[0:00]I like how in super set right here in a terminal tab, you know, at any given time I'm typically flipping between different work trees within one project. And sometimes I'm even working on additional projects and I can flip between them just like that, which is, which is really nice. Almost a year ago I published a video on this channel about multitasking with cloud code using get work trees. I knew it was possible back then, but it didn't really stick. Not for me, not for most of us. It's only recently here in 2026 that I actually find my of working with multiple agents on multiple tasks at the same time, almost all the time. Sure, the tools got better. We have stronger models now, harness like Claude Code keep getting better. That's all true. But that's not what caused me to change my behavior as a builder. In fact, in the last four years, I noticed a distinct change in my workflow and it's not until 2026 where multitasking with agents really shows up in my day-to-day. So I'll show you what multitasking looks like for me today, but before I do that, let me trace back how we got here. Because back in 2023, this was before AI really entered my work. I was designing, coding, writing, everything by hand. And this was the craft that I had been practicing my whole career. In 2024, I think we'll look back on as the early days of AI in our work. This was the enhancement phase. It started speeding up the work that I was already doing by hand. So things like tab completion, finishing lines of code using ChatGPT to punch up a headline or some marketing copy. So it was a power up, but it was not really changing my workflow. 2025 is when AI became my collaborator and this was the first real shift. So spec driven development started to take over. I started running all of my creative work through an agent, not just code, but writing too. And by the end of 2025, I wasn't writing a single line of code by hand. Here in 2026, I'm orchestrating agents and it wasn't until recently that I really embraced multitasking with agents. Because here's what I figured out. Spec writing is a practice just like any other craft. Right now about 90% of my hours are spent shaping and refining specs with agents. Only about 10% is reviewing what they've built. And the reason that balance works isn't because the models and hardware has got better and they did. It's that I kept pushing more of my focus and my creative energy into product thinking and shaping specs for my agents to build. So once it became clear that specs are the real output of my creative work and once those specs became larger and more ambitious, working with just one Asian at a time started to feel like a bottle neck. And look, at this point, multitasking isn't a hack or a power move. It's just what comes next as we evolve as builders with AI. So I'll show you what this looks like in my world today. By the way, if you're new here, I'm Brian Castle. Here on the channel, I show you how we level up as builders with AI. And every Friday I send my builder briefing. It's a free five-minute read on taking that next step as a builder with AI. You can get yours by going to builder methods.com. By the way, in builder methods Pro I just launched a new course called become the builder. It's about making that jump from vibe coding to building the way that the pros build. You can join us by going to builder methods.com/pro. So the first place multitasking shows up is in my actual work as a builder. So when I'm working on code, I'm typically moving between two to four different features, sometimes across different code bases at any given time. And so given how much time I'm multitasking with agents, I'm finally moving away from the IDE plus one terminal. And I'm feeling more comfortable in these modern interfaces for agentic development. So let me show you a few of these tools that I've been trying out lately as my daily driver for multitasking. So lately I've been on a bit of a quest to find my new daily driver for all of my coding and and building work. And you know, because it has occurred to me that I am really multitasking a lot more. And I'm interacting with multiple agents and that's really what all of these modern agentic development tools are sort of designed around these days. So, you know, I had been coming from cursor. I've been using cursor for, you know, a couple of years now. It started out as a VS code fork, you know, but they've since released a new version. I think this is cursor version three and and you know, this one has like the the agent side bar, so you can run multiple different repos, different projects, different agents. So that's interesting, but to be honest, I didn't really spend a lot of time with that because cursor doesn't have that native cloud code interface. You're using cursor's interface, which is nice, but I I really like to use Cloud Code uh the CLI interface. Right, this is the Cloud desktop app and I'm in the code tab, which is now over here. And so they've just recently redesigned that. So I wanted to give that a look, especially since it's from Anthropic and uh you know, since I'm primarily using Cloud Code most of the time. Sometimes I'll dabble with other models, but lately I've I've just been working really well with the Cloud Code CLI. And I used the Cloud mobile app all the time with with Cloud Code over there. So I wanted to give a look at the newly redesigned Cloud Code interface on Cloud desktop. This two has the sidebar interface where I can, you know, kind of flip through my different projects and repos and uh and kind of work on different things here. But, you know, I got to be honest, I I didn't really find this interface to be, you know, really kind of fits the way that my brain works. It's nice, but it doesn't really have all of the interface niceties that I would expect from from like a daily driver type of thing. It