Vollständiges Transkript anzeigen (2.093 Wörter)
Okay, so the A2A CLI shipped this week. And I want to talk about it because I feel it solves an interesting problem that you are probably sitting with already today. I know I definitely was. So, think about it. Different teams inside a company have all gone ahead and built their own AI agents. Marketing has one, finance has one, supply chain has one. Each one in its own project with its own files, and the problem is none of them really talk to each other very efficiently. Now, to get agents talking to each other, there is a protocol called A2A, which is Agent-to-Agent protocol. Google built it and open-sourced it last year, and it gives every agent a common way or common language to talk. So, it does not matter what each one was built with. For example, the marketing team could have used Google ADK, which is Agent Development Kit to build the agent, and the supply chain would have used LangChain, but they could still work with each other and talk to each other. And that part has been working for a while now. What the A2A CLI adds is a ready-made client. So anything that is not one of those agents can actually talk to them as well, you from a terminal, a script that runs overnight, or your coding assistant. Anything can talk to those agents. Before this week, you would have been writing that client yourself. They also ship the skill alongside with it, what lets Antigravity or Claude core or whichever assistant you use go and do it for you. Agents are really getting more and more useful, and I truly like where this is going. So as always, I wanted to test it out myself and show you what it looks like in a real business scenario, okay? And hey, if you're new here, I am Surya, 20 years in enterprise tech, and these days I lead an AI transformation team at a fang company. Basically, I take the cutting edge of AI and make it make sense. If that's useful to you, please hit that subscribe button. For now though, grab your favorite drink, and let's dive in. A quick disclaimer before we go there. All opinions are my own and do not belong to my employer. With that, let's get into it.
So let me put that in a simpler way because there are really three different things that might want to talk to an agent. The first one is obviously another AI agent, so my marketing agent talking to my supply chain agent. That is a few lines of code sitting inside individual agents, and the CLI does not change that at all. So that is not what the CLI is meant to do. The second is something that is not an agent, think of it as a tool or a coding assistant that knows nothing about A2A. Before this, you would have to actually build it into that particular thing. Now, you could run this particular command, and that's it. The third one is you yourself or a script or a job that runs overnight. Before this, you were writing a client to do it. Now, it is just one line. So the setup is actually very straightforward. I've got two agents running here, a marketing agent and a supply chain agent. Both of them have used ADK to build running on my laptop, and I've already got A2A established between the two of them. So first, let me show you how those two work together, and then we will bring in the CLI and see what it actually adds.
All right, so what you are seeing on the screen is the GitHub page for A2A CLI. This is brand new and you can see that this is actually part of the A2A project itself. So it was released yesterday, so you can see it's really brand new. And here, they explain why the A2A CLI is required. So something that I've already talked about. And you can see how to install this, so within Windows this is the this is how you would do it, and then if you have a macOS or Linux, these are the commands that you would do. So what I would recommend is use this GitHub link and I'm going to share this link so that you can copy some of the commands and install this. What they've also shared is the agent skill which I will be demonstrating on how you can actually install this directly in your agent. So I'll be using Antigravity here so that the agent directly can talk to any of the existing A2A compliant agents. So that's something which I wanted to show first.
So as I discussed before, what I've done is I've actually created two different ADK agents. One is a marketing agent and another one is a supply chain agent, okay? So what we're going to do here is these are both A2A compliant, so they can both talk to each other, but I just want to explain this and demonstrate this as a use case. So number one, let's select the supply chain agent, okay? So let me ask this particular question that how many Ridge trail jackets do we have, and what is the lead time? So this is the kind of question that you will ask from a supply chain agent and the supply chain agent goes ahead and gets the product inventory and then comes back with a simple response. Now we will ask a similar question when we are interacting only with the marketing agent, so I'm going to ask in general, what channels work best for a Q4 outdoor gear campaign? These are like individual agents which are responding only based on the individual department's question. Supply chain looking in the inventory, marketing looking at their marketing data. Now what I'm going to do here is now I'm going to ask the same agent, but in this case this would require it to actually work with the supply chain agent. So here you can already see that it is going and talking to the supply chain agent and then based on that it was able to give us those details. So here I asked this question, and you can see the state as well. And here you can see that it has actually used the marketing agent to get to the response and even in the traces you're able to see that marketing agent does the LLM call and here it actually requested the supply chain agent. You can see the A2A client was invoked and then based on that you are seeing the response. So this is possible because I made both supply chain and marketing agent A-compliant and that is why both are able to talk to each other. At this point, the assumption is you are already familiar with Agent-to-Agent protocol and you are able to build this, but if that is something you want me to go deeper and explain you what is an A2A protocol and how to build an A2A compliant agent, please let me know in the comment section and I'll be happy to go deeper and explain you how to create an A2A compliant ADK agent, okay? So let me know about that.
