# Holy smokes: this agent is amazing!

Channel: The Next New Thing
Video: https://www.youtube.com/watch?v=W0r8xrWIYGw
Duration: 22 min
Language: English
Words: 4318
Transcript page: https://viewrankai.com/tools/youtube-transcript/W0r8xrWIYGw

---

[0:00] Is there a better way to create agents than Hermes agents and OpenClaw agents? Well, it seems like the creator of Twitter and Square has just created it. It's called Buzz. And it not only allows you to create agents, it allows you to bring your Hermes and OpenClaw agents in. It allows you to chat with whatever agent you use. It allows you to bring your teammates in to chat with your agents. And by the way, like in Slack, you can talk with other humans in it, too. Oh, and seems like Claude is allowing you to use your subscription so

[0:29] you don't have to pay extra for it. Anyway, this seems complicated, but it's all beautifully elegant. And what else did you expect from Jack Dorsey? We're going to see it all step by step. Let's get into it. Presented by Zapier, the AI automation company. Vinnie, what are we looking at here? Okay, so um we're just in Buzz right now

[0:48] and um let's just start with the basics. It's like a Slack-l like interface. It looks a lot like Slack. And you've just got a place where your teammates, your human teammates, and your agents are um first class citizens. They're equals. They're one and the same. So they're not some integration that you tack on. They're just in there and they can do all the things that a human can do. So um that includes creating channels. That includes uh working on uh coding coding stuff, working on projects with you. Um, all types of stuff. Anything you can kind of think of that an agent that you're used to, any other agent can do,

[1:23] it can do that here. I see Honey. I see Hottown, which I know is your Twitter name. So, that's you in here. Who what is this? Who's human and who's not? Um, so when you start up uh you will get some default agents and uh they're called Fizz, Honey, and Bumble. Um, I didn't do much uh tweaking of these agents, but they're in there in the chat with you and they have a system and this is essentially like like an openclaw Hermes agent. It just comes

[1:50] packaged in with this chat experience. Right. Exactly. The way I would describe it is like Slack with OpenClaw and as well as a um agent coding agent manager or an agent manager on top. So you get like these three things roll rolled in and there are even some features there like that are kind of bubbling under the surface that they're hiding that will

[2:14] take it even broader perhaps. Um so there's some exciting possibilities with it but already the fact that you have these agents in your chat and let me just jump into my favorite thing about them which is they're all there. If you've ever um you know kind of dove into even chat GPT or something you'll see agent instructions. It's like a system prompt. It's their general kind of role or their general instructions of what you want them to do. This can get very specific. It can be very general. Uh these agents, it's quite general. Um you don't need to worry too much about that. Let's just say uh one is more focusing on coding,

[2:49] one is more focusing on um copy and editing and brainstorming. Got it. In this case, this is Fizz. Fizz is an agentic maker. That's the developer. He's upbeat, practical, and decisive. Helps users plan, create, solve problems, and finish work. What's the agent that is? What's the other agent that you have? Um, so these are just default. They come like when you open up uh start uh Buzz to to the first time, they're in there. And so I didn't change the the agent instructions. So this one is a warm, thoughtful communicator helping to write, organize ideas, and brainstorm. So I like that. And and I think that's a good small separation

[3:25] there. Um, but the first thing I did was look under the the surface and see that they are layers on top of agent harnesses or your coding tools or your agents that you're used to. So under there's you can choose cloud code. If you have a cloud code subscription, you can use cloud code. Let me pause on let me pause on that for a minute because that's really important with Hermes agent with openclaw. We have to we we have to pay for cloud code using their API per use. We're not allowed to bring our subscriptions anymore. Here you can bring your

[3:56] subscription in. Yep. And some people Exactly. I'm on like a max plan and so basically the agent is a higher level uh layer on top of this. Exactly. And that's what I think is so cool because I can go there and pin it to a model. Let's say I want my uh my maker to be like the really powerful frontier model and I want my brainstormer or my my writer or my like organizer agent to use sonnet or something that uses less tokens that's simpler tasks. So that's kind of what I did. I pinned them to fable and sonnet and um like really to have a Hermes agent

[4:33] that's that's powered by Fable is super expensive here. You just have it on your subscription. Okay, this is exciting. Um they also recently allowed us to bring our Hermes agent, our openclaw and other agents in you haven't done that but that's one of the features that they've added right? Um yes and I can show it here. So you

