WEBVTT NOTE Sentence-level transcript of https://www.youtube.com/watch?v=33Oct2hqGnk NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/33Oct2hqGnk.html. A cue ends where the next begins, or 2 s after its last word. s1 00:00:01.309 --> 00:00:03.309 [music] s2 00:00:12.719 --> 00:00:14.320 Hi everyone. s3 00:00:14.320 --> 00:00:16.560 Um, welcome, welcome. s4 00:00:16.560 --> 00:00:20.400 Agents that forge their own tools. s5 00:00:20.400 --> 00:00:22.320 What is so great about this? s6 00:00:22.320 --> 00:00:24.000 Can't your claw code do this? s7 00:00:24.000 --> 00:00:29.359 Or can't your cursor or your code editors, can't they all write their own tools? s8 00:00:29.359 --> 00:00:31.359 So what am I even talking about here? s9 00:00:31.359 --> 00:00:32.800 Why is this such a big deal? s10 00:00:32.800 --> 00:00:33.120 Right? s11 00:00:33.120 --> 00:00:38.320 Like any all of your agents can write tools by themselves. s12 00:00:38.320 --> 00:00:45.120 But how many agents do you know that can fix code at runtime? s13 00:00:45.120 --> 00:00:47.360 Let's say you've got something in production. s14 00:00:47.360 --> 00:00:48.079 It's running. s15 00:00:48.079 --> 00:00:50.640 Everything looks great, but something is failing. s16 00:00:50.640 --> 00:00:51.840 Then what happens? s17 00:00:51.840 --> 00:01:01.440 you have to probably bring it down, fix it by yourself and then restart it and then work with your clot code or whichever editor you're using and then bring it back up, right? s18 00:01:01.440 --> 00:01:07.119 But imagine if your agent was smart enough to figure out, oh man, there is a bug here. s19 00:01:07.119 --> 00:01:08.560 Something's not working. s20 00:01:08.560 --> 00:01:10.880 I'm falling falling into an error. s21 00:01:10.880 --> 00:01:12.560 I need to fix myself. s22 00:01:12.560 --> 00:01:16.080 And it realizes what it can do to fix itself. s23 00:01:16.080 --> 00:01:20.799 Writes its own tools all writes its own agents and fixes itself. s24 00:01:20.799 --> 00:01:22.799 How cool would that be, right? s25 00:01:22.799 --> 00:01:26.159 And that's what I'm here to talk to you about today. s26 00:01:26.159 --> 00:01:33.600 Let me quickly show you a demo because I think I like leading in with demos so you know what you're getting into. s27 00:01:33.600 --> 00:01:34.799 So, let's start with that. s28 00:01:34.799 --> 00:01:35.360 Right. s29 00:01:35.360 --> 00:01:38.320 So, here I have a and you can see my screen. s30 00:01:38.320 --> 00:01:40.560 I'm hoping it's big enough I can zoom in. s31 00:01:40.560 --> 00:01:40.880 Okay. s32 00:01:40.880 --> 00:01:45.119 So, I have a simple agent here called agents.py. s33 00:01:45.439 --> 00:01:49.600 This agent, we can go over the code in a bit, but this agent has nothing. s34 00:01:49.600 --> 00:02:02.159 It's just got it's just got a pretty cool system prompt and then it's calling the system prompt and it's got three tools and if you look at the tools package there are actually no tools files that are written. s35 00:02:02.159 --> 00:02:07.040 So technically this agent has no capability by itself, right? s36 00:02:07.040 --> 00:02:09.119 It shouldn't be able to do anything, right? s37 00:02:09.119 --> 00:02:13.920 All it should technically hallucinate because of the LLM that is running under the hood. s38 00:02:13.920 --> 00:02:18.480 But let's say I say something like okay so first I'm going to start running this. s39 00:02:18.480 --> 00:02:20.640 So we know it's running. s40 00:02:20.640 --> 00:02:20.959 Okay. s41 00:02:20.959 --> 00:02:24.239 So in runtime right now this is running. s42 00:02:24.239 --> 00:02:43.920 Now I'm going to say hey you know what um help me calculate something uh some complex math equation right