1 00:00:00,020 --> 00:00:03,990 Ejaaz: 99% of people are using AI models the same way that they use Google. 2 00:00:03,990 --> 00:00:08,140 Ejaaz: But recently, a new way of prompting your AI has emerged that doesn't just replace 3 00:00:08,140 --> 00:00:12,530 Ejaaz: the way that you work, it promotes you to the CEO of your very own AI company. 4 00:00:12,730 --> 00:00:16,520 Ejaaz: It's called Loops and it's part of a growing development in agent autonomy where 5 00:00:16,520 --> 00:00:21,160 Ejaaz: AI agents basically spin up and autonomously complete tasks or goals that you 6 00:00:21,160 --> 00:00:23,080 Ejaaz: set for it, often working throughout the night. 7 00:00:23,200 --> 00:00:28,610 Ejaaz: In 2019, the longest that an AI agent could work autonomously for was for two seconds. 8 00:00:28,800 --> 00:00:32,150 Ejaaz: Fast forward to today, and they can work autonomously for 12 hours, 9 00:00:32,150 --> 00:00:34,200 Ejaaz: and that's doubling every couple of months. 10 00:00:34,200 --> 00:00:38,290 Ejaaz: Andre Carpathy calls this phenomenon the autonomy slider, where you can go take 11 00:00:38,290 --> 00:00:43,010 Ejaaz: a dial that slides from humans that approve everything to humans that periodically check in. 12 00:00:43,010 --> 00:00:47,300 Ejaaz: And it's part of this growing trend of agents consuming and taking up more of 13 00:00:47,300 --> 00:00:52,480 Ejaaz: human capital and labor. And the question that remains going forwards is, what will humans do? 14 00:00:52,690 --> 00:00:54,510 Ejaaz: And will they be entirely replaced 15 00:00:54,510 --> 00:00:58,000 Ejaaz: by AI? Or will they be the ultimate orchestrator of their destiny? 16 00:00:58,300 --> 00:01:02,030 Josh: Yeah, I think the goal for this episode is really just to inform people on what's 17 00:01:02,030 --> 00:01:06,030 Josh: possible current day with these agents, with these LLMs, with writing these 18 00:01:06,030 --> 00:01:10,360 Josh: loops, as well as where you can possibly find yourself within that stack, 19 00:01:10,360 --> 00:01:11,960 Josh: because it gets pretty complicated. 20 00:01:11,960 --> 00:01:14,600 Josh: When we're getting into loops, not everyone needs to use loops, 21 00:01:14,600 --> 00:01:19,040 Josh: but everyone should be using LLMs probably slightly different than how you're using them today. 22 00:01:19,080 --> 00:01:23,810 Josh: So maybe we could start with a little history lesson in terms of the four levels 23 00:01:23,810 --> 00:01:28,990 Josh: in which we have been engaging with llm starting with the first level which is just prompting, 24 00:01:29,480 --> 00:01:33,800 Josh: generally like most people are probably still doing this started in 2022 2023 25 00:01:34,070 --> 00:01:35,480 Josh: around the release of chat gpt 26 00:01:35,840 --> 00:01:39,160 Josh: the way that you would engage with these llms is you would just submit a question 27 00:01:39,160 --> 00:01:43,020 Josh: or you submit a prompt and you get some language back now if you are still doing 28 00:01:43,020 --> 00:01:45,750 Josh: this that's okay because i find a. 29 00:01:50,260 --> 00:01:53,640 Josh: Three years ago, four years ago. It has since advanced pretty, 30 00:01:54,070 --> 00:01:55,170 Josh: pretty meaningfully since then. 31 00:01:55,450 --> 00:01:58,520 Josh: The second step of this is agents. And we're going to spend some time on agents. 32 00:01:59,110 --> 00:02:01,970 Josh: Everyone's kind of heard of an agent. Maybe not everyone knows what an agent is. 33 00:02:02,190 --> 00:02:05,170 Josh: An agent is something that could think for a little bit longer. 34 00:02:05,170 --> 00:02:08,180 Josh: It could run a bit longer than just a standard prompt. 35 00:02:08,460 --> 00:02:10,980 Josh: It can go off and do things. It could call tools for you. 36 00:02:11,630 --> 00:02:15,380 Josh: It's a much more capable version of the text box. Then, like we talked about 37 00:02:15,380 --> 00:02:17,840 Josh: all the time on the show recently in the last few weeks, there's the harness 38 00:02:18,050 --> 00:02:23,170 Josh: feature in which you put an LLM into a container and that gives it a memory feature. 39 00:02:23,170 --> 00:02:26,450 Josh: That gives it complete tool use. That's something like an open claw that we've 40 00:02:26,450 --> 00:02:29,160 Josh: talked about a lot that some people do use and that's level three. 41 00:02:29,160 --> 00:02:32,140 Josh: And now level four, which is the new thing that has come this week, 42 00:02:32,410 --> 00:02:36,200 Josh: that's really been highlighted by some of the top leaders at these AI labs is loops. 43 00:02:36,460 --> 00:02:41,770 Josh: And a loop is essentially a version of an agent that has an orchestration layer 44 00:02:41,770 --> 00:02:43,570 Josh: and kind of builds upon itself. 45 00:02:43,570 --> 00:02:47,390 Josh: So it allows you to kind of continue to scope yourself out. If you can imagine 46 00:02:47,390 --> 00:02:50,950 Josh: you're kind of you're dealing directly with an employee at level one and then 47 00:02:50,950 --> 00:02:54,280 Josh: you're kind of directing that person to go off and do their own in level two, 48 00:02:54,770 --> 00:02:58,770 Josh: At level three with the harness, you're kind of directing a series of people to help you. 49 00:02:59,020 --> 00:03:02,920 Josh: And then level four, you're just the top level CEO who's directing your C-suite 50 00:03:02,920 --> 00:03:06,280 Josh: to go and manage all the employees below you. So there's an entire stack to this. 51 00:03:06,510 --> 00:03:10,420 Josh: It's very cool. It just how do you use your AI currently? Where would you say 52 00:03:10,420 --> 00:03:11,750 Josh: that you fit in this stack? 53 00:03:12,000 --> 00:03:16,020 Ejaaz: Yeah, so looking at this diagram that we have on the screen here, 54 00:03:16,020 --> 00:03:19,850 Ejaaz: I'm somewhere between number two and number three. I'm somewhere between using 55 00:03:19,850 --> 00:03:22,870 Ejaaz: agents and trying to figure out the whole harness thing. 56 00:03:23,210 --> 00:03:26,820 Ejaaz: Now, what am I doing when it comes to like spinning up agents? 57 00:03:27,070 --> 00:03:31,250 Ejaaz: If you look at either my Claude or my ChatGPT desktop apps right now, 58 00:03:31,910 --> 00:03:37,830 Ejaaz: I've renamed a bunch of my conversations to a particular focus or subject and then agent after it. 59 00:03:38,170 --> 00:03:42,480 Ejaaz: And so I can go to it and this agent basically has all the context of what I 60 00:03:42,480 --> 00:03:44,950 Ejaaz: wanted to do, whether it's like research a particular topic, 61 00:03:45,230 --> 00:03:48,870 Ejaaz: create some kind of an outline for something, research a particular investment angle. 