ChatGPT or Superhuman Go: pick where to build your agents

If you open your LinkedIn or Twitter these days, you're going to see hundreds of posts about AI agents. I'm going to show you today how to build agents yourself in two popular platforms, both ChatGPT and Superhuman Go. By the end of this video, you'll not only be equipped to build agents yourself, but you're also going to know the strengths of each platform and be able to pick what platform you're going to build agents on.

As we look at building these agents, we're going to cover five core things. We're going to cover access, which is where to actually access and start building these agents in the platform. We're going to cover creating them, the experience of actually building them, either with AI or manually. We're going to talk about triggers, which is one of the most important things in agents, ensuring that these agents act when you want them to act. Connectors, which are the tools that you're going to give them access to, whether that be your email, calendar, communication tools or otherwise. And lastly, testing, to ensure that your agents work correctly. So come with me as we compare both platforms on these five.

All right, so here we are. I'm inside of ChatGPT. Let's look at it first. You're going to see a lot of different things here on the side. I have agents down here because I pinned it. If you don't see your agents down here, look on the sidebar or look under more. But as I go to build an agent, I'm simply going to press this plus button. When I press plus in ChatGPT, I get shown this screen. I can build an agent with chat, so I can simply tell it, "I want an agent that does all my work for me." You can build it with chat, you can start from a template down here below, or you can start blank and configure it manually. That's quick. That's easy. That's how you access them, and those are the two different ways that we can actually build them.

Let's look at Superhuman Go and see how theirs are built as well. When I'm in Superhuman Go, I am greeted with this screen. It says, "Here we go." On the left-hand side, I can see my tasks, and on the top, you're going to see agents. Agents really are the core aspect of Superhuman Go, so they're not going to be hidden under other menus. They're really going to be right in front of your face. I can create an agent with this plus button, and I'm going to be shown the manual setup right here. Or I can simply go to New Task and say, "I want to build a new agent that helps me with my email." All right. So those are the two different ways that you can actually access them and start building them.

Let's go ahead and start building a real agent together so we can really experience these platforms. We're going to build a very simple agent today. It's going to be an email triage agent that, in the morning, before I wake up, is going to look at my email inbox and start sorting the noise that is so common in our emails from the things that we have to actually reply to.

So let's start with ChatGPT and see what it does to develop this agent. You'll notice that ChatGPT is going to ask me different questions to clarify. I'm going to have it run every weekday at 5 a.m., because I really want to ensure that it's ready before I start. Personal must respond: I'm going to say, "You decide for me." The best thing about these is that this isn't final. ChatGPT is going to build your agent for you, and then you can go tweak it later with the manual settings. So let's go ahead and submit.

All right. So now I'm inside of the manual builder, and you'll see that I have something set up for me already, but it's not quite done. First off is my instructions. I have a larger instruction base that gives it a role. It tells ChatGPT about me so it can better understand what might or might not be important. It tells it which emails to review and how to sort. Of course, this was developed by AI itself. Anything that AI does develop, please go in, read it yourself, tweak it, and make it more relevant to your own self. Now, aside from the instructions, there are some other important things we're going to look at. Those are our connections, where we're going to connect apps, skills, files, and memory. But before we jump into those, let's look at Superhuman Go and build the same agent.

All right. So here we are with Superhuman Go. I want to give it the same prompt. I'm going to tell it that I want to build an email triage agent that helps me separate the noise in my inbox from personal emails and from people I really want to respond to. We're going to give it the exact same prompt and see what it does. Superhuman Go is doing a very similar experience. It's going to ask me clarifying questions. It's asking what email inbox I need to use. Great. Notice that I can tell it what to do as well. I want it every day at 5 a.m., because I want it really early. I'm going to submit that, and it's going to think through and build this agent for me.

All right. So it's created it, and now, similar to ChatGPT, it's given me the option to open it. Oops, I don't want to run it. Let's go ahead and edit it. So here I am inside of the manual editing screen for Go. Now I can tweak everything myself. So we've just built an agent with each platform using AI, and once it's drafted by AI, we're going to start manually tweaking it.

This next part is really important. It's about your connectors: how to connect your apps and tools in a secure way. They both handle it differently, but getting this part right ensures that your agents don't go rogue, expose information, or cause security vulnerabilities. So this is important to get right.

Let's start with ChatGPT. We're going to add a connector. From here, I am going to browse apps, and apps give me a whole host of different apps that I can connect, custom MCPs, or I can search. I'm going to connect Gmail, because that is the email that I use. We're going to add it. Once it's added, there is this configure button. I really want to press this. I really want to see what's going on here and configure it correctly. You'll see that it gives me the option of an agent account or an end user account. I want it to search my inbox, so I'm going to give it my account. But I am going to go tool by tool and not give it certain actions. I really don't want it to accidentally go rogue and start responding to things or sending emails. So these are all my write actions. I'm going to uncheck them all, and it will not be allowed to do any of those pieces. These are my read actions. I can go line by line and give it less access if I want. For example, labels: I don't really need it to do labels. I don't need drafts. I might need profile. I might need attachments. But we can leave it with the rest of those. So now I'm going to go back, and you'll see that Gmail is connected. I could browse and connect different apps as well if I wanted to add my calendar. But that's how you generally add connectors to an agent in ChatGPT.