it's kind of cool that you can see the plan here. There's some windows for viewing the tasks. You know, I haven't really been actively using it, so I'm not super familiar with how to use it. I think it has some keyboard shortcuts, but kind of limited on that front. I also found it a bit odd that I couldn't really in some cases I could access all of my skills, but in other cases I couldn't. That might be a user error, but there were just some like weird things that kept popping up here that I'm I still need to give it some more time. But that being said, I've been actually spending a lot more time in this app called Superset. And this is actually becoming my daily driver at least right now. I've I've been kind of comparing super set and conductor quite a bit. Now, one thing that you'll notice in all four of these, they really are all starting to converge on the same general layout, right? So we've got, you know, left side bar, repos and work spaces and work trees. I can show you a bit more of that in just a minute. And then like your main agent conversation happening in the middle and typically on the right side, we have the files. Uh you can look at the files in your repo or you can look at the changes that you making. And I like that, even though I'm not actively writing code anymore, I do like to see the structure of my project and see where all the files are. It just makes it a lot easier to understand how things are getting built. And that was one of the things that I found a little bit frustrating with the Cloud Code interface and Cloud desktop. I know that they have like a file browser somewhere, but again, I had a little a bit of trouble accessing it. I don't really see a button for it. In certain cases I can click on a single file and and review the file. But here I could act at least see the entire file system for my project, which is really nice. Now, for me, as I've been kind of comparing these different tools, it really came down to super set and conductor. I think they're both really nicely designed applications, especially here on the Mac. I I've kind of gone back and forth between them. They look very, very similar and I still might flip around in the next couple of weeks. One thing that I don't really love about conductor is that your primarily interacting with Cloud through conductor's interface like rapper over it. So if if I say, "Hello, are you on?" You know, it looks nice and I can see how it's working there. They're basically wrapping the Cloud Code interface uh using like the Cloud P command. This isn't actually like the native Cloud Code. Whereas in super set, I can actually run Cloud Code here just like I'm I'm used to running it, you know, as I normally would. Now, I know that in conductor, I think that they recently added the ability to just have a regular terminal tab in here. So you can actually run Cloud Code, but it's, you know, a Cloud Code CLI here in uh in conductor, but it's not exactly the same. Like you don't see the the active icon happening over here, you don't get the the sound notification, so it's sort of like a second class feature right now. I I really hope that they switched it to more of a native feel like we have in super set. So when it comes to multitasking and super set, that's what I really love. First of all, I like how by default the first workspace is the local branch, the local repository. That's another thing about conductor, every time we open up a new workspace, it's actually creating a new work tree, which is a copy of the entire workspace into a new folder. Which is sort of essential when you're doing multitasking because you don't want the two agents to like overwrite each other. But there is no built-in way to have access to the local branch, at least not easily. You have this terminal down here, which you can sort of do some things, but a little bit out of the way, whereas I like how in super set, we have that by default. So I can do like a quick change, or I can pull latest from GitHub right here in a terminal tab in the main window. But then when I'm ready to work on something new, especially if I'm multitasking, I can just do, you know, command N, and I can say add a joke to the footer of homepage. And so now that is going to actually fire up a new work tree and I can work on that here. It's actually gets to work right away. Sometimes I'll I'll escape out of it and uh and give it a little bit more instruction. So, you know, at any given time, I'm typically flipping between different work trees within one project and sometimes I'm even working on additional projects and I can flip between them just like that, which is, which is really nice. So, I'm still sort of jumping around tools right now. I really like super set. I believe that they have a new version coming out pretty soon. So, I'm looking forward to taking a look at that. But yeah, when it comes to building, I'm I'm really focusing most of my attention on this left side bar and using work trees. So, if I just finish something up, I'll commit that change to this branch. Again, we are working on a work tree, which is currently a copy of the whole code base. But when I'm ready to, you know, ship this or, you know, deploy it or merge it back into main, all I do is command shift P, and that sets me up for PR. I can create that, and then I can merge it back into into main very easy. And then I actually have a deployment process so that when I merge into main, we actually deploy. And depending on the project, I might deploy into staging first, and then main, so that's another story. So I don't know if the rise of multitasking is driven by these newer agentic development tools, or if we're just evolving as builders. I think it's a bit of both. Either way, it's clearly a pattern worth paying attention to. Now let me tell you about the sponsor for today's video, Consensus, because