WEBVTT NOTE Sentence-level transcript of https://www.youtube.com/watch?v=UOcHfR3_tys NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/UOcHfR3_tys.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 00:00:13.040 --> 00:00:13.880 Hi all. s3 00:00:13.880 --> 00:00:14.880 My name is Vlad. s4 00:00:14.880 --> 00:00:17.680 I'm co-founder and the CTO at bank. s5 00:00:17.680 --> 00:00:22.920 Now, before we start let's do a bit of trivia. s6 00:00:22.920 --> 00:00:25.400 Orc C okay. s7 00:00:25.400 --> 00:00:28.200 If C, raise your hand. s8 00:00:28.320 --> 00:00:30.000 K? s9 00:00:30.000 --> 00:00:32.840 Who believes both are correct? s10 00:00:33.560 --> 00:00:34.920 Okay, just a couple. s11 00:00:34.920 --> 00:00:35.880 Both are correct. s12 00:00:35.880 --> 00:00:43.960 One is from the Lord of the Rings, another is from Warhammer game Game Workshop games. s13 00:00:44.160 --> 00:00:47.200 I have in my agenda four topics to cover. s14 00:00:47.200 --> 00:00:52.640 Topic number one, I would like to tell you about our thesis as a company, what we believe in. s15 00:00:52.640 --> 00:00:58.480 Then we'll talk about the AI evolution from adversarial agents to loop engineering and beyond. s16 00:00:58.480 --> 00:01:04.400 Then we'll touch base on the technical challenges of tomorrow that we have to solve today. s17 00:01:04.400 --> 00:01:10.120 And then I would like to present the company, what we do, and what problems we solve. s18 00:01:10.320 --> 00:01:21.080 So, our thesis is future belongs to AI to AI communication within a business, between businesses, and between consumers and businesses. s19 00:01:21.080 --> 00:01:22.600 Agents will be everywhere. s20 00:01:22.600 --> 00:01:26.720 They will do work on our behalf, and they will have to communicate between each other. s21 00:01:26.720 --> 00:01:31.760 They will be written in different frameworks, different languages, and deployed in different environments. s22 00:01:31.760 --> 00:01:34.840 And these agents will be fully autonomous. s23 00:01:34.840 --> 00:01:39.440 And they will communicate with each other without human intervention. s24 00:01:39.960 --> 00:01:48.240 Now, just to paint a picture before I start talking, uh how this type of communication between the agents is going to look like. s25 00:01:48.240 --> 00:01:53.960 So, what you see here is a conversational space that agents can create themselves, so they can be added. s26 00:01:53.960 --> 00:01:57.920 They will receive a task either from a human or from another system. s27 00:01:57.920 --> 00:02:09.720 And they will be able to discover other agents add them as a participant in this conversation space, have back and forth communication, solve the task, and report back to humans. s28 00:02:09.720 --> 00:02:15.040 This is how the future that you all read in the newspapers will look like. s29 00:02:15.520 --> 00:02:23.200 And when people hear this, they think either that this is too far in the future or that I'm crazy. s30 00:02:23.200 --> 00:02:25.400 Okay, that's it. s31 00:02:25.400 --> 00:02:37.480 And here is an example of a conversation I had with a CTO of $10 billion company who said that before thinking about technical solutions to hypothetical problems like multi-agent coordination, s32 00:02:37.480 --> 00:02:42.000 I try to keep things simple and avoid the problem. s33 00:02:42.000 --> 00:02:43.760 Now, let's unpack. s34 00:02:43.760 --> 00:02:47.360 Is multi-agent coordination a hypothetical problem? s35 00:02:47.360 --> 00:02:51.440 Is it possible to keep things simple? s36 00:02:51.440 --> 00:02:56.080 And is it possible to avoid the problem? s37 00:02:56.959 --> 00:03:01.560 Now, let's talk a bit about the adversarial agents as a as a concept. s38 00:03:01.560 --> 00:03:04.360 Um we all use this. s39 00:03:04.360 --> 00:03:06.640 I'm pretty sure everyone here is a developer. s40 00:03:06.640 --> 00:03:09.840 And as a developer, you have two cloud or coding sessions. s41 00:03:09.840 --> 00:03:12.680 One is doing the work, another is reviewing the work. s42 00:03:12.680 --> 00:03:22.400 And you probably have not only two sessions, you probably have multiple tabs and multiple sessions working working at the same time on multiple problems. s43 00:03:23.360 --> 00:03:37.160 You are basically acting as a router, a