Let's look at Superhuman Go. They've got a similar model and a similar permission style. When I'm inside of Superhuman Go, if I scroll all the way down, you're going to see I've got tools and access. By default, it has three toggles: web access, Superhuman Docs, and Gmail, Outlook and Calendar. Then there are also connections, where you can add and you can search. Again, hundreds of different connections. I can search Gmail, and there it is. It's already on for me because the agent already knew I needed it. Let's go in and see what it can and cannot do. You'll see here are all the write actions. Again, I don't need those, so I'm going to toggle them off so it cannot do those things. But can it read? Yes, it can. It can read thread counts, message counts, messages, and search messages. So we're going to always allow that, and we're going to allow that it can read. If you do use Superhuman Mail, the Superhuman product, you can just toggle this on, and that's all you need to do. That's literally it.

All right. So those are our connectors inside of both agents. Again, connectors are what make agents powerful. They can work across and with your tools, but you need to ensure you configure them so you get the security aspects right.

Next, we're going to talk about triggers, and this is probably one of the areas where these two platforms vary the most. So let's look at how they vary and how you can set them up for yourself. When I'm inside of my email triage agent here in ChatGPT, there's actually not a very clear trigger section. All that I see is in the top right, I have this automations section. When I press that, you'll see that I can do intervals, so it can run every hour, and or it can run daily on a schedule. You can pick the specific days that it runs and the time at which it runs. That's all we've got for triggers here. There actually are some hidden triggers. It's a little bit different. If we add a channel, there are two other ones we can use: Slack or an API. We can trigger it via API, or it can reply to messages in Slack.

Now, this does get a bit more complex. This video is just showing you the basics of how to set up this agent. If you want a ton more technical information that lets you build advanced agents, you can look in the description below. I've got a large resource that dives really deep into each one of these agent builders and talks about every last thing you need to know. So go ahead in the description and grab that. We're not going to dive into that today, but know that you can also trigger these ChatGPT agents either with Slack messages or with the API.

Let's go over and look at Superhuman Go and see their triggers. There are actually a lot more triggers, and it's very clear where they are. They're one of the blocks inside of your agent builder. Notice if I say add trigger, I actually have a whole section that's going to give me a ton of different triggers. I've been working with Superhuman Go since it was in beta, and this is constantly growing, so I know that there are only going to be more and more triggers that come. Here is their schedule, though. For example, I can do it daily, weekdays, weekly, or custom. I'm going to choose weekdays for right now, and I'm going to choose 5 a.m. in my time zone.

But you'll notice here that they have so many other triggers you can use. First off, they've got a proactive trigger, which means that while you're typing, it's going to be watching what you type, and it may be able to trigger as you're typing based on certain keywords or things like that. This one's really powerful, and ChatGPT and Claude don't have this type of trigger. I've got email triggers as well, like every time an email is received or sent, or if you add a specific label. I've got Slack ones as well. And then I've got docs. If you use Superhuman Docs, there are a lot of doc-based triggers as well. So there are a lot more robust trigger options inside of Superhuman Go. But for this one, I want it to be a scheduled trigger, which I already have. So those are our triggers. Triggers simply tell an agent when it should run, and Superhuman Go is going to win in this area with its more robust triggers.

All right. And lastly, testing. So now we've built our agent. We've got our prompt. We've got our triggers. We've got our names. Now we just have to test it to ensure that it works correctly. Inside of ChatGPT, we'll start here. I can press create, and once I press create, I'm going to be able to test it. The testing across these two platforms is very similar. You'll see I've got "Try in ChatGPT." I can try it. It's going to run by itself, but I can just say, "Run this agent." It's going to run the agent now and work through it. We'll come back to this in a moment.

If we go back to Superhuman Go, it's very similar. You'll see Run agent. I can simply run this agent. We'll save it. When I run it, it goes to a chat screen, just like ChatGPT. It says, "I want to run this agent," and then the agent's going to start running, searching my inbox and looking for me.

All right. So they've both been running for a while. You'll see that ChatGPT is running right now. It's still reading and checking. It's been about three to four minutes. Superhuman Go is done with its report. There we go. It's got priority and a lot more information on my email. I'm not going to scroll through it all because it is my personal email. But those are our two agent platforms.

Again, there is far more to know about how these platforms work and how to build agents effectively. If you want a deeper comparison between the two, go into the description below to get that resource. If you've got any questions about these platforms, let me know in the comments below. Thank you.