 00:25:13.000 --> 00:25:21.280 After 2 years of burning several million tokens on many max plans, I feel everything else is free now. s264 00:25:21.280 --> 00:25:25.840 And the highest return of investment is a novelty budget. s265 00:25:25.840 --> 00:25:31.920 So before the bill starts, name the one thing that is novel. s266 00:25:31.920 --> 00:25:40.640 If you don't have one, then you can take the median because there are a lot of people who have built the solution for the problem that you have and I'm sure of it. s267 00:25:40.640 --> 00:25:44.080 You tell me a problem for which there are no solutions. s268 00:25:44.080 --> 00:25:46.720 I'll go and find it for you. s269 00:25:46.720 --> 00:25:52.560 And then put this novelty plan as part of your road map and your project planning. s270 00:25:52.560 --> 00:25:55.360 I do this now. s271 00:25:55.360 --> 00:26:01.520 And review it the way you actually review spending like a project. s272 00:26:03.320 --> 00:26:14.000 Honestly, if you cannot name the novelty in your project, then you should concede that it's a learning exercise. s273 00:26:14.000 --> 00:26:16.360 That's fine. s274 00:26:19.040 --> 00:26:26.080 Uh the failures, the learnings, and the mechanics to beat medians for 2 years, I came up with this four rules. s275 00:26:26.080 --> 00:26:29.280 And these are not side project rules. s276 00:26:29.280 --> 00:26:35.480 I want you to run them as your road map for the next quarter speeches. s277 00:26:35.480 --> 00:26:40.520 Because your products will definitely have a twin and sometimes it might be open source. s278 00:26:40.520 --> 00:26:43.600 Why take all the effort in building it? s279 00:26:43.600 --> 00:26:45.920 Leverage it, fork it. s280 00:26:45.920 --> 00:26:49.240 So one of the things is be highly opinionated. s281 00:26:49.240 --> 00:26:53.440 The AI assistants most of the times are going to give you a probable answers. s282 00:26:53.440 --> 00:26:55.720 Be it anything. s283 00:26:56.160 --> 00:26:59.800 And those things are going to be same for almost everybody. s284 00:26:59.800 --> 00:27:05.240 It might be similar, but that's in the spectrum of already solved. s285 00:27:05.240 --> 00:27:13.000 In the world of medians, an opinion is the cheapest novelty you could own. s286 00:27:13.000 --> 00:27:16.560 And go for it because the model cannot supply. s287 00:27:16.560 --> 00:27:18.560 Two. s288 00:27:18.600 --> 00:27:22.560 The prompt to production is a median path. s289 00:27:22.560 --> 00:27:24.920 And there's absolutely nothing wrong with it. s290 00:27:24.920 --> 00:27:31.480 If you feel that it's an internal consumption you want to build and your team has to use it, go for it. s291 00:27:31.480 --> 00:27:34.120 I'm all for it. s292 00:27:34.280 --> 00:27:36.360 And then the third. s293 00:27:36.360 --> 00:27:37.640 This is the most fascinating things. s294 00:27:37.640 --> 00:27:44.480 When AI says done, then it means that to me, okay, AI has given its median product. s295 00:27:44.480 --> 00:27:47.440 I have to start working from here. s296 00:27:47.440 --> 00:27:53.800 That's the signal from any of your coding agents because your work begins there. s297 00:27:53.800 --> 00:27:56.400 And remember this. s298 00:27:57.160 --> 00:28:05.080 These coding agents so far has given the last great design for free. s299 00:28:05.080 --> 00:28:12.600 It's not going to invent the next Kubernetes or Zookeeper or Kafka or whatever those building blocks are. s300 00:28:12.600 --> 00:28:21.120 It's going to do a scaffolding and it does a really good job in multiple places, but it's it's it's going to be the last great design for free. s301 00:28:21.120 --> 00:28:23.560 And what's your job? s302 00:28:23.560 --> 00:28:28.640 If you want to thrive, you need to find the next one. s303 00:28:29.800 --> 00:28:32.840 And here are some methods I found useful. s304 00:28:32.840 --> 