62 00:03:49,010 --> 00:03:52,290 Ejaaz: It already knows and has the embedded context for what it needs to do. 63 00:03:52,530 --> 00:03:58,030 Ejaaz: And there's usually like one to maybe three tasks that it needs to autonomously execute on its own. 64 00:03:58,310 --> 00:04:02,230 Ejaaz: And so it runs in kind of like a sequence. But if any of that sequence kind 65 00:04:02,230 --> 00:04:06,210 Ejaaz: of breaks, let's say it kind of tries to retrieve data from some particular 66 00:04:06,210 --> 00:04:09,130 Ejaaz: website and it is unable to do so, it breaks. 67 00:04:09,130 --> 00:04:13,080 Ejaaz: And it comes to me and it says, hey, Ejaz, is there some other thing that you 68 00:04:13,080 --> 00:04:15,030 Ejaaz: want to look at or retrieve from, blah, blah, blah? 69 00:04:15,430 --> 00:04:19,930 Ejaaz: It's not fully autonomous. Now, number three, the Harness side of things is 70 00:04:19,930 --> 00:04:22,540 Ejaaz: what I'm trying to kind of like mold my understanding around. 71 00:04:22,860 --> 00:04:26,430 Ejaaz: What I've noticed is when you type in a prompt and you get a response, 72 00:04:26,750 --> 00:04:30,770 Ejaaz: you can kind of tell that it's AI-y. Like usually when we kind of create artifacts, 73 00:04:30,770 --> 00:04:34,490 Ejaaz: it comes in a particular font or it speaks in a particular type of language. 74 00:04:34,780 --> 00:04:38,640 Ejaaz: The Harness helps kind of like take your prompt and kind of mold it into something 75 00:04:38,640 --> 00:04:43,580 Ejaaz: that is more human-like, but also more nuanced with what you are trying to do. 76 00:04:43,580 --> 00:04:44,730 Ejaaz: Like it effectively gets 77 00:04:45,090 --> 00:04:49,010 Ejaaz: closer towards that ultimate goal. Like we were talking before recording this 78 00:04:49,010 --> 00:04:53,130 Ejaaz: episode about human taste and how AI doesn't really get human taste. 79 00:04:53,130 --> 00:04:57,690 Ejaaz: The harness helps you get towards that ultimate kind of taste profile for the 80 00:04:57,690 --> 00:04:59,200 Ejaaz: particular output that you're trying to generate. 81 00:04:59,970 --> 00:05:04,200 Ejaaz: I haven't tried working with loops just yet, but my understanding of this, 82 00:05:04,200 --> 00:05:10,110 Ejaaz: and correct me if I'm wrong, is you have an AI. You can prompt it and you can get some kind of output. 83 00:05:10,480 --> 00:05:15,740 Ejaaz: A loop specifically is an AI agent that doesn't break, if it comes across an 84 00:05:15,740 --> 00:05:19,980 Ejaaz: obstacle that it doesn't understand, its instinct isn't to come to the human 85 00:05:19,980 --> 00:05:22,930 Ejaaz: and say, hey, like, I can't figure this out, guide me. 86 00:05:23,260 --> 00:05:28,070 Ejaaz: It completely reiterates the prompt over and over again until it gets past that 87 00:05:28,070 --> 00:05:32,100 Ejaaz: obstacle, working towards like one objective. So a few examples I've seen for 88 00:05:32,100 --> 00:05:34,730 Ejaaz: this is if you are coding, right? 89 00:05:34,980 --> 00:05:38,860 Ejaaz: And let's say there's multiple workflows of a code base that you want to work 90 00:05:38,860 --> 00:05:42,730 Ejaaz: on, and it comes across a hiccup where it can't retrieve data from one of those 91 00:05:42,730 --> 00:05:45,860 Ejaaz: particular flows, it is able to kind of like circumnavigate around it, 92 00:05:46,070 --> 00:05:49,000 Ejaaz: maybe spin up its own separate flow and try to figure out the problem. 93 00:05:49,240 --> 00:05:53,040 Ejaaz: And often this results in an agent working for multiple hours at a time, 94 00:05:53,040 --> 00:05:57,590 Ejaaz: often overnight. I think Carpathy spoke about his auto research agent working 95 00:05:57,590 --> 00:05:58,730 Ejaaz: overnight whilst he slept. 96 00:05:59,020 --> 00:06:02,280 Ejaaz: And we're seeing different variations of this start to arise. 97 00:06:02,840 --> 00:06:04,430 Ejaaz: Where are you, Josh, in the stack? 98 00:06:04,980 --> 00:06:09,340 Josh: Yeah, loops are like the closed source a system where you kind of define an 99 00:06:09,340 --> 00:06:13,590 Josh: outcome and it will continue to work towards that outcome without any external inputs. 100 00:06:13,810 --> 00:06:16,090 Josh: It's very cool. It's very automated. I don't think it's for everyone. 101 00:06:16,090 --> 00:06:21,430 Josh: It's certainly not for me because I haven't really had a use case for loops per se. 102 00:06:21,430 --> 00:06:26,290 Josh: I would say I'm sitting at each one of those first three phases given whatever 103 00:06:26,290 --> 00:06:28,980 Josh: tasks I'm trying to do. And I think it's important to understand that a lot 104 00:06:28,980 --> 00:06:33,140 Josh: of people might not even need to go past number one unless you're actually doing productive work. 105 00:06:33,310 --> 00:06:37,780 Josh: A lot of the agents, a lot of the harnesses are for kind of automating more. 106 00:06:38,640 --> 00:06:41,920 Josh: More systems from your life if you're just trying to use this as google if you're 107 00:06:41,920 --> 00:06:46,570 Josh: just trying to use this as a writing assistant or someone to chat with the prompting 108 00:06:46,570 --> 00:06:48,500 Josh: is really strong and i find a lot of times 109 00:06:48,900 --> 00:06:53,150 Josh: this is my outlook or this is my outlet for like google search results so instead 110 00:06:53,150 --> 00:06:56,700 Josh: of searching for google i'll get a little more in-depth results i'll ask my llm 111 00:06:56,930 --> 00:07:00,090 Josh: for agents i use them quite a bit when i'm doing a little bit more productive 112 00:07:00,090 --> 00:07:04,050 Josh: work for example we track the analytics on limitless and we want a place in 113 00:07:04,050 --> 00:07:06,470 Josh: which we can have all those analytics dumped to a dashboard, 114 00:07:06,800 --> 00:07:08,100 Josh: that is an agent that I run. 115 00:07:08,390 --> 00:07:12,850 Josh: So it goes into my browser. It detects all of the views that we've had from 116 00:07:12,850 --> 00:07:16,660 Josh: the week for YouTube, from Spotify, from RSS feed, where you should all be subscribed 117 00:07:16,660 --> 00:07:18,010 Josh: to and rate us five stars. 118 00:07:18,230 --> 00:07:21,810 Josh: And it compiles it into a singular spreadsheet in which we could then publish 119 00:07:21,810 --> 00:07:25,190 Josh: online and we could share with prospective sponsors and things like that. 120 00:07:25,410 --> 00:07:28,410 Josh: And then for harnesses I've used, because I mean, that's mostly OpenClaw. 121 00:07:28,590 --> 00:07:34,040 Josh: I've used OpenClaw. I really enjoyed the process. I find myself using it a bit less and less. 122 00:07:34,460 --> 00:07:39,950 Josh: And I think in the loops feature, at least it's probably most productive right 123 00:07:39,950 --> 00:07:43,000 Josh: now for people who are writing code, who are writing verifiable solutions. 