8. Google Anti-gravity Interface Tour: Mastering Agents, Rules & Workflows

Key Takeaways

Hello learners and welcome back. In this video, we will get familiar with the antigravity interface. The goal here is to understand how different components are organized and how they work together when building AI agents. By the end of this video, you should be comfortable navigating the interface and configuring agents effectively. So, let's get into it. This is the main antigravity interface. It is structured into multiple sections. The agent workspace, the configuration panels, the tools and extensions, and the execution and debugging. Each section plays a role in defining, running, and managing agents. So, for the first component is the agent manager. Let's look at that. From here, you can open the agent manager. After you click on the open agent manager, this window will appear. This is the agent manager. Here, you can create and manage different agents. Each agent consists of a system prompt, the selected model. Each agent consists of a system prompt, the selected model, rules, and the workflows, and access to different kind of tools. Think of an agent as a combination of instructions and capabilities. So, let's try prompting something. Let's configure the different models from here. I chose Gemini 3 Flash. The agent is starting, so let's see the response it gives. Inside an agent, the most important component is the system prompt. It is basically searching the workspace that I already have. This defines what the agent does, how it behaves, and when it should use tools. A clear and structured prompt significantly improves the agent performance. Since I did not give a very detailed prompt, it is looking at the different files that I already have. But, if you give a very detailed prompt, like give the positioning or basically the folder path as well, then it will give you a direct response. So, your prompt needs to be very detailed and very specific. Moving ahead, we can see the different kind of models we have in the drop-down. Even in the antigravity interface, we have different options for the model as well. So, let's look at that. Here, you can choose different kind of models that we have. In the drop-down, we can see Gemini 3.1 Pro, Gemini 3 Flash, Claude Sonnet 4.6, Claude Opus 4.6, and GPT 4 with 120 billion model. Each model has different characteristics. Some are faster, some are better at reasoning, some are optimized for tool usage. For workflows involving tools, models that supported structured reasoning are preferred. For example, if I switch to a faster model, responses may be quicker but less detailed. If I switch to a more capable model, responses become more accurate and structured, especially when tools are involved. Now, let's look at how we can customize the rules and the workflows that we learned earlier. If you click on these three dots, you have the option to see the customization. In the customizations, we have rules and workflows. Rules basically help enforce consistent behavior across the agent. These are additional instructions laid on top of the system prompt. Let's try adding a rule. It's your choice whether you want to add it globally or in the workspace particularly. I'll try adding it in the workspace. I'll give it a rule. So, you can see that even while adding the rule name, we have certain rules that we need to follow. You can't enter the name in capital letters, so it has to be small. So, we over here, we have activation mode. It basically gives us the choice whether you want to apply that rule to a particular workspace or always want to keep it on, or basically move the decision to the model. So, let's try to keep it manual and First, let's try to describe the rules for the agent to follow using markdown. I'll just write it in in plain English. Always keep the We have added the rule in our workspace. Now, we'll try making use of it by asking a particular question to our agent. Let's look at that. So, over here, we can see the multiple options that we have. If you want to mention a particular file directly, or basically use the directories, MCP servers, rules, or the conversation, we can make use of at the rate. Over here, we'll choose the rules option. If you have multiple rules, then we can go for that as well. Since we only have one rule to choose from, we'll go for rule.md. Then, we click on send. Let's see what it responds. Over here, we can see that it gave us the response particularly in three lines because of the rule that we added. You will notice that the responses become shorter and more concise. Rules are useful when you want to standardize behavior without rewriting the entire prompt. You can add multiple rules as well, or basically choose the activation mode. If you want to add the rule globally for all the tasks that you will be performing in the future, you can choose the global method. Now, let's move on to the next component. In the customizations, we saw that we have another thing called workflows. Workflows basically allow you to define reusable prompt templates. Instead of writing prompts repeatedly, you can save structured tasks and reuse them. Let's try adding one in the workspace. I gave the name to the workflow as workflow one. You can add a description to the workflow in maximum of 250 characters. After that, you have to enter the content for the rule that the agent will follow. So, in description, I'll add testing. And then, in the content, I'll add summarize the following text in three bullet points. For that, I'll give this only. Now, instead of typing the full instruction every time, I can just trigger this workflow and provide input. So, let's give it in the agent and type. I'll save this one. Over here, I'll just make use of the workflow that I have added. We can see that it has showed us the option to choose the workflow. I'll choose the workflow one, and over here, I'll give my content. Let's see if it works or not. Ideally, it should give me the content in three bullet points as it has been described in the rule of the workflow. Okay, it says that the content was not saved. So, let's try saving it again, and then we'll run. Let's run it again. So, we can see that it has provided us the content in exactly three bullet points. Now, instead of typing the full instruction every time, I can just trigger this workflow and provide input. This is especially useful for repeated tasks, like content creation, summarization, or analysis. Now, let's move on to the next component. Over here, we can see that we have many options, like extensions, running and debugging. Let's look at one of them. This is where we connect external tools to the agent. Like we have Jupyter Notebook, Docker, NPM IntelliSense. You can choose any one of them to connect and run it inside the antigravity. For example, earlier we created the weather tool using MCP. Once connected here, the agent can call it whenever required. We have remote explorer. The remote explorer allows you to connect to external environments. This is easy. Cut. This is useful when accessing remote systems. Like we have version controls, GitLab, GitHub. It basically allows us while working with external files, or basically running code outside your local setup. Other than that, we have running and debugging. This is where you test and debug your agent. You can run prompts, observe responses, monitor tool calls, and identify the different issues we have. So, if you click on run and debug, you can choose the option that we have. Here, you can see the agent response, whether a tool was used, and how the output was generated. This helps in understanding and improving the agent behavior. Now, the basic question that we mostly come across is how do you connect your Google account? You can connect your Google account to enable integrations like Google Drive, Gmail. This allows the agent to interact with the real-world data securely. Over here, you can just sign out and then sign in quickly. Another component that we can quickly discuss is the agent modes. Agent modes usually define the how the agent operates. For example, the basic response mode, the tool augmented mode, and the multi-step reasoning mode. You can switch between the modes and run the same query. You will notice that in advanced modes, the agent performs more structured reasoning and may use tools more effectively. At this point, we have explored all the key components, agents, models, rules, workflows, tools, and extensions, debugging, and execution. All these elements work together to define how an agent behaves and interacts with external systems. So, this was a walk-through of the antigravity interface. The key takeaway is that antigravity provides a structured and unified environment to build, configure, and scale AI agents efficiently. In the next video, we will start building more advanced workflows using these components. Thank you for watching.