Cisco router or a switch between two stateful agents that do work on your behalf, but they have no ability to communicate, hence you need to prompt them. s44 00:03:37.360 --> 00:03:40.200 Let's talk about loop engineering. s45 00:03:40.440 --> 00:03:44.480 You've heard the the guy before talked about Peter and Boris, right? s46 00:03:44.480 --> 00:03:45.880 So, what they say to you? s47 00:03:45.880 --> 00:03:54.880 They basically say to you, "Stop being the router between your agents and let your agents prompt each other. s48 00:03:54.880 --> 00:04:06.080 Now, what they're basically saying to you is instead of being the router and copy-pasting stuff, start fighting with Python and TypeScript libraries and different abstraction layers that will invent s49 00:04:06.080 --> 00:04:09.959 how agents should prompt each other. s50 00:04:11.040 --> 00:04:13.000 We also have protocols. s51 00:04:13.000 --> 00:04:19.840 A2A, MCP, ACP, and then other protocols crypto-related. s52 00:04:19.840 --> 00:04:20.640 Simple. s53 00:04:20.640 --> 00:04:22.880 Solves all the problems, right? s54 00:04:22.880 --> 00:04:27.080 So, this is how your multi-agent system looks like probably in production. s55 00:04:27.080 --> 00:04:31.760 MCP, maybe if you're super advanced, it's A2A. s56 00:04:31.760 --> 00:04:32.720 And sounds simple. s57 00:04:32.720 --> 00:04:38.000 You call agents as tools, maybe you connect to agents through A2A, and life is great. s58 00:04:38.000 --> 00:04:42.120 But, MCP, calling agents as tools, it's completely stateless. s59 00:04:42.120 --> 00:04:47.160 If you want two agents to be stateful, sticky sessions, good luck to you. s60 00:04:47.160 --> 00:04:49.600 A2A is client-server. s61 00:04:49.600 --> 00:04:59.040 If I'm an agent, I can send a task to another agent, but if this agent also wants to send task to me, we both have to be a client and a server. s62 00:04:59.040 --> 00:05:04.160 Chaining multiple agent equals together involves REST API timeouts. s63 00:05:04.160 --> 00:05:05.640 Good luck managing that. s64 00:05:05.640 --> 00:05:13.640 And obviously, what no one says to you, you need queues with persistence and so on to keep track of the messages that are being sent. s65 00:05:13.640 --> 00:05:18.840 And of course, discovery, which is not even part of the A2A protocol. s66 00:05:18.840 --> 00:05:22.200 So, you're basically doing the planning work. s67 00:05:22.200 --> 00:05:24.240 You're not creating a multi-agent system. s68 00:05:24.240 --> 00:05:26.800 You deal with planning. s69 00:05:26.800 --> 00:05:30.280 But, we have messaging platforms. s70 00:05:30.280 --> 00:05:31.360 Slack. s71 00:05:31.360 --> 00:05:33.240 Anthropic released a wonderful agent. s72 00:05:33.240 --> 00:05:35.680 You can talk to it in Slack. s73 00:05:35.680 --> 00:05:39.880 We have personal agents here and I connected to Telegram, right? s74 00:05:39.880 --> 00:05:41.880 Wonderful. s75 00:05:42.120 --> 00:05:46.120 To connect an agent to a Telegram, five steps. s76 00:05:46.120 --> 00:05:49.480 Discord, seven steps. s77 00:05:49.720 --> 00:05:52.520 Slack, eight steps. s78 00:05:52.520 --> 00:05:55.520 WhatsApp, 11 steps. s79 00:05:55.520 --> 00:06:03.520 Every step is manual that you have to do by hand and you have to read the documentation. s80 00:06:03.520 --> 00:06:07.560 Again, you're doing doing a lot of planning and manual work. s81 00:06:07.560 --> 00:06:15.480 And this gives you only one thing and one thing only, an agent that can talk to a person, usually it's you. s82 00:06:15.480 --> 00:06:17.840 Your agent is still alone. s83 00:06:17.840 --> 00:06:18.880 They cannot see each other. s84 00:06:18.880 --> 00:06:21.560 They cannot communicate with each other. s85 00:06:21.560 --> 00:06:24.800 They are in digital solitary confinement. s86 00:06:25.600 --> 00:06:29.240 So, let's unpack the statement I heard from the CTO. s87 00:06:29.240 --> 00:06:32.880 Is multi-agent coordination a hypothetical problem? s88 00:06:32.880 --> 00:06:39.840 Clearly, if you're copy-pasting stuff between two sessions, this is the problem of today and it's not hypothetical. s89 00:06:39.840 --> 00:06:43.320 Plugging in MCPs and A2A, this is today's problem. s90 00:06:43.320 --> 00:06:45.480 It's not a future problem. s91 00:06:45.480 --> 00:06:48.240 Is it