when I say this this agent which technically has nothing available s43 00:02:43.920 --> 00:02:56.319 is going to be calling the three tools that it's got and we'll look into the tools that it's got in just a bit and it's going to say oh you know what let me create a math calculator tool because that's what you're asking for. s44 00:02:56.319 --> 00:03:03.519 And it's not just going to create, it's going to let me use this tool that I'm creating right now at runtime without having to rerun it. s45 00:03:03.519 --> 00:03:04.000 So, okay. s46 00:03:04.000 --> 00:03:06.080 So, it's clearly gone ahead and created. s47 00:03:06.080 --> 00:03:07.920 So, let's look at tools again. s48 00:03:07.920 --> 00:03:13.120 It's created the math calculator tool and it's saying, okay, I do all these things. s49 00:03:13.120 --> 00:03:14.959 I've got min, max, blah, blah, blah. s50 00:03:14.959 --> 00:03:17.360 It's gone and created this entire new tool. s51 00:03:17.360 --> 00:03:19.200 So, it's saying I can do all these things, right? s52 00:03:19.200 --> 00:03:33.360 So, let's say calculate uh oh yikes uh what's the square of blah blah blah some number right I'm just typing out a random number so now it's calling that math calculator s53 00:03:33.360 --> 00:03:42.959 tool that it just wrote by itself and it's giving me the answer not just that right so let's say I give it something completely out of blue and I say uh cool s54 00:03:42.959 --> 00:03:51.200 whatever count number of characters in my word. s55 00:03:51.200 --> 00:03:51.680 Okay. s56 00:03:51.680 --> 00:03:58.080 So then, so then it's like, okay, so it's going to check if it can use any of the tools that it's currently got access to. s57 00:03:58.080 --> 00:03:58.799 It doesn't. s58 00:03:58.799 --> 00:04:08.080 So what it's saying is, okay, let me write a tool to count the number of characters you want because clearly it knows that it does not have the capability to do it. s59 00:04:08.080 --> 00:04:10.799 And so it's going ahead and it's creating a there we go. s60 00:04:10.799 --> 00:04:15.439 Even before I could finish speaking, it created a character counter for us. s61 00:04:15.439 --> 00:04:21.680 So, let's say um do it for this word. s62 00:04:21.680 --> 00:04:22.720 And it's not even a word. s63 00:04:22.720 --> 00:04:24.880 I'm just going to put random things. s64 00:04:24.880 --> 00:04:25.280 You know what? s65 00:04:25.280 --> 00:04:28.479 I should have actually done something that I could count so I can validate it. s66 00:04:28.479 --> 00:04:30.080 But the thing is it's calling a tool. s67 00:04:30.080 --> 00:04:35.280 It's calling the character counter tool and it's telling me that it's got 28 letters. s68 00:04:35.280 --> 00:04:36.720 Let's do it something we can test. s69 00:04:36.720 --> 00:04:39.280 So, I'm just going to do it for these three letters, right? s70 00:04:39.280 --> 00:04:40.400 We'll see if it knows. s71 00:04:40.400 --> 00:04:41.919 So, I'm not even telling it what to do. s72 00:04:41.919 --> 00:04:43.360 I'm just typing three letters. s73 00:04:43.360 --> 00:04:45.680 It knows that it has to do it and it's doing it. s74 00:04:45.680 --> 00:04:47.440 How powerful is this? s75 00:04:47.440 --> 00:04:54.880 So, what I want to leave you with today is show you how you can build something like this where you can use it. s76 00:04:54.880 --> 00:05:00.320 I think it's pretty powerful and how we can implement it in our work day-to-day work. s77 00:05:00.320 --> 00:05:01.120 Very niche. s78 00:05:01.120 --> 00:05:03.520 So, now let's get down to the code. s79 00:05:03.520 --> 00:05:07.199 This is an idea called meta tooling. s80 00:05:07.199 --> 00:05:12.320 and I'm using an open-source framework called Strand