this one actually fits what we were just talking about. Here's the thing. When your agents are doing research on your behalf, whether that's during spec planning, or when you're reviewing what they've built, they're usually pulling from a web search, which means they're pulling from the same noisy blog posts and SEO content that everyone else is. That might be fine for low stakes stuff, but if you're building anything where decisions actually matter, health care, Fintech, enterprise tools, anything where you might have to report or justify your technical choices, the quality of the data that your agents are working from suddenly matters a lot. Consensus is a different data layer for your agents. It's an MCP server that you can plug into Cloud, Cloud Code, Cursor, Codex, any MCP client. And it searches over 200 million peer-reviewed academic papers, not just blog posts, actual published research where every result comes with citeable sources that you can point to. And the way consensus returns information is what really makes it useful in an agent workflow. Because every finding comes back with line by line citations linking to the peer-reviewed papers that back it up. So your agent isn't just giving you the answer, it's showing you exactly which studies it pulled from, where the claims came from and letting you verify the source. That's the kind of trail that you actually want when decisions matter. Now, while consensus has typically been used for academic researchers, their MCP makes it a great tool for builders too. And I can see this being especially useful in two places in a typical workflow. The first is during spec planning. When you're deciding on an approach for a new feature, a UX pattern, a user behavior, accessibility, a security trade-off, you could have your agent pull peer-reviewed research on the decisions before the spec gets finalized. The second is during review. So when your agent ships a PR, you're not just reviewing the code, you're reviewing the decisions behind it. Consensus is a way to validate that those decisions line up with actual research, not just what an LLM thinks sounds right. So you can try it for free by going to consensus.app, and you can connect it via MCP in a snap. All right, now let's talk about context switching for a second because that's the mental barrier that I think prevents many of us from really embracing multitasking, even though we know that the tools are capable of it now. But I'm starting to think of it more like context blending because more often than not my product work is directly related to my marketing work. And I'm even building agent skills to tie these two together even more closely. Let me show you.
[15:19]So, lately I've been building a lot of these internal tools. So this one is called Sparkdrop and this is where me and my agents sort of manage my pipeline of new content ideas that were developing and then scheduling out and building out my newsletter. So this is, you know, internal and uh designed around my own workflow. I've got another one called brain down and this is where I'm able to edit and read mark down files and easily share them with my agents. It all integrates nicely with Dropbox. It's just a really nice kind of markdown file manager, if you will. So these are my internal tools, but I'm starting to release them publicly in the form of build kits or starter kits, if you will, that I'm releasing to build methods pro members. So that, you know, members can build their own versions of these same tools starting from my prompts and plans and PRDs and specs and then my video guides that come along with it. So it's a really nice way for builders to be able to build exactly the same thing I built or deviate and customize it to your own internal needs. That's kind of the beauty being able to build your own internal tools like this. I'm going to show you a bit more of Sparkdrop later in this video, especially how I have my agents working and interacting with it. While I'm also interacting with it as well, I'll get into that like agent multitasking thing. But what I want to talk about here is how I'm sort of blending marketing with these products. So if I, if I sign out of these accounts and then just look at the homepage, so you can see at brain down.app and at sparkdrop.co. I have these kind of marketing pages built out for these. Now, I'm not actually offering these as like stand alone software tools or SAS tools at this time. Again, they're they're just tools for me, but I still like to have like a a public marketing page at least to explain what this tool is since I like to share the tools that I build here on the channel and share the starter kits. So to create these marketing pages, I actually did that directly in the code base for the products themselves. So like in Sparkdrop, this is just the the root page for it. But I built a specialized skill that I can always come back to any time I built a new tool to quickly and easily spin up a new one-page marketing page for the homepage, right? So I can easily call it up by just calling like app marketing page and then I can just have it run the process. Since I've already done that, I'll just show you the skill itself. So in my collection of internal skills, I've got this one called app marketing page. And then if we look at that, it's uh this is how I tend to organize the skills that I have my agents work on. So I usually have like the one main skill.md, which is the overall process and it has, you know, multiple phases. What it's going to do is this one will analyze the code base and then interview me about like what I think the most important messaging points are about this. Then it's going to use the front end design skill to design the page and then it will implement the page. And then I I usually have this folder called steps where I will kind of flesh out the individual