All right, so now that we have seen that both these agents are working and both are A2A compliant and both can talk to each other, now the next thing that I want to show is actually using the CLI. So for that, what we will do is we'll actually get into Antigravity because what I want to show you is how you can directly talk to any of these agents from your terminal as well as your AI coding agent like an Antigravity or Cloud Coding. So we will open a new terminal here inside Antigravity, you can do it outside as well, and this is a skill command that you already saw here. So I'm just copy-pasting this and then going into Antigravity and running it. So you will see that it gives me this and says that okay, this these are all the places where I will install, and then you want some other places as well, so I'm just going to select Cloud Code and then will go with Symlink. So it's as simple as installing this, it does a security assessment pretty quickly and that's it. So it has installed that. Now once it has installed, what we will do is I'm going to ask a question directly to the agent over here. So here I'm going to say that there's an A2A agent running at http://localhost:8001. This is this particular agent, which is running here, which is the marketing agent. I'm going to ask, sorry, the supply chain agent, and find out what it can do then ask it how many Ridge trail jackets we have and what the lead time is. So for it to work, this just couldn't have worked directly like this. You would have to write like an A2A based SDK, basically some sort of a code before you can actually make this happen. But now, because of the skill and the CLI, it is able to go talk to any compliant A2A agent, and you're able to get to the answer pretty quickly. And that is what this A2A CLI brings to the table. So we'll give it a second for it to come back, but you will see that it is able to provide a similar answer. So you can see that it already says I've already inspected the agent running at this and queried the inventory, and it is going to come back with a quick response.
Okay, so there you go. It has provided the response. It also explains from inspecting the agent card via A2A, this is a supply chain agent, this is a description, these are the protocols and then this is the command that it executed, and this is the result that we have got. Now I can ask another question, there is another agent running at localhost:8002. Ask it to plan a Q4 campaign for the trail jacket. Now this is interesting because now this one should talk to marketing agent via A2A, and it should provide that response. So we are talking to an agent which will talk to another agent and then provide a combined response. So I want to test this out as well, and you can see that it is already working through that. So give it a second and then we'll come back and show you the answer. All right, so it has gone ahead and provided this and you can see that it executed this particular command. This was the prompt that was given, the marketing agent actually did an A2A tool call, so it is very clear and explaining that and this is the response that you have got. So this is a pretty cool thing because now you are actually directly talking to an A2A compliant agent whatever it is and wherever it is, from your coding assistant or in this case Antigravity which is an agent itself. So A2A CLI is here now, and you can already see that this is only going to help developers as well as anybody who wants to talk to an A2A compliant agent directly using a terminal or an IDE or any other coding agent like Cloud Code or Antigravity.
All right, so the big question, can you use it this week? Yes, and I think you should. Install it, point it at an agent, show your team what this looks like. So that's the A2A CLI. If you want to try it, the install is one download and the repository has a set of worked examples you can get through in about 20 minutes. That is exactly what I did. If you have got two agents sitting in your company right now that cannot talk to each other, I would like to know what they are. Drop it in the comments and as always, I'll do my best to answer your questions. If this was useful, do subscribe to the channel and give the video a like, it really does help the channel. Thank you very much for your time. Thank you for watching, and I will see you in the next one.
Video