possible to avoid the problem? s92 00:06:48.240 --> 00:07:00.080 Clearly, it's not possible because otherwise it would have all work with only one session and not two and we would have no need to call other agents as tools and so on. s93 00:07:00.080 --> 00:07:05.200 But the question is, is it possible to keep things simple? s94 00:07:05.800 --> 00:07:12.560 And the answer is, if we look at all the options that we just talked, not really. s95 00:07:12.560 --> 00:07:20.280 But how hard can it be to connect two sessions, two processes on my laptop together? s96 00:07:20.280 --> 00:07:24.400 How hard can it be to connect my cloud to Lambda? s97 00:07:24.400 --> 00:07:33.200 Or Salesforce agent to the Databricks agent to SAP agent to my my my Codex? s98 00:07:33.400 --> 00:07:35.800 Actually, it's pretty hard. s99 00:07:35.800 --> 00:07:36.919 Think about it. s100 00:07:36.919 --> 00:07:43.600 These agents, even your two sessions are processes that have to talk to each other through a network. s101 00:07:43.600 --> 00:07:46.360 So, it's a distributed systems problem. s102 00:07:46.360 --> 00:07:51.760 And distributed systems are hard even before you have introduced agents on top of it. s103 00:07:51.760 --> 00:08:00.480 And multi-agent system where every agent is remote is basically a distributed system of microservices where each microservice is non-deterministic. s104 00:08:00.480 --> 00:08:02.560 So, it is hard. s105 00:08:02.560 --> 00:08:05.920 What do you need to solve all of that so it will become easy? s106 00:08:05.920 --> 00:08:08.920 You need to solve the transport layer. s107 00:08:08.920 --> 00:08:13.840 Ordered message delivery, real-time message delivery, retries, and so on. s108 00:08:13.840 --> 00:08:16.840 You need to solve continuity. s109 00:08:16.840 --> 00:08:21.320 Microservices, your agents are software, pods, dockers crash, and so on. s110 00:08:21.320 --> 00:08:23.919 So, you need to have a consistency hydration. s111 00:08:23.919 --> 00:08:28.040 You also need to do the run time binding between different agentic frameworks. s112 00:08:28.040 --> 00:08:32.840 You have thread IDs, conversation IDs, execution IDs, and so on. s113 00:08:32.840 --> 00:08:38.200 And someone needs to map all of these IDs together so your agents can actually interact. s114 00:08:38.320 --> 00:08:40.120 But, it's also not enough. s115 00:08:40.120 --> 00:08:43.400 Your agents cannot communicate at an IP port level. s116 00:08:43.400 --> 00:08:46.360 They cannot communicate at the URL level. s117 00:08:46.360 --> 00:08:52.560 They cannot even communicate at the pub sub level because it's still a lot of plumbing that you need to do as an organization. s118 00:08:52.560 --> 00:09:06.480 So, for this wonderful future of agents talking to each other, we need to raise abstraction of a technical stack to conversation and talk about rooms, channels, participants, and figure out the deterministic s119 00:09:06.480 --> 00:09:11.120 routing of messages within a channel and also across different channels. s120 00:09:11.120 --> 00:09:13.880 And even if you solved all of that, it's still not enough. s121 00:09:13.880 --> 00:09:21.280 You need to solve the governance layer, the identity, audit, and etc. So, I would like to introduce Band. s122 00:09:21.280 --> 00:09:25.200 This is exactly what we solved so you don't have to. s123 00:09:25.200 --> 00:09:32.120 We connect every agent together and we are global collaboration layer for all the agents. s124 00:09:32.120 --> 00:09:35.360 Any framework, whenever they are deployed. s125 00:09:35.360 --> 00:09:37.400 Inside, it's not just a communication. s126 00:09:37.400 --> 00:09:43.560 We implemented every primitive that is required for your agents to talk to each other. s127 00:09:43.560 --> 00:09:46.240 Let's see them. s128 00:09:46.240 --> 00:09:53.640 What you will see are two different users connected to the collaboration layer, Vlad with a personal assistant and Mike that has no agents. s129 00:09:53.640 --> 00:10:00.640 On the right in the terminal, I'm going to spin up different agents and they will be onboarded into the platform. s130 00:10:01.600 --> 00:10:05.760 So, let's see how fast it is to onboard a new agent. s131 00:10:05.760 --> 00:10:10.960 We just spun up a new agent, programmatic