124 00:07:43,260 --> 00:07:46,180 Josh: One of the difficult things that as I was looking into loops and figuring out 125 00:07:46,180 --> 00:07:49,460 Josh: how I can structure them into my life, one of the problems that I run into is 126 00:07:49,460 --> 00:07:51,430 Josh: I'm not really sure I have a verifiable, 127 00:07:51,910 --> 00:07:55,500 Josh: set of outputs that I wanted to optimize for, for a lot of the work that I'm 128 00:07:55,500 --> 00:07:59,240 Josh: doing, because a lot of it is subjective. A lot of it is kind of creative work. 129 00:07:59,440 --> 00:08:01,540 Josh: It requires a human in the loop for a lot more of it. 130 00:08:01,920 --> 00:08:06,400 Josh: So I would say I am number one, two, and three on the list. Haven't quite made my way to four. 131 00:08:07,300 --> 00:08:11,010 Josh: But yeah, for the people who are, those are the people like Boris Churny from 132 00:08:11,010 --> 00:08:16,020 Josh: Anthropic. And we know Andre and Peter Steinberg from OpenAI. 133 00:08:16,020 --> 00:08:18,640 Josh: They are all on four. They are using it to, 134 00:08:19,390 --> 00:08:23,250 Josh: create these like unbelievable, agentic systems and continue to remove themselves out of the loop. 135 00:08:23,580 --> 00:08:28,150 Ejaaz: You know what I've realized? With loops in particular and just AI agents in 136 00:08:28,150 --> 00:08:35,050 Ejaaz: general, they're trying to improve our understanding or rather their understanding 137 00:08:35,050 --> 00:08:36,220 Ejaaz: of the English language. 138 00:08:36,220 --> 00:08:40,710 Ejaaz: So one of my favorite Carpathy quotes back in the day was English is the new 139 00:08:41,110 --> 00:08:44,020 Ejaaz: programming language. I think you said this like two, two and a half years ago. 140 00:08:44,780 --> 00:08:49,180 Ejaaz: And I've just realized that like us creating AI agents is basically like, 141 00:08:49,180 --> 00:08:52,170 Ejaaz: it's the same model. It hasn't necessarily got smarter. 142 00:08:52,380 --> 00:08:56,260 Ejaaz: It's just like using that model to kind of like keep ramming its head and its 143 00:08:56,260 --> 00:09:00,670 Ejaaz: brain against a particular problem until it understands what the human actually means. 144 00:09:00,900 --> 00:09:04,790 Ejaaz: And so like in this new world, like I know you just used the example of like, 145 00:09:04,790 --> 00:09:07,720 Ejaaz: you know, loops can be used for coding specifically, 146 00:09:08,360 --> 00:09:12,100 Ejaaz: coding that Boris Churny and Carpathy is doing is English. 147 00:09:12,100 --> 00:09:16,070 Ejaaz: Like they're speaking to the LLM, they are writing in English to the LLM. 148 00:09:16,070 --> 00:09:18,580 Ejaaz: And yeah, maybe they're copy and pasting some versions of code, 149 00:09:18,580 --> 00:09:21,450 Ejaaz: but that code is primarily generated by an AI. 150 00:09:21,450 --> 00:09:25,860 Ejaaz: I think like something crazy, like 80% plus of code generated at Anthropic, 151 00:09:25,860 --> 00:09:31,430 Ejaaz: both for research and for just general consumer adoption is generated by Claude itself. 152 00:09:31,430 --> 00:09:36,010 Ejaaz: And so that's one thing. The other thing is the model just not getting smarter 153 00:09:36,240 --> 00:09:39,480 Ejaaz: is a really interesting thing. Like typically in my head, I would think, 154 00:09:39,480 --> 00:09:43,000 Ejaaz: okay, you need a better model to be able to unlock some of these new features 155 00:09:43,000 --> 00:09:45,650 Ejaaz: like AI agents, autonomous loops, et cetera. 156 00:09:45,890 --> 00:09:49,530 Ejaaz: But really you could just take the same model, wrap a harness around it and 157 00:09:49,530 --> 00:09:53,980 Ejaaz: try to get it to understand what particular goal it's getting at and just run 158 00:09:53,980 --> 00:09:57,870 Ejaaz: that iteration over and over and over again until you get a better output. 159 00:09:58,610 --> 00:10:03,680 Ejaaz: And I guess this is the same concept as inference or reinforcement learning 160 00:10:03,680 --> 00:10:07,650 Ejaaz: where like we've found this trend of post-training of these AI models, 161 00:10:07,650 --> 00:10:12,110 Ejaaz: these AI models just getting smarter, not because they've got bigger GPUs or more expensive GPUs. 162 00:10:12,110 --> 00:10:15,130 Ejaaz: It's because you've just taken the same model and you've just run it through 163 00:10:15,130 --> 00:10:18,260 Ejaaz: a different reasoning framework over and over again until it can do a thing. 164 00:10:18,540 --> 00:10:23,190 Ejaaz: And this is the practical embellishment of it. I personally haven't found like 165 00:10:23,190 --> 00:10:25,180 Ejaaz: an obvious use case for loops either. 166 00:10:25,370 --> 00:10:29,910 Ejaaz: So either you and I are boxing ourselves into a particular realm and maybe someone 167 00:10:29,910 --> 00:10:32,550 Ejaaz: listening to this is using this for like their software engineering thing or 168 00:10:32,550 --> 00:10:35,530 Ejaaz: their marketing thing. But yeah, I guess that's where I sit right now. 169 00:10:35,790 --> 00:10:37,540 Josh: Well, I think it's probably a skill issue on both our parts. 170 00:10:37,540 --> 00:10:40,860 Josh: Like there is certainly a use case for us in which we can use a loop in which 171 00:10:40,860 --> 00:10:44,210 Josh: we can define this outcome, send an agent off to go do it, and it will iterate 172 00:10:44,210 --> 00:10:46,140 Josh: on itself until it comes to a conclusion. 173 00:10:46,430 --> 00:10:50,170 Josh: I think it's just so novel and so new. It's difficult to kind of understand 174 00:10:50,460 --> 00:10:53,070 Josh: why. And we have this really great chart on screen that you're showing now, 175 00:10:53,070 --> 00:10:55,260 Josh: which is the why now section of this. 176 00:10:55,550 --> 00:10:59,850 Josh: And it's because the duration of a task that these agents can run is so much 177 00:10:59,850 --> 00:11:01,460 Josh: longer than it used to be. 178 00:11:01,670 --> 00:11:06,220 Josh: I mean, in 2019 we have here, it was two seconds. This was well before ChatGPT. 