00:28:40.560 I've shared this method with multiple leaders across organizations and I've got good feedback on them. s305 00:28:40.560 --> 00:28:47.200 If you are leading and don't build now with this AI, this is exactly for you. s306 00:28:47.200 --> 00:28:48.680 You'll be seeing a bunch of demos. s307 00:28:48.680 --> 00:28:51.480 I'm sure you're bombarded with them every week. s308 00:28:51.480 --> 00:29:07.240 And it looks like a big deal because I've seen the instincts of leadership were not calibrated because back when they were building their intuition the deliverable that they are seeing s309 00:29:07.240 --> 00:29:10.720 took them a quarter or more time and a team. s310 00:29:10.720 --> 00:29:22.120 Today a new engineer shows up and claims to have built the same that would have taken a team and a quarter uh to build it. s311 00:29:22.680 --> 00:29:29.480 So, it's one of the hardest time to be on deciding where to put your eggs on. s312 00:29:29.720 --> 00:29:32.640 Especially when you're not immersed in this tech. s313 00:29:32.640 --> 00:29:37.960 Because nobody can recalibrate by reading or watching YouTube or LinkedIn posts. s314 00:29:37.960 --> 00:29:40.080 You recalibrate by building. s315 00:29:40.080 --> 00:29:42.400 You build intuition by building. s316 00:29:42.400 --> 00:29:43.760 So, the playbook is simple. s317 00:29:43.760 --> 00:29:44.600 Three questions. s318 00:29:44.600 --> 00:29:49.480 It's not going to solve your problems, but these are good handy questions to ask. s319 00:29:49.680 --> 00:29:50.680 One. s320 00:29:50.680 --> 00:29:52.280 Who has already done this? s321 00:29:52.280 --> 00:30:03.720 This question forces the search because the world of exciting AI, we need to tame the build first reflex powered by AI. s322 00:30:03.720 --> 00:30:05.720 Two. s323 00:30:05.840 --> 00:30:15.000 What do we have that the next team or our competitors or another person in the team don't have? s324 00:30:15.000 --> 00:30:21.960 This forces novelty as a first-class citizen into the minds of the team members. s325 00:30:22.240 --> 00:30:29.800 Because a better prompt, somebody would say, it's just going to throw you a median problem median solution. s326 00:30:30.160 --> 00:30:36.200 And third, where will this application live? s327 00:30:36.680 --> 00:30:50.120 In the age of exciting AI, forcing outcome-based thinking when everyone is high on building, it's a way to show them the demo is the expensive way to find out s328 00:30:50.120 --> 00:30:53.120 if you have no distribution mode. s329 00:30:53.360 --> 00:30:57.480 And I applied all this rigorously in the last 1 year. s330 00:30:57.480 --> 00:31:03.496 And you may be asking, what happened to the one project that I was talking to you? s331 00:31:03.496 --> 00:31:04.280 [snorts] s332 00:31:04.280 --> 00:31:06.800 I asked these questions. s333 00:31:06.960 --> 00:31:19.480 Okay, if I have to be providing that value, which is above median, then what stands between an open model and a token generating system? s334 00:31:19.480 --> 00:31:22.480 And what cost does an enterprise have to pay for it? s335 00:31:22.480 --> 00:31:30.720 Because open weights are at rage, but open weights are not open inferences. s336 00:31:31.600 --> 00:31:41.480 And I found that the hard way because there are 10 layers that sit between an open weight and a system that generates these tokens. s337 00:31:41.480 --> 00:31:54.160 We take the labs for granted for providing us there is ML ops, there is SRE on a different scale altogether that the world has not seen. s338 00:31:54.160 --> 00:32:01.840 So, precision KV cache, uh batching, execution parallelism, and there are different slice and dices of them. s339 00:32:01.840 --> 00:32:04.360 So, some of them you have to hire experts. s340 00:32:04.360 --> 00:32:13.280 And I don't know where these experts come from because this is not a tech or a skill