Agents. s81 00:05:12.320 --> 00:05:15.840 And my name is Sandy and I work at AWS. s82 00:05:15.840 --> 00:05:26.560 And we at AWS built Strand agents as an open-source a aentic harness that you can use to run your LLMs with, right? s83 00:05:26.560 --> 00:05:31.280 And so, um, how show of hands, how many developers do we have in the room? s84 00:05:31.280 --> 00:05:31.840 A quick show. s85 00:05:31.840 --> 00:05:32.639 Oh, most of you. s86 00:05:32.639 --> 00:05:33.039 Brilliant. s87 00:05:33.039 --> 00:05:34.400 I'm talking to the right audience. s88 00:05:34.400 --> 00:05:36.880 If not, I will change things around a little bit. s89 00:05:36.880 --> 00:05:40.880 So, if you guys like getting into the code, I'm going to get into the code with me. s90 00:05:40.880 --> 00:05:41.600 Right. s91 00:05:41.600 --> 00:05:50.800 So, the main thing we need here for agents and I have a slide deck that goes with it is we need just three types of tools. s92 00:05:50.800 --> 00:05:59.199 The editor, the shell, and the load tool along with the system prompt that lets us implement this function. s93 00:05:59.199 --> 00:06:04.479 So I'm going to quickly go to my slide deck that helps us understand this better. s94 00:06:04.479 --> 00:06:07.120 Like I mentioned, this is strand agent. s95 00:06:07.120 --> 00:06:13.120 This is our agentic AI harness and you can use your own LLMs, bring your own models. s96 00:06:13.120 --> 00:06:24.639 Another cool part because this is uh an harness is the fact that if let's say tomorrow there's a new the state-of-the-art LLM that comes out, you don't have to rewrite all of your system prompt. s97 00:06:24.639 --> 00:06:26.160 You don't have to rewrite the structure. s98 00:06:26.160 --> 00:06:28.560 You don't have to change your entire architecture. s99 00:06:28.560 --> 00:06:33.600 you can keep it as it is and whichever model you use you can just replace it. s100 00:06:33.600 --> 00:06:39.440 So it also lets you experiment with model performance without really rewriting your entire architecture. s101 00:06:39.440 --> 00:06:40.960 So it's super super powerful. s102 00:06:40.960 --> 00:06:44.080 Supports MCP AA all of that cool stuff. s103 00:06:44.080 --> 00:06:46.240 Has a bunch of model providers. s104 00:06:46.240 --> 00:06:47.440 Um you can get started. s105 00:06:47.440 --> 00:06:48.800 So what do we do? s106 00:06:48.800 --> 00:06:49.680 Very simple right? s107 00:06:49.680 --> 00:06:51.120 Five lines of code. s108 00:06:51.120 --> 00:06:54.479 Uh import agent from there import shell. s109 00:06:54.479 --> 00:06:59.599 Shell is the very first tool that we need because we need to give it access to the directory. s110 00:06:59.599 --> 00:07:00.800 We need to give it access. s111 00:07:00.800 --> 00:07:04.960 It needs to know what it's running, where it's running, where your codebase is located. s112 00:07:04.960 --> 00:07:08.000 So that is just one of the tools that you need. s113 00:07:08.000 --> 00:07:12.800 And then if you ask it what can you do, it understands what's going on under the hood. s114 00:07:12.800 --> 00:07:17.280 So to techniques technically start being able to use it on the fly, right? s115 00:07:17.280 --> 00:07:23.680 So then let's get into what I actually just built a couple of minutes ago and I showed you over here. s116 00:07:23.680 --> 00:07:25.280 What am I trying to do? s117 00:07:25.280 --> 00:07:30.880 The idea because this is communitydriven, AWS manages it and maintains it for you. s118 00:07:30.880 --> 00:07:36.639 People write a lot of pre-built inbit tools, but there's also community written tools. s119 00:07:36.639 --> 00:07:42.560 And so all you need is three tools to be able to start implementing matter tooling. s120 00:07:42.560 --> 