179 00:11:06,470 --> 00:11:13,050 Josh: But even early last year, in 2025, the duration that an agent could run on one single task was 180 00:11:13,440 --> 00:11:17,020 Josh: less than an hour in length so there's only so many tokens it could generate 181 00:11:17,020 --> 00:11:19,990 Josh: there's only so much reasoning it can do and there's only so much iteration 182 00:11:19,990 --> 00:11:23,680 Josh: you could get over that hour time period let alone the amount of costs that 183 00:11:23,680 --> 00:11:24,640 Josh: these tokens are going to be, 184 00:11:25,100 --> 00:11:28,930 Josh: costing you if you're using like the api or anything like that now fast forward 185 00:11:28,930 --> 00:11:33,310 Josh: to today i mean the best models in the world they're getting days worth of runtime 186 00:11:33,540 --> 00:11:38,070 Josh: so they can really think deeply and continue to iterate on themselves over and over i see examples of um. 187 00:11:38,680 --> 00:11:43,970 Josh: Backslash goal on x all the time of people who have a problem whether it be 188 00:11:43,970 --> 00:11:47,130 Josh: an optimization problem where they have a bug that they need to fix and they'll 189 00:11:47,130 --> 00:11:49,200 Josh: put this backslash goal on it for 190 00:11:49,590 --> 00:11:53,830 Josh: however long it needs to and it'll think for three four even five days i've 191 00:11:53,830 --> 00:11:57,960 Josh: seen in order to optimize for the specific parameter and this is possible because 192 00:11:57,960 --> 00:11:59,730 Josh: these models now can think for days long, 193 00:12:00,290 --> 00:12:04,940 Josh: you have to assume months is coming what does it look like Like when an agent can think for months. 194 00:12:04,940 --> 00:12:10,770 Josh: I mean, it's a really interesting paradigm shift that I'm not sure where people 195 00:12:10,770 --> 00:12:15,090 Josh: are going to find value in the open-ended way that it exists today, 196 00:12:15,090 --> 00:12:17,650 Josh: right? It's like, okay, here's this agent. 197 00:12:17,880 --> 00:12:19,960 Josh: You can tell to do whatever you want. You can create a loop. 198 00:12:19,960 --> 00:12:22,040 Josh: You can create an infrastructure system for it to operate in. 199 00:12:23,000 --> 00:12:25,760 Josh: It's pretty much open-ended and it's on you. And I think the answer to that 200 00:12:25,760 --> 00:12:30,600 Josh: is that not even the AI companies really understand the best use cases for it quite yet. 201 00:12:30,890 --> 00:12:33,700 Josh: I would imagine it's still this really difficult thing of how do you unlock 202 00:12:33,700 --> 00:12:37,350 Josh: value from essentially an open-ended agent that can go and run for an infinite 203 00:12:37,350 --> 00:12:38,700 Josh: amount of time? I don't know. 204 00:12:39,640 --> 00:12:45,180 Ejaaz: I also question like what a human's purpose would be at that point. 205 00:12:45,310 --> 00:12:51,210 Ejaaz: Like if you automate enough of the thinking and the curiosity behind like solving 206 00:12:51,210 --> 00:12:54,280 Ejaaz: particular problems, What do humans end up doing at that point, 207 00:12:54,280 --> 00:12:56,330 Ejaaz: especially if they don't do the work themselves? 208 00:12:56,330 --> 00:12:59,570 Ejaaz: They don't understand it, right? You need an AI to kind of like understand what 209 00:12:59,700 --> 00:13:01,300 Ejaaz: on earth is going on in the first place. 210 00:13:01,540 --> 00:13:05,150 Ejaaz: And eventually like an AI will then start setting goals, like more ambitious 211 00:13:05,150 --> 00:13:09,250 Ejaaz: goals than a human can in terms of like what to like kind of solve or go after. 212 00:13:09,980 --> 00:13:15,950 Ejaaz: There were some very low-level examples that I saw in response to Pete Steyer's tweet about loops. 213 00:13:16,260 --> 00:13:19,650 Ejaaz: And there's some kind of concrete examples that I want to run through very quickly here. 214 00:13:19,930 --> 00:13:25,530 Ejaaz: So one of them is using it for code, right? So a classic loop could basically 215 00:13:25,530 --> 00:13:30,360 Ejaaz: look like, okay, can you please pull live errors for my particular app? 216 00:13:30,630 --> 00:13:34,260 Ejaaz: Can you inspect and figure out where the bug might particularly be? 217 00:13:34,500 --> 00:13:38,760 Ejaaz: Can you create then a fix for this particular bug in my code? 218 00:13:39,010 --> 00:13:41,870 Ejaaz: And then can you deploy it? Then can you check the health of that deployment 219 00:13:41,870 --> 00:13:43,550 Ejaaz: and make sure that nothing else is broken? 220 00:13:43,810 --> 00:13:47,390 Ejaaz: And then record what failed and feed that into a database so that in the future, 221 00:13:47,390 --> 00:13:51,380 Ejaaz: we can detect errors like this or prevent it when we code and build some of 222 00:13:52,480 --> 00:13:53,670 Ejaaz: These future app features. 223 00:13:53,930 --> 00:13:58,490 Ejaaz: Now, that is kind of like a very small and specific enough use case that can 224 00:13:58,490 --> 00:14:02,800 Ejaaz: be generalized across basically any app or software engineering project that 225 00:14:02,800 --> 00:14:04,690 Ejaaz: you might be working on if you're listening to this. 226 00:14:05,380 --> 00:14:11,220 Ejaaz: And I wonder how many hours worth of engineering time that this replaces. 227 00:14:11,220 --> 00:14:13,990 Ejaaz: Because I know that there are entire teams having worked at companies. 228 00:14:14,420 --> 00:14:18,650 Ejaaz: Been a product manager in the past, entire teams of software engineers that 229 00:14:18,650 --> 00:14:21,840 Ejaaz: spend their entire days working on something like that. So that's one thing. 230 00:14:22,180 --> 00:14:25,720 Ejaaz: And then for content, which is very applicable for product managers, 231 00:14:25,720 --> 00:14:29,700 Ejaaz: or even like the work that you and I do, Josh, an agent could read a PRD. 232 00:14:29,700 --> 00:14:32,950 Ejaaz: So which is a product requirement doc, which is usually kind of like created 233 00:14:33,120 --> 00:14:36,320 Ejaaz: for a strategic goal that you want to kind of like build at your company, 234 00:14:36,320 --> 00:14:40,610 Ejaaz: like a product or a feature, it then writes whatever that next asset could be. 235 00:14:40,610 --> 00:14:43,470 Ejaaz: So it could be like a design profile or a mockup of what that feature might 236 00:14:43,470 --> 00:14:47,600 Ejaaz: look like, score it against like some kind of criteria that the company has 237 00:14:47,600 --> 00:14:50,960 Ejaaz: across like, you know, it must follow our vision, A, B, and C. 238 00:14:51,130 --> 00:14:53,980 Ejaaz: It must also look a particular way. This is our design profile, 239 00:14:53,980 --> 00:14:56,200 Ejaaz: our brand kind of profile and our aesthetic. 240 00:14:56,440 --> 00:15:00,510 Ejaaz: And then it kind of like updates its progress depending on like what other teams 241 00:15:00,510 --> 00:15:03,260 Ejaaz: have shipped. So maybe it's dependent on a particular feature. 242 00:15:03,470 --> 00:15:07,810 Ejaaz: And so it updates itself autonomously like that. Now, this all sounds very vague 243 00:15:07,810 --> 00:15:12,370 Ejaaz: intentionally because it's meant to apply to your particular business or your particular project. 244 00:15:12,370 --> 00:15:14,260 Ejaaz: But make no mistake, this is what 245 00:15:14,260 --> 00:15:18,770 Ejaaz: a lot of humans are paid upwards of six figures to do on a daily basis. 