Claude Cluster is Insane… Upgrade Your Claude Code Workflow

The developers of Claude Code just silently announced massive updates that almost no one is talking about. The biggest of which is the release of Async background agents. What's strange is that you won't find these features mentioned anywhere on Anthropic's main site or even their blog. Now, you might be wondering if these are actually useful, and honestly, I wasn't sure at first either, but I found an unexpected angle that's completely changed how I use them. So, in this video, I'm going to break down exactly what that is, why it makes a lot of existing AI coding frameworks basically unnecessary, and then walk through several other Claude features. They announced these new features in the change logs on Claude Codes GitHub repo. They mentioned that these agents and bash commands can run asynchronously and send messages to wake up the main agent. The idea here is to achieve true parallel working. It is quite simple to use them. We first ask Claude to handle a task that benefits from a sub agent and then the main agent creates a sub aent for the task. When the agent has started running, we push it to the background using controlB and continue working with the main agent on other tasks. When the sub agent has completed the task, it wakes up the main agent with the results. We can also check our running agents using the tasks command. Under the hood, the background agent works as a separate process that is able to execute complex multi-step tasks autonomously, meaning it can independently handle full-scale tasks on its own. They allow you to continue interacting with Claude Code's main thread without blocking their execution, so you can use your time effectively instead of waiting. When they're done executing, they report the results back to the main agent. Previously, you had to open multiple terminals with separate Clawude instances for this purpose. Now, Claude does this itself, managing background agents internally, coordinating multiple agents and integrating the results on its own without the manual setup we were using earlier. They say background agents are ideal for researching while coding, reviewing workflows, performing parallel tasks, and handling any task that shouldn't pollute your main context window. Before we discuss whether these agents actually save us time, let's take a quick break to talk about Automata. After teaching millions of people how to build with AI, we started implementing these workflows ourselves. We discovered we could build better products faster than ever before. We help bring your ideas to life, whether it's apps or websites. Maybe you've watched our videos thinking, "I have a great idea, but I don't have a tech team to build it." That's exactly where we come in. Think of us as your technical co-pilot. We apply the same workflows we've taught millions directly to your project. Turning concepts into real working solutions without the headaches of hiring or managing a dev team. Ready to accelerate your idea into reality? Reach out at hello@automator.dev. So, I tried out what they were suggesting and from the time I've spent with them so far, I really like what I'm seeing. I think these agents have a lot of potential. These background agents can handle any research task that takes a lot of time or might bloat your context window. You can simply move that research to the background and assign a separate task to the main agent. This allows the main agent to work on something completely different while remaining totally isolated from the research task. These agents make it much easier to implement changes on separate pages at the same time by parallelizing tasks. Instead of waiting for one agent to handle everything sequentially, you can update them simultaneously. For example, if you're switching all your components to a new library, you can use background agents to update them all at once. Other than that, you're often left with tasks like reviews that take up a lot of time. You can use async agents to launch several of them simultaneously. One for a security review, one to identify performance bottlenecks, one for refactoring code, and another to check for duplications or unused code. You can run all of these tasks in the background while your main agent stays focused on a completely different feature update. This way, each agent works separately and handles its own task independently from the others. Now, here's the unexpected angle I mentioned earlier. You can even equip background agents with the ability to manage their work in a git work tree, allowing them to commit and push changes in isolation. Some of you might have already thought about this, but for those who don't know, Git work trees are an extremely underutilized feature that lets you work on multiple branches in separate directories. This means you can avoid constant branch switching in a single workspace and simply spawn as many agents for as many features as you like. To use this workflow, I set up a custom/command that creates a background agent to work on tasks in separate work trees. All the instructions for this command live right in the commands folder, which you can find inside the do.clude folder. Let's say I'm working on an app, but I'm not happy with the theme. I have three different theme ideas and I want to implement all of them to see which one looks best. Without work trees, I would have had to clone the repo separately, create three different branches, or open separate clawed instances just to keep the tasks apart. But with background agents, all of that changes. The three agents worked in parallel to implement those themes separately, all while leaving my main design completely unaffected. This allows me to simply merge the theme I like into the main project and discard the others. This setup made the implementation process significantly faster because all the changes are isolated. I can choose the theme I like best from the different options. It gives me the freedom to experiment and execute without worrying about breaking the main design or having to revert unwanted changes. In the change logs, they announced some other really cool features and one that I personally am really thankful for is instant autocompact. Compacting used to take a good 2 to 3 minutes, but now it happens significantly faster because it's likely maintaining histories for each project right in thecloud folder. Claude Code also added a prompt suggestion feature where you can simply press enter to accept a suggestion or just type a new prompt in its place. You can use the agent flag while starting Claude to run it as the specific agent you have in your project, allowing you to delegate tasks to it directly. Claude code now allows you to fork sessions and choose between them when you launch Claude with the resume flag. All you have to do is press the escape key twice and your current session will be forked. Now, even though background agents are great, there are certain areas where you shouldn't use them as they will hinder your workflow instead of making it better. For example, you don't want to background any task that requires your input because that would cause the agent to get blocked and unable to proceed. You also don't want to background any task that requires your approval while it's running. It would just get blocked in the background and you could actually lose track of it if the main agent is busy with another task. It is also highly recommended not to assign tasks that depend on each other as this can create conflicts in your agents work. Instead, it's much better to assign isolated tasks to each agent. There are a few best practices you need to keep in mind to get the most out of this workflow. While using background agents, you need to keep token usage in mind. Even though they speed up your work, each agent consumes a significant number of tokens. So, it's important to track this if you want to avoid high bills. Since agents return their output to the main agent, you have to make sure your tasks are isolated. Also, when you're spawning an agent, give it a descriptive name. That way, if you're managing multiple agents at once, you'll know exactly what each one is doing. That brings us to the end of this video. If you'd like to support the channel and help us keep making videos like this, you can do so by using the super thanks button below. As always, thank you for watching and I'll see you in the next one.