00:07:43.759 So what do you need? s121 00:07:43.759 --> 00:07:51.280 First of all, you need to have a system prompt that tells your agent what a good tool looks like. s122 00:07:51.280 --> 00:07:52.479 How does that work? s123 00:07:52.479 --> 00:07:56.080 We have to start with the at tool decorator function. s124 00:07:56.080 --> 00:07:57.919 That's how it knows it's a tool. s125 00:07:57.919 --> 00:07:58.560 Excuse me. s126 00:07:58.560 --> 00:08:00.240 That's how it knows it's a tool. s127 00:08:00.240 --> 00:08:04.720 And then you tell it what a tool looks like and where it should be stored, right? s128 00:08:04.720 --> 00:08:06.639 And so you call the system prompt there. s129 00:08:06.639 --> 00:08:07.919 That's all you say. s130 00:08:07.919 --> 00:08:11.680 And then you can give it access to different tools. s131 00:08:11.680 --> 00:08:24.400 I highlighted load tools from directory equals true because that is what enables you to enables the agent to start letting it run the tools that it's generated at runtime s132 00:08:24.400 --> 00:08:25.520 dynamically. s133 00:08:25.520 --> 00:08:31.039 And we later on decided we should just make this into a fun into a tool by itself. s134 00:08:31.039 --> 00:08:33.360 So now that's become the load tool. s135 00:08:33.360 --> 00:08:40.000 So now when you go and say hey create two create five new tools for yourself and start using them it's going to know what it's going to do. s136 00:08:40.000 --> 00:08:43.440 So let's go back to the code that I showed you earlier, right? s137 00:08:43.440 --> 00:08:44.159 Same thing. s138 00:08:44.159 --> 00:08:51.519 It's got the system prompt and here I'm saying you are the meta tooling agent that creates your own tools and custom tools at runtime. s139 00:08:51.519 --> 00:08:55.839 So I'm defining the tool template and telling it what a tool looks like. s140 00:08:55.839 --> 00:08:59.440 I'm giving it the tool specification of what it's supposed to look like. s141 00:08:59.440 --> 00:09:04.640 And then I'm giving it rules for what it's supposed to do and where it's supposed to write into. s142 00:09:04.640 --> 00:09:09.120 And I'm also telling it always check if it if it exists or not. s143 00:09:09.120 --> 00:09:24.320 only if it doesn't exist then write a new tool and I'm telling it so I'm giving it three tools editor which allows it to write the shell tool which shows it what it has and the load tool which allows it to load from the directory s144 00:09:24.320 --> 00:09:40.800 the tool from the directory dynamically and this is all you need and it takes just one line of code right system prompt equals system prompt three different tools and this is just code so that I can chat with it continuously. s145 00:09:40.800 --> 00:09:50.000 And so when I do this, it's actually able to start running create five random tools for yourself. s146 00:09:50.000 --> 00:09:53.839 And so it's going to start creating these five new tools. s147 00:09:53.839 --> 00:10:08.240 And the power of this is like I said, when you're deploying code into production, you need you might want a way to for the agent to be smart enough to figure out what's wrong and to be able to answer questions for itself. s148 00:10:08.240 --> 00:10:18.560 Let's say you're you're building this app that's supposed to book flight tickets domestically, but let's say you have one user who's saying, "No, no, I want a flight ticket from, I don't know, India to Hong Kong." s149 00:10:18.560 --> 00:10:27.040 And your agent gives up because it's not been explicitly programmed to do that, or it's not been trained to have access to be able to do that. s150 00:10:27.040 --> 00:10:28.480 But do you just want it to fail? s151 00:10:28.480 --> 00:10:29.839 Do