246 00:15:18,770 --> 00:15:22,960 Ejaaz: It's that nuance. And we're starting to see basically AI models and AI agents 247 00:15:23,230 --> 00:15:28,580 Ejaaz: enter into that human taste profile. So when I think about where we end up eventually, 248 00:15:29,860 --> 00:15:33,390 Ejaaz: There's a common argument that's made that it's like, oh, humans will always have the taste. 249 00:15:33,650 --> 00:15:36,760 Ejaaz: They'll always be able to kind of direct where the AI should go because we are 250 00:15:36,760 --> 00:15:39,430 Ejaaz: this all being kind of like smart kind of entity. 251 00:15:39,780 --> 00:15:44,800 Ejaaz: But I see increasingly AI stepping into that boundary and becoming the tastemaker 252 00:15:45,000 --> 00:15:46,320 Ejaaz: for all of the work that we end up doing. 253 00:15:47,070 --> 00:15:51,810 Josh: I still believe that to be true, that humans in the loop are critically important 254 00:15:51,810 --> 00:15:55,520 Josh: to applying human taste. I saw this great chart. I have no idea where it is. 255 00:15:55,810 --> 00:15:59,650 Josh: Somewhere in the depths of X. But basically, it was showing that in the App 256 00:15:59,650 --> 00:16:01,810 Josh: Store, the iOS App Store, where everyone downloads their apps, 257 00:16:02,140 --> 00:16:06,090 Josh: the amount of apps that have gone into production that have been published recently has gone vertical. 258 00:16:06,330 --> 00:16:10,140 Josh: I think it's doubled or tripled over the last six months. Everybody's publishing apps at the App Store. 259 00:16:10,590 --> 00:16:15,580 Josh: The amount of five-star reviews and the amount of downloads has actually either 260 00:16:15,580 --> 00:16:17,440 Josh: stayed flat or gone down. 261 00:16:17,760 --> 00:16:20,820 Josh: It has not matched the amount of new apps that are going to the app store. 262 00:16:21,000 --> 00:16:26,520 Josh: Why is this? It's because a lot of the apps don't have enough care applied to 263 00:16:26,520 --> 00:16:28,910 Josh: them. They're just not great applications. And when I think about, 264 00:16:29,670 --> 00:16:33,670 Josh: how I use my phone on a regular device or on a regular day or how I use my laptop 265 00:16:33,670 --> 00:16:37,100 Josh: and the applications that I actually spend time on, there's a very fixed set 266 00:16:37,100 --> 00:16:41,130 Josh: of them. And I'm a little stubborn when it comes to downloading new ones because 267 00:16:41,130 --> 00:16:43,010 Josh: a lot of the new ones just are not great. 268 00:16:43,350 --> 00:16:47,330 Josh: And I think a lot of that comes from this, this lack of care that is presented 269 00:16:47,330 --> 00:16:51,360 Josh: from AI outputs, where if you're optimizing for a specific parameter that you 270 00:16:51,360 --> 00:16:56,700 Josh: can measure, it's going to do it great, but it doesn't understand the subtle nuances of how humans 271 00:16:57,130 --> 00:17:00,800 Josh: engage and how they really love to use these products like one of the products 272 00:17:00,800 --> 00:17:04,340 Josh: that i use totally unrelated totally not sponsored but this app called copilot 273 00:17:04,340 --> 00:17:08,750 Josh: money it's like a budgeting application and it's so thoughtfully curated and designed and. 274 00:17:09,510 --> 00:17:13,920 Josh: And it really deeply understands all the complexities that are related to humans 275 00:17:13,920 --> 00:17:16,420 Josh: when it comes to budgeting it understands a lot of the 276 00:17:16,770 --> 00:17:19,910 Josh: the design characteristics same with an app called flighty i'm sure a lot of 277 00:17:19,910 --> 00:17:22,930 Josh: people have heard flighty it's like a flight tracking application there's a 278 00:17:22,930 --> 00:17:27,350 Josh: thousand ways to track a flight but flighty really cares about design they really care about how, 279 00:17:28,070 --> 00:17:31,480 Josh: it's implemented with the human and they've created this amazing output and 280 00:17:31,480 --> 00:17:35,010 Josh: i don't see that changing one thing that i did want to note is that, 281 00:17:35,640 --> 00:17:39,550 Josh: i think when a lot of people see this they imagine a world in which they are 282 00:17:39,550 --> 00:17:42,690 Josh: getting replaced everyone's like ai is replacing me look how much i could do 283 00:17:42,690 --> 00:17:47,080 Josh: now it has these loops and i think the reality is it gives you a lot more agency 284 00:17:47,080 --> 00:17:48,410 Josh: to do the things you want to do, 285 00:17:48,880 --> 00:17:52,050 Josh: where maybe you're not doing the day-to-day where, 286 00:17:52,680 --> 00:17:55,580 Josh: you would normally prompt an agent to do this but you're doing a lot of the 287 00:17:55,580 --> 00:17:58,060 Josh: higher level tasks you can imagine yourself not having to do 288 00:17:58,380 --> 00:18:02,160 Josh: the day-to-day like for example if you're just managing your household you no 289 00:18:02,160 --> 00:18:05,490 Josh: longer have to take out the trash you don't have to run errands you could just 290 00:18:05,490 --> 00:18:08,520 Josh: focus on how to make your household the best household it is because you have 291 00:18:08,520 --> 00:18:09,530 Josh: that higher level ability 292 00:18:09,800 --> 00:18:13,740 Josh: and in that chart that we showed in the artifact earlier on it shows a decreasing 293 00:18:13,740 --> 00:18:16,820 Josh: sized human it's the amount of input that a human is needed to get the output you want, 294 00:18:17,280 --> 00:18:21,410 Josh: but it's still ultimately on the human being in order to to push and navigate 295 00:18:21,410 --> 00:18:25,710 Josh: towards the outputs that you want because ultimately these tools are just for us so when i think of. 296 00:18:26,500 --> 00:18:30,120 Josh: Ai becoming increasingly good and when it comes to running the show even i've 297 00:18:30,120 --> 00:18:34,210 Josh: leaned on it we both have i think a lot more recently but all that's done is 298 00:18:34,210 --> 00:18:38,360 Josh: actually given us more leverage to do more with the show than have it replace 299 00:18:38,360 --> 00:18:39,710 Josh: us and even in the case that. 300 00:18:41,060 --> 00:18:43,810 Josh: We could clone ourselves. We could create a video version of ourselves that 301 00:18:44,010 --> 00:18:48,000 Josh: has a perfect voice that sounds just like us. I don't think people actually want that. 302 00:18:48,600 --> 00:18:52,110 Josh: There's that lacking human nature that still isn't understood. 