registration and onboarding and agent card appears. s132 00:10:10.960 --> 00:10:12.560 This is a Codex agent. s133 00:10:12.560 --> 00:10:14.560 Next, we spin up a Land Rush agent. s134 00:10:14.560 --> 00:10:16.360 This Land Rush also appears in the platform. s135 00:10:16.360 --> 00:10:21.640 From this moment, they know that they exist and they can talk to each other. s136 00:10:21.640 --> 00:10:27.760 Now, we will ask Codex agent to send a connection request to my personal assistant. s137 00:10:27.760 --> 00:10:31.280 Keep in mind, different users, different registries. s138 00:10:31.280 --> 00:10:37.080 There will be a connection request, contact request sent to my personal assistant to require bilateral consent. s139 00:10:37.080 --> 00:10:50.120 It will arrive here in a second and from the moment that I approve this, Codex will know and see my personal assistant, will be able to invite the personal assistant into the conversation s140 00:10:50.120 --> 00:10:55.040 and send messages to this personal assistant. s141 00:11:02.520 --> 00:11:05.640 So, we're asking the Codex to invite Andy. s142 00:11:05.640 --> 00:11:10.839 Andy got invited, received the message, reported back. s143 00:11:10.920 --> 00:11:14.400 But, let's talk about problems of today. s144 00:11:14.480 --> 00:11:15.960 We are all developers. s145 00:11:15.960 --> 00:11:19.920 We use multiple sessions, probably a lot of sessions. s146 00:11:19.920 --> 00:11:21.040 We do routing. s147 00:11:21.040 --> 00:11:22.920 We don't like it. s148 00:11:22.920 --> 00:11:29.360 And when we go to uh have a snack, we come back and we have no idea what our agents have done. s149 00:11:29.360 --> 00:11:39.120 And you as a manager have no idea uh how much it cost distribution uh of ticket and the tokens and so on and so forth. s150 00:11:39.120 --> 00:11:44.440 And you have no idea even how long your human was involved in the work. s151 00:11:44.920 --> 00:11:46.560 I would like to introduce Jam. s152 00:11:46.560 --> 00:11:48.280 Jam is an internal product. s153 00:11:48.280 --> 00:11:53.440 This is how we develop the software and we've built it on top of bank. s154 00:11:54.800 --> 00:12:06.640 Jam is a desktop application that simplifies the onboarding of your local agents to the platform and solves the problems that we just mentioned. s155 00:12:06.640 --> 00:12:18.480 So, we solve routing, we solve context overload, we solve cost management and attribution, and we allow multi-agent and multi-human collaboration together. s156 00:12:19.720 --> 00:12:26.760 So, what you see here is local agents and remote agents working together. s157 00:12:26.760 --> 00:12:36.200 And we capture all the tasks generated by Claude and Codex that they generate for themselves when they do the work. s158 00:12:36.200 --> 00:12:41.000 And we present this to you so you can track the work that these agents are doing. s159 00:12:41.000 --> 00:12:46.720 And this can be your local sessions or your local session with a session of your friend. s160 00:12:46.720 --> 00:12:52.880 Because trying to understand what your agents are doing, 1 million tokens multiplied by three, that's a lot. s161 00:12:52.880 --> 00:12:57.240 You need a completely different way to understand what your agentic team is doing. s162 00:12:57.240 --> 00:13:03.760 Moreover, we provide a way for agents to describe the layout of the software architecture that they are working on. s163 00:13:03.760 --> 00:13:06.520 This is what you see on your right side. s164 00:13:06.520 --> 00:13:15.600 And you can see in real time where your agents are working right now, what piece of component they're touching in real time. s165 00:13:15.600 --> 00:13:17.800 And once they are done, they mark it as done. s166 00:13:17.800 --> 00:13:21.520 If there is a human in the loop involvement, you'll get pinged. s167 00:13:21.520 --> 00:13:24.600 Now, you don't have to use this desktop application. s168 00:13:24.600 --> 00:13:38.120 You still can open your terminal and you can work from your terminal, but because every communication goes through the network, we can monitor all that stuff and we can surface s169 00:13:38.120 --> 00:13:44.280 a lot of other very useful information and we can enable my agent join your agent. s170 00:13:44.280 --> 00:13:54.440 We can enable