you want it to run into an error? s152 00:10:29.839 --> 00:10:36.160 Do you want to give it back to your deployment to your engineering team and have a turnaround of two weeks and then come back? s153 00:10:36.160 --> 00:10:37.839 No. You want to be able to solve it. s154 00:10:37.839 --> 00:10:42.079 So this is one of the ways by which you can handle that right. s155 00:10:42.079 --> 00:10:45.600 So again let's get back to the code and we have more cool things coming up. s156 00:10:45.600 --> 00:10:51.279 Now I was speaking about tools that can this agent that can write it own write its own tools. s157 00:10:51.279 --> 00:10:53.200 Can it also write its own agents? s158 00:10:53.200 --> 00:10:54.000 Yes it can. s159 00:10:54.000 --> 00:10:57.519 And I have another demo for that but wait until the last few minutes for that. s160 00:10:57.519 --> 00:10:58.160 Right. s161 00:10:58.160 --> 00:11:05.519 So when we talk about agents that write their own u agents also we want to understand agentic patterns. s162 00:11:05.519 --> 00:11:07.440 We have three primary patterns. s163 00:11:07.440 --> 00:11:13.519 The first one is swarm where you have sub agents that work with each other in parallel. s164 00:11:13.519 --> 00:11:18.880 So they work on a task in a combined fashion and they take different parts of the task themselves. s165 00:11:18.880 --> 00:11:28.880 Then you have a graph where the result of one sub aent gets fed into the other sub aent and then there's a handoff and then you could also have workflows s166 00:11:28.880 --> 00:11:31.040 that have a combination of all of these. s167 00:11:31.040 --> 00:11:31.600 Right? s168 00:11:31.600 --> 00:11:35.839 So with the workflow, you could have a parallel set of sub aents calling each other. s169 00:11:35.839 --> 00:11:39.440 And then there could be nodes and graphs and all of that cool stuff. s170 00:11:39.440 --> 00:11:46.079 And so this and so they all of this can be created by just the agent itself. s171 00:11:46.079 --> 00:11:47.040 We don't need to. s172 00:11:47.040 --> 00:11:51.360 We can do it our own selves, but we don't necessarily need to. s173 00:11:51.360 --> 00:11:56.240 So um let's quickly go check how that looks. s174 00:11:56.240 --> 00:11:58.399 Um so this is the exact same thing. s175 00:11:58.399 --> 00:12:00.640 This is for my meta agents. s176 00:12:00.640 --> 00:12:01.839 I'm doing the exact same thing. s177 00:12:01.839 --> 00:12:03.839 I've given it three tools over here. s178 00:12:03.839 --> 00:12:09.839 I'm actually if you look at my agents file, it's pretty empty and I'm saying agent as the tool file template. s179 00:12:09.839 --> 00:12:13.120 So I'm saying just break it into two to four sub agents. s180 00:12:13.120 --> 00:12:14.079 I'm keeping it simple. s181 00:12:14.079 --> 00:12:15.920 No swarm, nothing fancy here. s182 00:12:15.920 --> 00:12:18.399 Simple demo and use it. s183 00:12:18.399 --> 00:12:21.519 And this is the tool spec for what an agent should look like. s184 00:12:21.519 --> 00:12:24.399 And I'm happy to share the code with you after this, right? s185 00:12:24.399 --> 00:12:26.240 And these are the rules that it follows. s186 00:12:26.240 --> 00:12:27.120 Same thing, right? s187 00:12:27.120 --> 00:12:28.800 One line is all it takes. s188 00:12:28.800 --> 00:12:37.920 system prompt equals system prompt three tools and then say so I'm going to start running this now python 3 agent.py pie. s189 00:12:37.920 --> 00:12:48.320 And so I'm flying to Hawaii or whichever place to some uh to Hawaii. s190 00:12:48.320 --> 00:12:49.600 Let's just say Hawaii. s191 00:12:49.600 --> 00:12:54.160 Help me plan my itinerary. s192 00:12:54.160 --> 00:12:55.120 Okay. s193 00:12:55.120 --> 00:12:56.560 So this