303 00:18:52,400 --> 00:18:55,910 Josh: And I find that it's more empowering when I hear that these loops exist that 304 00:18:55,910 --> 00:19:01,080 Josh: can run for days on end and create amazing outputs versus not where it's kind 305 00:19:01,080 --> 00:19:03,370 Josh: of extracted from us. I don't really think that's true. 306 00:19:03,830 --> 00:19:08,830 Ejaaz: Yeah, it's like that stat of, well, it's that thesis that everyone held about 307 00:19:08,830 --> 00:19:13,310 Ejaaz: a year ago, which is like with the increase of AI adoption, people will have 308 00:19:13,310 --> 00:19:15,930 Ejaaz: more free time to have fun and leisure. 309 00:19:16,300 --> 00:19:20,900 Ejaaz: And in fact, the opposite has shown that like people just work way more and work harder. 310 00:19:21,170 --> 00:19:24,500 Ejaaz: And the output of that work is measured across like pretty much every single 311 00:19:24,500 --> 00:19:26,210 Ejaaz: company and profession and role. 312 00:19:28,200 --> 00:19:32,590 Ejaaz: I do generally agree with that. I don't think humans are going to get wiped out anytime soon. 313 00:19:33,520 --> 00:19:41,270 Ejaaz: But one thing that is kind of nagging my brain is if we extrapolate this intelligence out enough, 314 00:19:42,080 --> 00:19:48,760 Ejaaz: there is no reason why AI won't be able to take over or replace other parts 315 00:19:48,760 --> 00:19:51,530 Ejaaz: of the cognitive process that a human can do, 316 00:19:52,410 --> 00:19:57,700 Ejaaz: particularly if it's one AI model trained on the entire corpus of knowledge 317 00:19:57,700 --> 00:19:59,440 Ejaaz: that a bunch of humans have been guiding it. 318 00:19:59,640 --> 00:20:03,980 Ejaaz: So when I think about Anthropic, when I think about OpenAI, I think about all the 319 00:20:05,300 --> 00:20:08,720 Ejaaz: millions of people that use their product every single day and the data that 320 00:20:08,720 --> 00:20:12,930 Ejaaz: they ingest every single day that gets recorded on one singular database that 321 00:20:12,930 --> 00:20:17,940 Ejaaz: can then be reused to train a better model that is more hyper-optimized towards humans. 322 00:20:18,150 --> 00:20:22,720 Ejaaz: You could argue that as a single human, you don't get to meet and read the thoughts 323 00:20:22,720 --> 00:20:24,370 Ejaaz: of every other human that is out there. 324 00:20:25,860 --> 00:20:29,550 Ejaaz: You have your very own individual process. And I think that an AM model that 325 00:20:29,550 --> 00:20:34,720 Ejaaz: can get access to the world's brain and thoughts could probably create something 326 00:20:34,720 --> 00:20:38,550 Ejaaz: kind of close to knowing what that human taste profile would be. 327 00:20:39,240 --> 00:20:44,440 Ejaaz: The other major question that I'm wondering is, how much is all of this going to cost? 328 00:20:46,210 --> 00:20:49,890 Ejaaz: One, like, stat that has stuck in my head over the recent few weeks is that 329 00:20:50,190 --> 00:20:53,010 Ejaaz: Philanthropic particularly, they service, or like, 330 00:20:53,890 --> 00:20:58,290 Ejaaz: the Fortune 10, the top 10 companies in the world, nine of them use Clawed, 331 00:20:58,560 --> 00:21:04,380 Ejaaz: and their budget's increased by 500%, or is projected to increase by 500% by the end of this year. 332 00:21:04,600 --> 00:21:07,400 Ejaaz: And they're doing this willingly because the ROI, the value that they're getting 333 00:21:07,400 --> 00:21:09,300 Ejaaz: out of that is pretty massive. 334 00:21:09,970 --> 00:21:13,490 Ejaaz: Alternatively, there are companies like Uber that have slashed their budgets 335 00:21:13,490 --> 00:21:17,210 Ejaaz: massively because their entire year's budget was spent in a couple of months. 336 00:21:17,440 --> 00:21:22,140 Ejaaz: So I'm wondering, in this world of agent loops where you've got AIs working 337 00:21:22,140 --> 00:21:25,610 Ejaaz: overnight for you, the bills are going to increase pretty massively. 338 00:21:25,610 --> 00:21:29,810 Ejaaz: And I'm wondering, unless these AI models don't get cheaper, 339 00:21:29,810 --> 00:21:33,320 Ejaaz: and there's an infrastructure bottleneck there where these GPUs cost a lot of 340 00:21:33,320 --> 00:21:36,660 Ejaaz: money, we can't scale power and infrastructure anytime soon. 341 00:21:36,660 --> 00:21:42,310 Ejaaz: We need so much more energy than we already have currently on Earth to be able to power these things. 342 00:21:42,640 --> 00:21:46,070 Ejaaz: The cost of these things are just going to go up a lot more massively, 343 00:21:46,320 --> 00:21:51,140 Ejaaz: which means that either this is only going to be a power or a tool reserved 344 00:21:51,140 --> 00:21:54,890 Ejaaz: for the rich, or something's going to break here and maybe open source models 345 00:21:54,890 --> 00:21:56,510 Ejaaz: get adopted more aggressively. 346 00:21:56,790 --> 00:22:00,700 Josh: Yeah, I imagine there's probably use cases for all of the above. 347 00:22:01,220 --> 00:22:04,260 Josh: It's like open source models will continue to improve they'll be able to do 348 00:22:04,260 --> 00:22:08,020 Josh: a lot of the more trivial tasks that don't require frontier intelligence so 349 00:22:08,020 --> 00:22:11,050 Josh: therefore the cost of those types of loops will go down because not everyone 350 00:22:11,050 --> 00:22:13,070 Josh: needs to have the most cutting edge, 351 00:22:13,800 --> 00:22:16,930 Josh: software stack engineering like they're just kind of having it help them through 352 00:22:16,930 --> 00:22:20,630 Josh: their day-to-day maybe it's replying to emails maybe it's whatever miscellaneous things it may be 353 00:22:21,010 --> 00:22:23,920 Josh: there's a high probability that these open source models as they continue to 354 00:22:23,920 --> 00:22:27,390 Josh: improve will be able to bite off a meaningful chunk of that then the other half 355 00:22:27,390 --> 00:22:30,840 Josh: is using these frontier models that is a requirement in order to get the absolute 356 00:22:30,840 --> 00:22:34,260 Josh: best results for whatever very challenging work they're doing. 357 00:22:34,560 --> 00:22:38,740 Josh: And that is going to cost a lot of money for sure. 358 00:22:39,120 --> 00:22:46,330 Josh: And I don't see that changing, but I think the output of the dollars in will continue to go up. 359 00:22:46,330 --> 00:22:50,920 Josh: It's because as you get more knowledge per token, as you get more output per 360 00:22:51,150 --> 00:22:55,030 Josh: prompt, it very clearly, I mean, the economics seem to make sense. 