that the industry already has. s341 00:32:14.600 --> 00:32:17.440 And some of them, layer nine, is utilization. s342 00:32:17.440 --> 00:32:18.440 It is structural. s343 00:32:18.440 --> 00:32:21.000 You cannot buy it. s344 00:32:23.040 --> 00:32:36.520 So, unless like I said, you take one problem and want it to be the best at solving that particular problem, you're not going to cross the median just by using these AI tools. s345 00:32:36.800 --> 00:32:44.520 And I searched and I found that none of the 244 calculator that I was showing you earlier did this. s346 00:32:44.640 --> 00:33:03.200 And I changed the entire application into kind of what the identical weights cost to run inside your enterprise versus running it in the Neo cloud and taking tokens. s347 00:33:03.480 --> 00:33:06.480 That is what this calculator gives you. s348 00:33:06.480 --> 00:33:14.280 Because if you have open models and want to go and build it, then what's the difference? s349 00:33:14.280 --> 00:33:24.720 And received great feedback from a lot of enterprises who want to actually go that path because when people say open models, they're excited. s350 00:33:24.720 --> 00:33:26.840 For what? s351 00:33:28.440 --> 00:33:34.320 Organizations considering deploying open weight models, the weights are free. s352 00:33:34.320 --> 00:33:35.720 But you may want to check. s353 00:33:35.720 --> 00:33:37.640 That's the QR code. s354 00:33:37.640 --> 00:33:43.240 So, scan it, put it into your workload, and see which way it goes for you. s355 00:33:43.240 --> 00:33:46.520 It's open source and free. s356 00:33:49.240 --> 00:33:54.320 So, here is where it all lands and ends. s357 00:33:54.400 --> 00:34:04.280 With the power of the LLMs on the harness, the last 3 years building has got exponentially easier. s358 00:34:04.320 --> 00:34:14.760 But the changes and the dynamics of software development still hasn't changed at the end. s359 00:34:14.760 --> 00:34:17.320 The prompt-based median products are useful. s360 00:34:17.320 --> 00:34:18.000 I ship them, too. s361 00:34:18.000 --> 00:34:20.159 They're not slops. s362 00:34:20.159 --> 00:34:23.440 But they have little durability when they go to open competition. s363 00:34:23.440 --> 00:34:27.120 And they may survive if you have a huge distribution mode. s364 00:34:27.120 --> 00:34:40.960 So, the one attitude if you want to try thrive in this field is again I'm I may be repeating this, but take that one problem and pursue it with the intent of s365 00:34:40.960 --> 00:34:44.399 being the best at it in the entire world. s366 00:34:44.399 --> 00:34:52.040 Yeah, we may not be able to crack it, but that intent will definitely put you above median. s367 00:34:54.560 --> 00:34:59.760 The only question left is can your AI-assisted products thrive? s368 00:34:59.760 --> 00:35:03.440 I spent 2 years on the median, but that was education. s369 00:35:03.440 --> 00:35:06.200 It was worth every token I burnt it. s370 00:35:06.200 --> 00:35:08.480 I asked two questions at the start. s371 00:35:08.480 --> 00:35:11.440 What happens to the software? s372 00:35:11.440 --> 00:35:17.240 To me, the software that is running this 100 trillion-dollar economy, nothing happens to it. s373 00:35:17.240 --> 00:35:24.720 The world still runs on deterministic software, at least for the foreseeable future, and it might always be. s374 00:35:24.720 --> 00:35:36.160 The age of AI means that we can write it more faster, but it is not going to take out the deterministic software. s375 00:35:38.080 --> 00:35:53.360 So, I would like to wrap up by saying that spend everything you have on one thing, which is the novelty. s376 00:35:53.360 --> 00:36:00.880 Squeeze every vein, every ounce of your brain on novelty, because novelty is scarce, and the median is free. s377 00:36:00.880 --> 00:36:03.160 It's up to you to choose. s378 00:36:03.160 --> 00:36:05.360 Thank you.