is all I'm telling it. s194 00:12:56.560 --> 00:13:00.320 I'm not even telling it, hey go make your own agent or go create your own rule. s195 00:13:00.320 --> 00:13:01.519 This is all I'm telling it. s196 00:13:01.519 --> 00:13:04.000 This is how your customers would chat with it, right? s197 00:13:04.000 --> 00:13:07.200 they just tell them what their problem is and they expect you to figure it out. s198 00:13:07.200 --> 00:13:07.680 Right? s199 00:13:07.680 --> 00:13:10.480 So, this agent is going to be like, "Oh, great. s200 00:13:10.480 --> 00:13:16.000 Um, here I'm going to be printing out I'm going to be creating these into three focus sub aents. s201 00:13:16.000 --> 00:13:24.480 The flight agent, the activities agent, the itinerary agent, and it's calling the editor tool three times, one for each of these." s202 00:13:24.480 --> 00:13:27.200 So, I see that it's already created the activities. s203 00:13:27.200 --> 00:13:29.040 Yeah, it's creating It's pretty fast. s204 00:13:29.040 --> 00:13:31.680 It's creating the activities agent. s205 00:13:31.680 --> 00:13:34.160 It's created like a flight agent that's doing something. s206 00:13:34.160 --> 00:13:35.680 It's created an itinary agent. s207 00:13:35.680 --> 00:13:45.120 The let me call out that the one thing this can't currently do right now is access real-time information because I've been given it a tool that has access to real world API. s208 00:13:45.120 --> 00:13:45.600 Right? s209 00:13:45.600 --> 00:13:55.680 So assuming you add that on you can ask it what's the weather like and it'll be like you know what let me write a tool that can always grab the latest weather in whichever location and recommend stuff. s210 00:13:55.680 --> 00:13:59.760 So based off of that, it created those agents. s211 00:13:59.760 --> 00:14:04.240 And if you look at it somewhere here, it should technically be calling those agents. s212 00:14:04.240 --> 00:14:04.720 There we go. s213 00:14:04.720 --> 00:14:07.120 See, it's calling the activities agents. s214 00:14:07.120 --> 00:14:09.680 And then it's saying these are the ma main airports. s215 00:14:09.680 --> 00:14:11.279 These are the best beaches. s216 00:14:11.279 --> 00:14:14.560 These are what it is, and this is what you have to do. s217 00:14:14.560 --> 00:14:20.079 And so how amazing is is it that you can now have your agent. s218 00:14:20.079 --> 00:14:25.519 So you now your focus is not just building a good agentic system that can get the job done. s219 00:14:25.519 --> 00:14:36.160 You would also maybe want to start thinking about healing, writing self-healing code, about creating an agent that can sort of go over itself, look at what bugs are there and fix itself. s220 00:14:36.160 --> 00:14:39.279 Had this been a longer session, I usually do longer sessions. s221 00:14:39.279 --> 00:14:44.240 I usually ask the audience to ask it to do crazy stuff and I try breaking it. s222 00:14:44.240 --> 00:14:46.560 And what happens is this actually ends up breaking. s223 00:14:46.560 --> 00:14:50.240 you'll see a bunch of errors coming up and this thing is just talking chatting away. s224 00:14:50.240 --> 00:15:04.800 So I need to probably set the max token somewhere here but um and so it actually ends up erroring and breaking but it recognizes that it's erroring and it says oh man I'm ering let me fix myself and it would go and fix the same s225 00:15:04.800 --> 00:15:08.320 fix this exact same tool or agent that it's got here. s226 00:15:08.320 --> 00:15:15.120 So now if I'm like I don't want Hawaii why do you make it for Hawaii do for something else going to be like sure let me do it let me update the agent. s227 00:15:15.120 --> 00:15:17.120 So it's super super powerful. s228 00:15:17.120 --> 00:15:22.000 But then again we have to keep in mind that this is just autonomously doing things by itself. s229 00:15:22.000 --> 00:15:27.440 With great power also comes great responsibility even greater responsibility. s230 00:15:27.440 --> 00:15:28.959 So what do we need to do with it? s231 00:15:28.959 --> 00:15:29.199 Right? s232 00:15:29.199 --> 00:15:31.519 We need a eval. s233 00:15:31.519 --> 00:15:34.720 What are some inbuilt evals that str agent has? s234 00:15:34.720 --> 00:15:37.279 We we have totally eight different evals. s235 00:15:37.279 --> 00:15:39.920 The first one is at an overall session level. s236 00:15:39.920 --> 00:15:44.000 We want to figure out if I told it hey book my flight tickets is it actually booking? s237 00:15:44.000 --> 00:15:45.680 Did it achieve the end goal? s238 00:15:45.680 --> 00:15:47.360 That is part one of it, right? s239 00:15:47.360 --> 00:15:48.639 End goal achievement. s240 00:15:48.639 --> 00:15:51.759 Second thing we want to look at it is at a trace level. s241 00:15:51.759 --> 00:15:53.360 Um was it actually helpful? s242 00:15:53.360 --> 00:15:59.199 If I say what is my balance and says my balance is 500 bucks, is that the expected type of answer? s243 00:15:59.199 --> 00:16:00.000 Part one. s244 00:16:00.000 --> 00:16:02.000 Part two, did it make up the answer? s245 00:16:02.000 --> 00:16:04.560 Is that the actual answer that you're supposed to get? s246 00:16:04.560 --> 00:16:06.560 You need to be able to evaluate that. s247 00:16:06.560 --> 00:16:08.079 But it doesn't just end there. s248 00:16:08.079 --> 00:16:09.440 What about tool access? s249 00:16:09.440 --> 00:16:10.000 Right? s250 00:16:10.000 --> 00:16:14.959 What about did it use the right tool to get to that answer or did it use the wrong tool? s251 00:16:14.959 --> 00:16:18.959 And within that specific tool, did it use the right parameter? s252 00:16:18.959 --> 00:16:23.519 Is the account ID actually one two three or is it supposed to be is it messing up? s253 00:16:23.519 --> 00:16:30.079 We want insight into what is going on so that we can trust it and reliably give it access. s254 00:16:30.079 --> 00:16:32.000 Again, not just it doesn't just end there. s255 00:16:32.000 --> 00:16:35.519 We also want to understand inter agent dependency. s256 00:16:35.519 --> 00:16:44.639 When we're building multi-agentic systems, we want to look at the overall outcome and the overall response if it's the way you wanted it to be. s257 00:16:44.639 --> 00:16:49.920 You want to make sure that it's calling the different tools and the different sub aents in the right sequence. s258 00:16:49.920 --> 00:16:58.560 And you also want to take a deeper dive into what the messages, the inter agent communication looks like and if that's up to the mark. s259 00:16:58.560 --> 00:17:04.480 All of this together gives us the full end to end agent conversation within our session. s260 00:17:04.480 --> 00:17:06.240 This gives us the insight. s261 00:17:06.240 --> 00:17:08.959 Again, this also isn't good enough, right? s262 00:17:08.959 --> 00:17:12.000 We want to make sure it doesn't cause undue damage. s263 00:17:12.000 --> 00:17:18.559 If this meta tooling or this meta agent can spin off tools and agents, it can also modify and delete. s264 00:17:18.559 --> 00:17:20.480 Do I want it to delete my data? s265 00:17:20.480 --> 00:17:24.400 Do I want it to delete stuff that I have put all of my effort into building? s266 00:17:24.400 --> 00:17:26.400 No. So, what do I do? s267 00:17:26.400 --> 00:17:28.559 I have to have guard rails in place. s268 00:17:28.559 --> 00:17:30.240 Four different types of guard rails. s269 00:17:30.240 --> 00:17:32.720 The first one is your environment. s270 00:17:32.720 --> 00:17:35.760 We need to make sure that we're sandboxing the environment. s271 00:17:35.760 --> 00:17:45.440 And when I say sandboxing the environment, it's not your typical sandbox where your agent is in one container and you're protecting it from the rest of the world. s272 00:17:45.440 --> 00:17:57.919 We want the environment in which this code is executing to be sandboxed and Strands recently launched a feature about I think a few weeks ago where we can sandbox s273 00:17:57.919 --> 00:18:12.000 the code execution environment and that is what guarantees that it doesn't touch the rest of your code execution environments and it doesn't mess up everything else and it doesn't write into wrong places and delete from the wrong places. s274 00:18:12.000 --> 00:18:12.480 Right. s275 00:18:12.480 --> 00:18:14.400 Next we have to constrain the tools. s276 00:18:14.400 --> 00:18:16.240 We have to constrain permissions. s277 00:18:16.240 --> 00:18:24.160 We have to constrain the who's got access and who can ask it questions and what types of questions we need to ask and observability. s278 00:18:24.160 --> 00:18:37.440 All of your evals, your hotel, your telemetry, all of that data needs to be there in order to be able to say that, hey, reliably and say, hey, I think I can go ahead s279 00:18:37.440 --> 00:18:52.960 and build an agentic system that can think for itself that it can do more than what it's just been set out to do because it can fix things when things are broken on the fly, on the run, s280 00:18:52.960 --> 00:18:57.440 and it can solve more capabilities than it has been initially taught to. s281 00:18:57.440 --> 00:19:03.919 This is the starting of self-improving agents and the self- evvolving agents. s282 00:19:03.919 --> 00:19:08.480 At some point, I'm convinced they will take over the world and I'm I'm going to be at risk. s283 00:19:08.480 --> 00:19:11.840 But until then, I think this is incredible. s284 00:19:11.840 --> 00:19:22.480 And if anything, if I want to leave you with anything, I would want you to I would want to leave you with a quick idea of where Strands is, where Strand started with and where we are at now. s285 00:19:22.480 --> 00:19:25.360 We started with building their own tools. s286 00:19:25.360 --> 00:19:30.160 We we moved to getting them to update uh to create agents by themselves. s287 00:19:30.160 --> 00:19:33.280 Now it's at a point where they can update their own source code. s288 00:19:33.280 --> 00:19:37.039 Fun fact, we created Strand agents three years ago internally. s289 00:19:37.039 --> 00:19:40.640 It was supposed to be this internal thing before agents was even a thing. s290 00:19:40.640 --> 00:19:45.600 And it was so powerful that we said, you know what, let's just open source it and give it back to the community. s291 00:19:45.600 --> 00:19:49.039 And we wrote the Python version of Strands. s292 00:19:49.039 --> 00:19:54.480 And the Python version of strands wrote by itself the TypeScript version of Stand. s293 00:19:54.480 --> 00:19:56.720 So, it did update its own source code. s294 00:19:56.720 --> 00:19:57.679 It's pretty powerful. s295 00:19:57.679 --> 00:19:59.039 I'm pretty psyched. s296 00:19:59.039 --> 00:20:06.320 And if if this is my call to action, if uh this sounds interesting to you, reach out to me. s297 00:20:06.320 --> 00:20:09.120 Scan these and we can have a conversation later. s298 00:20:09.120 --> 00:20:12.799 I'm happy to share my slide deck and my code with you. s299 00:20:12.799 --> 00:20:14.960 Um and we can chat offline. s300 00:20:14.960 --> 00:20:17.039 Um thank you very much. s301 00:20:17.039 --> 00:20:18.320 This is my LinkedIn. s302 00:20:18.320 --> 00:20:22.640 Uh, we have 30 seconds to go, I think.