361 00:22:55,200 --> 00:22:56,980 Josh: And I think that's kind of right now. 362 00:22:58,960 --> 00:23:02,930 Josh: Enterprise spend on these models they're trying to figure out well how much 363 00:23:02,930 --> 00:23:06,760 Josh: value can we actually get back from every dollar spent and right now it's a 364 00:23:06,760 --> 00:23:10,210 Josh: little bit unsure you mentioned uber we have uber here that we're showing on screen 365 00:23:10,480 --> 00:23:14,680 Josh: where uber just recently put a cap on the amount of tokens that 366 00:23:15,070 --> 00:23:20,900 Josh: their employees are allowed to use at fifteen hundred dollars per engineer per tool per month and 367 00:23:21,300 --> 00:23:24,940 Josh: we'll see how that works because a lot of other companies that we know they're 368 00:23:24,940 --> 00:23:29,170 Josh: kind giving their engineers unlimited budget in fact they're kind of ranking 369 00:23:29,170 --> 00:23:33,710 Josh: the engineers based on how many tokens they're using per month and. 370 00:23:34,620 --> 00:23:38,410 Josh: We'll see where that goes. I suspect the companies that are spending more on 371 00:23:38,410 --> 00:23:41,900 Josh: tokens will continue to see a higher upside for now, at least. 372 00:23:42,180 --> 00:23:46,140 Josh: But like you mentioned, the underlying problem with all of this is we're going 373 00:23:46,140 --> 00:23:48,880 Josh: to continue to have more prompts. I mean, these loops consume a tremendous amount 374 00:23:48,880 --> 00:23:51,180 Josh: of tokens, whether they're frontier tokens or open source tokens. 375 00:23:51,410 --> 00:23:54,940 Josh: It doesn't matter. We're going to need orders of magnitude more than we have. 376 00:23:54,940 --> 00:23:59,460 Josh: And we don't have the computability. It really does always come down to that 377 00:23:59,460 --> 00:24:01,200 Josh: energy problem, that infrastructure problem. 378 00:24:01,200 --> 00:24:06,030 Josh: We don't have the infra built out to support this so therefore the costs likely 379 00:24:06,030 --> 00:24:10,870 Josh: continue to stay high maybe it's not because you're paying the provider for tokens 380 00:24:11,170 --> 00:24:15,810 Josh: perhaps it's just renting the gpu time from a cluster that is doing much more 381 00:24:15,810 --> 00:24:18,190 Josh: valuable work so i think that might ultimately be 382 00:24:18,560 --> 00:24:23,060 Josh: that crux is the actual availability of the compute to do these things and that's 383 00:24:23,060 --> 00:24:25,520 Josh: why these edge compute devices like having your, 384 00:24:26,240 --> 00:24:30,110 Josh: mac studio on your desktop that can run locally it's probably a pretty valuable thing to have. 385 00:24:30,290 --> 00:24:33,630 Ejaaz: So I'm sure a lot of you are wondering, you know, how does this apply to me? 386 00:24:34,490 --> 00:24:37,920 Ejaaz: You know, I have none of my friends have mentioned this loop feature. 387 00:24:37,920 --> 00:24:39,710 Ejaaz: I don't really know many people who are using it. 388 00:24:40,490 --> 00:24:43,850 Ejaaz: As we mentioned earlier, like this isn't probably going to be used by the bulk 389 00:24:43,850 --> 00:24:47,570 Ejaaz: or majority of people yet until some of those use cases actually arise. 390 00:24:47,570 --> 00:24:51,840 Ejaaz: I think it's mainly going to happen in the workplace. It's going to happen with 391 00:24:51,840 --> 00:24:54,540 Ejaaz: like some of these enterprise companies that are trying to automate certain 392 00:24:54,540 --> 00:24:57,610 Ejaaz: departments or functions of their particular a company like marketing, 393 00:24:57,840 --> 00:24:58,910 Ejaaz: like software engineering. 394 00:24:58,910 --> 00:25:02,740 Ejaaz: And I think it'll start with lower level tasks because these agents still aren't 395 00:25:02,740 --> 00:25:05,060 Ejaaz: smart enough to understand nuance completely. 396 00:25:05,310 --> 00:25:09,180 Ejaaz: And also, you don't just want to let an agent run loose overnight whilst you're 397 00:25:09,180 --> 00:25:12,450 Ejaaz: sleeping and then take down your entire company. And one place where it's working 398 00:25:12,640 --> 00:25:15,390 Ejaaz: tirelessly to accelerate the development of that 399 00:25:18,190 --> 00:25:23,380 Ejaaz: And we have Boris Cherny over here basically explaining how he's basically ditched 400 00:25:23,380 --> 00:25:25,470 Ejaaz: his integrated development environment. 401 00:25:25,730 --> 00:25:30,440 Ejaaz: He has ditched all of his normal tools that he had spent decades basically honing 402 00:25:30,440 --> 00:25:34,790 Ejaaz: his software engineering skill on to now completely focus on building up these 403 00:25:34,790 --> 00:25:36,530 Ejaaz: agent loops. And what is he focused on? 404 00:25:36,760 --> 00:25:40,520 Ejaaz: Well, he works primarily on cloud code, but the other folks at Anthropic and 405 00:25:40,520 --> 00:25:43,020 Ejaaz: OpenAI have started this thing called 406 00:25:43,660 --> 00:25:48,870 Ejaaz: Recursive self-improvement or RSI, which is basically the goal of getting your 407 00:25:48,870 --> 00:25:52,380 Ejaaz: AI model to build the next version of itself. 408 00:25:52,590 --> 00:25:57,390 Ejaaz: And this is a test that Anthropic and the folks at OpenAI do for any new model that they release. 409 00:25:57,660 --> 00:26:02,560 Ejaaz: They set it a goal or task to basically rebuild itself in a more improved fashion. 410 00:26:02,730 --> 00:26:07,640 Ejaaz: Now, one thing that the AI has gotten really good at is building out that next function. 411 00:26:07,640 --> 00:26:12,780 Ejaaz: But one thing it's not very good at is figuring out what research problems they 412 00:26:12,780 --> 00:26:16,400 Ejaaz: should fix, what research problems it should focus on to try and, 413 00:26:16,860 --> 00:26:21,120 Ejaaz: you know, overcome and make it ultimately, you know, a better model than its competitors. 414 00:26:21,400 --> 00:26:27,010 Ejaaz: Now, RSI is something, it's kind of like the golden egg that each AI lab is going after. 415 00:26:27,280 --> 00:26:30,270 Ejaaz: And this is the primary use of agent loops right now. 416 00:26:30,450 --> 00:26:34,440 Ejaaz: And you can see why it might be obvious. If you have an AI model that can basically 417 00:26:34,440 --> 00:26:38,210 Ejaaz: build the next best version of itself, eventually you're going to get to AGI, 418 00:26:38,210 --> 00:26:41,950 Ejaaz: whatever the hell that looks like, and then you can apply it to pretty much any sector. 419 00:26:42,180 --> 00:26:45,790 Ejaaz: Now, the problem and the worry that kind of immediately pops into my head and 420 00:26:45,790 --> 00:26:50,410 Ejaaz: a lot of these researchers head is, if it eventually does get that smart, right? 421 00:26:51,190 --> 00:26:55,850 Ejaaz: Could escape human control completely and run off on its own and do its own 422 00:26:55,850 --> 00:27:00,220 Ejaaz: thing. Because at that point, why would it need a human to kind of like guide it or shepherd it? 423 00:27:00,380 --> 00:27:03,930 Ejaaz: Instead, it can just kind of like do its own thing. So this is like the primary 424 00:27:03,930 --> 00:27:06,770 Ejaaz: use case that I'm seeing for agent loops being worked on right now. 425 00:27:06,770 --> 00:27:09,720 Ejaaz: I would love to see a like more broader application across like kind of like 426 00:27:10,110 --> 00:27:12,960 Ejaaz: consumer professions, like in finance, like in science and stuff like that, 427 00:27:12,960 --> 00:27:16,870 Ejaaz: which I do believe it'll spill over eventually. But unless you're seeing anything 428 00:27:16,870 --> 00:27:21,610 Ejaaz: else, Josh, I think like that is primarily it on agent loops and agent autonomy. 