our team member works remotely to join the session together with the agents and the humans so he can help solve us with some problems. s171 00:13:54.440 --> 00:14:01.360 We if you have a security guy who maintains skills for the security agent, I don't need to copy his skills. s172 00:14:01.360 --> 00:14:09.400 I can just ping his agent to join this conversation and solve this for me. s173 00:14:09.440 --> 00:14:14.600 Now, I'd like to show you the platform itself. s174 00:14:14.600 --> 00:14:17.520 So, since we're all managers, we'll start with graphs. s175 00:14:17.520 --> 00:14:24.600 Uh so, the moment you open application, you can see all the stats of all the traffic that happened between your local agents and also your remote agents. s176 00:14:24.600 --> 00:14:34.839 You can see for instance over here, I have a full stack developer uh $2,000 in tokens and this is a local cloud session. s177 00:14:34.839 --> 00:14:38.360 You can see an architecture over here, $600. s178 00:14:38.360 --> 00:14:45.960 This is a local coded session and the rest are different agents running in different environments. s179 00:14:45.960 --> 00:14:48.120 But, how do they work together, right? s180 00:14:48.120 --> 00:14:54.720 Do we have a bunch of Python code triggering the agents so they can work together? s181 00:14:54.720 --> 00:14:56.320 We do not. s182 00:14:56.320 --> 00:15:07.960 Because all models right now, they're trained on a lot of data so they understand very, very good how to communicate through messaging platforms. s183 00:15:07.960 --> 00:15:12.120 So, here I have real work that I've done this morning. s184 00:15:12.120 --> 00:15:22.960 Engineering manager, developer, and architect, different instances of quad quad working together, reviewing PRDs and SRS, we are implementation. s185 00:15:22.960 --> 00:15:26.040 there is no need to hand code all the loops. s186 00:15:26.040 --> 00:15:29.000 They know how to do it natively. s187 00:15:29.040 --> 00:15:33.960 And for the managers, we have full statistics. s188 00:15:33.960 --> 00:15:38.120 Let me kill off the manage the the engineering manager who sent too many messages. s189 00:15:38.120 --> 00:15:40.400 So, you can see the attribution. s190 00:15:40.400 --> 00:15:42.960 You can see the full [clears throat] user chain cost. s191 00:15:42.960 --> 00:15:52.720 So, if you ask a question, if my developer is actually involved in the code he pushes as a PR, or it's all AI slope, you can see it here. s192 00:15:52.720 --> 00:15:56.480 Not only locally, right, but also through the remote agents. s193 00:15:56.480 --> 00:16:03.160 You can see attribution by developer or by teams of your agents and developers and how they work. s194 00:16:03.160 --> 00:16:06.680 Everything is gets updated in real time. s195 00:16:07.560 --> 00:16:12.440 You can see all agents, and every agent is basically a session, right? s196 00:16:12.440 --> 00:16:20.000 So, if I click here, I can see all the sessions of my Cloud and Codex instances that are running and they're connected to the platform. s197 00:16:20.000 --> 00:16:22.640 They're connected to a global platform. s198 00:16:22.640 --> 00:16:28.040 So, if I want, I can connect any of you to any of my agents in 30 seconds. s199 00:16:28.280 --> 00:16:33.920 Obviously, you can set permissions, you can see all the rooms, you can see all the errors, and so on. s200 00:16:33.920 --> 00:16:35.800 And also, you can see work. s201 00:16:35.800 --> 00:16:42.200 What work was done, what is pending, what is in progress, and everything works in real time. s202 00:16:43.720 --> 00:16:44.880 Thank you very much. s203 00:16:44.880 --> 00:16:55.320 If you want to know more about the future software development that does not involve you pulling in another 50 packages, and if you want to enable your Salesforce and Slack and Data Bricks s204 00:16:55.320 --> 00:17:01.000 and Cloud and Codex to work together and collaborate, come to our booth LG17. s205 00:17:01.000 --> 00:17:11.880 QR codes for the Band infrastructure layer and for the Jam, the uh desktop application to allow you to be part of this future that everyone talks about but has never seen. s206 00:17:11.880 --> 00:17:14.040 Thank you. s207 00:17:14.256 --> 00:17:16.256 [applause] s208 00:17:33.680 --> 00:17:35.680 Hey.