instructions for each of these steps. Now, you know, this is how I wanted to design the process for making these these marketing pages the way that I want them. So it's just really easy to work with Cloud to craft the process however fits your business and your workflow. So I'd like to have these like reusable skills built in here. So that's an example of how I'll often just switch between building on the tool itself and then have an agent work on the marketing page directly in the same code base, sometimes in two different work trees in the app. Now, in addition to the marketing pages, I'm also starting to extract more ideas and assets from the products that I'm building. So like I talked about those starter kits, I actually created an agent skill and this one is more complex for generating the starter kit from the actual code base that I built out. So the the my plan for the next couple of weeks is to build more of these tools similar to like spark spark drop and what I call brain down and I've got a few more in the pipeline that I'm that I'll be building. But then I'll be turning them into starter kits that that members can uh can get access to and build from. And so I have an agent's skill that will do the same analysis, build out the PRD, the tech stack, the design system, and then, you know, build out some ready-made prompts. And there's a lot more to it than that. I'm I'm giving a lot more input along the way. And and so each of these has some like interview steps for me to to work through. These are all the different steps in the process of creating a starter kit. Um so, you know, again, it's like building out these operations in my business, turning them into agent skills. And then building them into my actual workflow where I can easily flip between having an agent work on a marketing task, creating an asset or actually building out one of my products and really doing these side by side sometimes at the same time. One more of these skills that I just recently set up is called Plan videos from build. So again, I'm trying to extract more ideas and takeaways and design patterns from my actual work itself building in the products to turn them into concepts that I can share on videos for YouTube, right? So I created a skill for that and I'm going to be doing more of these in the coming weeks here on the channel. So again, it's going to analyze the code base, it'll analyze my conversations that I've had with Cloud and the process for how I went about building the building whatever tool it was that I built. And then we have a bunch of steps for actually drafting and structuring and outlining a video for me to then develop and record and and come out with. So product and marketing used to be two totally separate workflows and now they run in parallel sometimes inside the same code base. Now here's a form of multitasking that I've been waiting for for a year but it's only really doable now. And that's having agents keep working on my projects even when I'm not at work. I'm talking about mobile and it's kind of addicting.
[23:13]Now, I did a deep dive into all the ways you can use Cloud Code on mobile just a few weeks ago here on the channel. But to give you more context, I'm seeing a few patterns in my day-to-day. Before I finish up my day here in the office, I give Cloud a big task to keep working on while I go to dinner and hang out with my kids. And most nights I grab my phone and kick off a task just before bed so that it's working while I sleep. And the next morning, you guessed it. I'm grabbing my phone and I'm having it plan a new task while I go do some exercise and get my morning coffee going. You know, a lot of this is about optimizing the timing of our interaction with AI agents. Because a lot of it is just waiting for Cloud to analyze an entire code base or do some research or churn through a big spec. So instead of waiting there and watching Cloud do its thing, I'd rather kick it off and either go enjoy my free time or start planning the next feature in a different work tree. Now 2026 is also the year of agents, which means that on most days there are tasks being done by an open claw agent or a cloud work agent that I didn't need to prompt or even watch it work. They're just members of my team doing their jobs on schedule, following the process every time. Now, I've covered my agent setup in a few videos recently here on the channel and inside builder methods Pro. But here's a fresh look at what my agents are actually doing most days. So, I showed some of this in my previous videos on Openclaw. This is my custom dashboard that I use to assign and dispatch recurring scheduled tasks to all of my agents. So on open cloud, which is running on a dedicated Mac mini here on my desk. I have four different agents and each one of them is responsible for a number of different tests. For example, my agent named Gumbo does like a daily synthesis of my work and to do and intakes a lot of like files and stuff and then sends me a a morning report that looks sort of like this. I get the Met score, sometimes the nick score, my to do list and a summary of what happened yesterday. We capture all that stuff into like a a brain file system. And so I can manage it all in this custom dashboard here and so each of these tasks can map to a custom skill that I've built into my system. By the way, this dashboard for dispatching tasks and scheduling them on a recurring basis to a team of agents. This is another one of those apps that I have released a starter kit, like a builder's kit for you to be able to build your own version of this. That's available inside of builder methods pro, along with my entire course on setting up a team of agents with open cloud. So that's a quick look at my agents on open claw. Again, I I shared some of this in my other videos, but this this agent handles a task that I call content