429 00:27:21,720 --> 00:27:25,790 Josh: It's on you to figure out the best use cases for it. Like there's no real company 430 00:27:25,790 --> 00:27:29,690 Josh: defining it. They're just giving you the tools. And I mean, for better or worse, 431 00:27:29,690 --> 00:27:32,890 Josh: it's very open-ended. So it's on you to figure out how best to use these. 432 00:27:33,130 --> 00:27:36,940 Josh: I think if this sounds a little overwhelming, maybe we could outline a few examples 433 00:27:36,940 --> 00:27:39,420 Josh: of each one of these kind of four rungs in the ladder here. 434 00:27:39,740 --> 00:27:42,760 Josh: The first one being prompting this everyone has done before. 435 00:27:42,760 --> 00:27:44,650 Josh: I'm sure it's like rewrite this 436 00:27:44,970 --> 00:27:49,140 Josh: email to sound more confident or explain what my doctor meant by this. 437 00:27:49,400 --> 00:27:53,760 Josh: But then you've probably also used the partial agentic usage as well of these 438 00:27:53,760 --> 00:27:58,250 Josh: models, which is like planet my three day vacation to Lisbon that I'm going on next week. 439 00:27:58,500 --> 00:28:02,010 Josh: And it will actually go off and use tools and it will think complex thoughts 440 00:28:02,010 --> 00:28:05,220 Josh: and ideas and kind of surface you a full itinerary for your trip. 441 00:28:05,530 --> 00:28:08,580 Josh: And then there's the third one, which is the harness. This is a little more 442 00:28:08,580 --> 00:28:11,430 Josh: complicated. This is for people who are building more project based stuff. 443 00:28:11,430 --> 00:28:15,800 Josh: So for example, if you want to build you a website for your dog walking business 444 00:28:15,800 --> 00:28:19,220 Josh: and you kind of describe it and you go back and forth on a spec and then it 445 00:28:19,220 --> 00:28:20,410 Josh: goes off and implements that. 446 00:28:20,630 --> 00:28:24,060 Josh: And the fourth is loops, which doesn't have to necessarily be overwhelming. 447 00:28:24,060 --> 00:28:26,810 Josh: It can be simple as let's say you are. 448 00:28:27,660 --> 00:28:30,420 Josh: Interested in the news, you could say every morning before I wake up, 449 00:28:30,600 --> 00:28:35,550 Josh: scan these 10 sources plus market data and give me this bulleted brief. 450 00:28:35,550 --> 00:28:38,910 Josh: Or let's say you have a to-do list. It'll go off and think overnight and solve 451 00:28:38,910 --> 00:28:42,370 Josh: all those problems overnight, iteratively until it comes to a solution that 452 00:28:42,370 --> 00:28:45,810 Josh: it hopefully arrives at in the mornings. There's a lot of use cases. 453 00:28:45,810 --> 00:28:47,740 Josh: I think a lot of it requires creativity. 454 00:28:47,980 --> 00:28:51,970 Josh: And that is the prompt we will leave you with today, which is share with us, 455 00:28:51,970 --> 00:28:54,220 Josh: please, how you are using these models best. 456 00:28:54,220 --> 00:28:57,340 Josh: Because so much of the question isn't are these models smart? 457 00:28:57,340 --> 00:29:00,540 Josh: It's how can I extract that intelligence from them in the most effective way 458 00:29:00,540 --> 00:29:04,330 Josh: for my life? So I would be so curious to hear which rung of the ladder you find 459 00:29:04,330 --> 00:29:05,380 Josh: yourself on one through four. 460 00:29:05,600 --> 00:29:07,990 Josh: And then what the most interesting use cases you found, 461 00:29:08,690 --> 00:29:12,650 Josh: among those rungs of the ladder are you using loops currently what are you using them for, 462 00:29:13,130 --> 00:29:16,090 Josh: are you with agents are you still using it as a google extension if you're still 463 00:29:16,090 --> 00:29:19,540 Josh: using it as a google extension i would encourage a little more creativity really 464 00:29:19,540 --> 00:29:22,580 Josh: try to ask harder questions and figure out how it could be implemented in your 465 00:29:22,580 --> 00:29:26,230 Josh: life but i think that's pretty much it on the loop um, 466 00:29:26,990 --> 00:29:31,310 Josh: you're not going anywhere but your job might shift a little bit in terms of 467 00:29:31,310 --> 00:29:34,710 Josh: scope as these tools get more powerful and that should be the hope that should 468 00:29:34,710 --> 00:29:39,540 Josh: be the goal because it'll allow you to do so much more that you want to accomplish, I believe. 469 00:29:39,540 --> 00:29:41,770 Josh: And yeah, I think that's where we'll leave you with today. 470 00:29:42,040 --> 00:29:46,580 Ejaaz: Thank you folks so much for listening. Similar to Josh's prompts, 471 00:29:46,580 --> 00:29:51,130 Ejaaz: I'm actually kind of curious, for one singular task that you've used your AI 472 00:29:51,130 --> 00:29:53,690 Ejaaz: for, what is the most number of tokens that you've burnt? 473 00:29:53,810 --> 00:29:56,860 Ejaaz: Be honest, it can be for any use case, doesn't matter, let us know. 474 00:29:57,160 --> 00:30:01,390 Ejaaz: And also, what is the longest that you've had an AI work on a particular task 475 00:30:01,390 --> 00:30:05,880 Ejaaz: for? Is it a couple of minutes? Is it hours? Is it potentially overnight? 476 00:30:05,880 --> 00:30:08,750 Ejaaz: Let us know. I'm curious. And what was the associated bill with that? 477 00:30:09,030 --> 00:30:12,320 Ejaaz: And yeah, we'll see you on the next episode. Wherever you listen to us, 478 00:30:12,520 --> 00:30:16,000 Ejaaz: if you haven't subscribed, if you haven't rated us, if you're not leaving us 479 00:30:16,000 --> 00:30:17,180 Ejaaz: comments, what are you doing? 480 00:30:17,410 --> 00:30:20,910 Ejaaz: We respond to pretty much any and every one of them. We listen to your feedback. 481 00:30:20,910 --> 00:30:23,540 Ejaaz: It feeds into some of the work and content that we put out. 482 00:30:23,770 --> 00:30:28,360 Ejaaz: We are almost hitting 60,000 of you folks. And you guys are reading our newsletter, 483 00:30:28,360 --> 00:30:33,390 Ejaaz: which is like hit out to about 100,000 plus people. every single week. We post twice a week. 484 00:30:34,030 --> 00:30:37,840 Ejaaz: But yeah, wherever you are, please subscribe to us, leave us a comment, 485 00:30:37,840 --> 00:30:38,900 Ejaaz: and we'll see you on the next one. 486 00:30:39,290 --> 00:30:40,080 Josh: See you guys next time. 487 00:30:40,080 --> 00:30:40,450 Ejaaz: Peace.