radar. And that one is basically following some tweets and news from the industry for me to pay attention to and we can maybe, you know, use that as material to create some new content. Now, all of these are tied to skills. So you can see like daily synthesis, content radar scan, intake processing, like a weekly report. And so these are defined in my collection of skills. For example, here is the daily synthesis skill that my agent follows. And in my open cloud series in Builder Methods Pro, I go through my entire setup, how I wire it up, how I configure the agents, how I have them able to read a specific skills and especially dealing with multiple agents. It can get pretty involved, but I share all the cheat cheats and all the setups and everything in there. Now, lately, I've started to use Cloud co-work agents as well. And this is a really interesting pattern that I think is going to be become more common. So I mentioned that I recently built this tool for myself called Sparkdrop. And this is my me and my team, and when I say my team, I mean my agents, we use this to develop content and move it through the pipeline, whether it's ideas for new videos here on the channel, or my weekly newsletter, or things that I might put up on Twitter, or X, or LinkedIn. So they start out as sparks. These are like new potential ideas that are being pitched that I can react to, and I can either revise them, I can comment them. And I can, you know, say, yeah, let's green light this for development, and then these ideas, you know, turn into what we call flames. And so these are like drafts that my agent will give me a first draft. I can comment on it, give it some feedback. I can rewrite it myself, which I often do. And then we can, you know, schedule it into the schedule and send it out on X, or market it as something that I'm going to include in this week's newsletter. So this is an interface for me to be able to do all that stuff, but my agent also interacts with this same interface using our APIs. So I built an API and it has documentation in here. So this custom built app has an API and all these API endpoints that my agent can use to submit a new spark idea or develop a draft on a flame. And I'm actually using a Cloud co-work agent. The reason I use Cloud for that is because I wanted to be using the latest model of Opus using my Cloud Max plan. And so for that, we're using Cloud co-work. Now, here I'm actually looking into a window that looks into the dedicated Mac mini where I have my agents running. I have two different users on that Mac mini. One is for open cloud agents. And the other is like my account on the Mac mini, and that's where I run Cloud co-work inside that Mac mini. So this is like a window from my main machine into my Mac mini and this is Cloud co-work and I have this scheduled content development task which happens every morning at 6:00 a.m. and we can see that it happened today at 6:00 a.m. and yesterday. For some reason it happens like 4 minutes later, so 6:04. The agent will go through and do its thing. Now, now it is configured to use my skill, which I've named content development. And I define that and I and I maintain that back in my code base for content development. That is here. So here's the main skill for that. This is the content development pipeline, and this one goes deep. So I've got instructions to the agent on how to interact with my Sparkdrop API. And then it goes through and instructs the agent on where to find the training data. You know, we have a lot of training built into Sparkdrop on like my voice and preferences and topics that I like to talk about. And then how to read my comments and take my feedback and how to post its own comments. We also have a system for the agent to create learnings. You know, it learns from me. It learns my voice, learns my new preferences and we fold that back into the training. And then here are the pipeline steps. Again, here's that pattern where each step in our main process will link off into these sub steps. So how to generate sparks or generate new ideas for content? A whole process for that. How to integrate learnings back into uh into our work. How to write drafts and then how to like move drafts into the newsletter sections. So, you know, a bunch of steps here. I'm not going to bore you with all the nitty gritty, but that's how I structure it. And so basically what happens in the background automatically using Cloud co-work. This is running every morning at 6:00 a.m. like clockwork. Right, it's just always running and the agent using the latest Cloud Opus model is going to run my skill with all those processes and submit new ideas, new content into my Sparkdrop application. This is like the user interface, and then I can access this here or on my mobile device, and I can just give feedback and and we can approve and move content through the pipeline. So, I just really like that trifecta of like me, human in the loop with the custom built internal tool, the interface, and the agent being able to all kind of converge and interact this way. I think it works really well, and this is how we're able to, you know, develop content and just push it through the schedule while I work on other stuff, while I work on the next project or the next video. So all of this is to show that whether you're becoming a builder or you're adapting your work as a builder, it's a progression. And once you focus on your role as the human in the loop, you start to see patterns emerge, which makes multitasking the natural next step. Now, as I showed, one of those unlocks has been getting truly productive with managing agents from mobile. So you won't want to miss my recent video showing every way that you can use Cloud Code from your mobile device. First, hit subscribe so you don't miss my next video, and then I'll see you over there next. Let's keep building.