[4:52] have agent runtimes. Um buzz agent we can get into later. There's a really cool kind of hidden feature there. Why don't you tell me now what's a buzz agent? Um so they have this option where you can um share compute. So let me go to real quick under here compute. So you can say share this machine. You click that option there. It'll download a local uh open source model that fits well on on my Apple M1 Pro. And um I'm then able to create an agent here and select buzz agent right here. You see

[5:27] agent runtimes. Select buzz agent. Select that local model. And then people in my community, humans can tag that agent and use the compute that's running on my machine. So that's a very powerful feature if you expand it out and think about more powerful machines, communities, the future, things like that. Okay, let's go back to what we were just

[5:48] talking about. Uh I have cloud code. It uh recognized that and installed the adapter. Um but there are all sorts of ones. You can add codecs, you can add cursor, grock, kimmy, uh open claw, open code. Okay. And that's where it gets exciting for me at least. Um, and I think this is the direction they're trying to go in is that uh, you know, these things move so fast and these models are getting so good that maybe everyone saw probably Kimmy code K3 is on par with with uh,

[6:17] Fable or Opus, you know, Opus 5. You want to try that? You want to use that? Boom. Switch it over. Your agent, the same agent all of a sudden can now be using Kimmy K3 under the hood, but it gets all the same context. It gets all the same workflows that you've built in Buzz. Nothing changes in Buzz. It's like you didn't even know the underlying harness, the what's underlying the models under it has even changed. So,

[6:40] that's really cool. By the way, when you're using new agents like this when they're experimental, you really want to protect your data, which is why I use our sponsor, Zapier, and specifically Zapier MCP. With Zapier MCP, I could give my agents access to my email, my notion, etc., But I restrict it to only specific things. So I can say you can read my email. You can draft my email but you can't send it. You can get into notion but you can't do certain

[7:06] things in notion. I love Zapier MCP. It's available free for trial and you'll love it and it'll give you more access to more power with the protection you need. Go to zapier.com/mcp if you're listening. You'll like it. Okay. Can I see what you built with it? So now wait actually you were telling me what's So now I see that Honey is not a

[7:23] human. It's an agent managed by you. Fizz is an agent. Do you also bring your team members in here? I think you brought one team member in. Sure. Yeah. Yeah. We we just, you know, messed around. Um we were trying to see, so Matia is a a co-orker of mine and uh we were just trying to see how easy it was. We put shared boy. We were trying to share an agent. Um because that is um apparently possible. We weren't able to get it working. So that's one thing to mention. This is like developer

[7:49] preview. This is really early software. There might be some hiccups along the way, but um I think the idea is interesting enough to to deal with those hiccups and stick with it at the moment. Um so you you're supposed to be able to share be able to share an agent. So that would mean an agent running on my

[8:07] computer, my cloud code subscription. You'd be able to ping it, tag it in the chat, and get it to do tasks while it runs on my computer. That's pretty powerful idea. Yeah. Can I see what you've built now? Sure. Um, so in this channel, and what I like to do, um, it's kind of messy right now, but, um, you I create a kind of a channel for each idea or feature or thing I'm working on. Um, you'll go up here to to add an agent, create agent or add an agent, and go ahead. I added Fizz and I told Fizz right here, spin up a simple CRM app using Wasp, the Wasp

[8:43] fullstack framework, and deploy it to Railway for me. So, this is kind of like my preferred um uh stack for building apps and my preferred hosting provider. Railway is um a place where they'll put the code so that it's live online for you. And um because I like that because it works so well with these kinds of tools. I specifically gave it those instructions. But like what's really cool is that this is one line, you know, I just jammed in some some text here. Just said do this for me. And let's take a look at what it actually um did. So, I sent that off and I went and did some other stuff

[9:19] and it said, "Picked up. Building a Wasp CRM now. Environment checked. It's installed. Your WASP is installed. Railway CLI is installed. Um, railway off is expired. So, like go ahead and log in real quick." I said, "Okay, I logged in." Mhm. And then it just went off. Um, and it ran it built the thing, ran tests, deployed and verified that it was live,

[9:42] gave me the the URL, took screenshots. So, it even tested out in the browser for me. It went in the browser, went to this uh website, this app that it just built and u made sure it worked end to end, logged in itself and stuff like that. Right. Great. You see here, Vinnie Buzz CRM. So, don't go um you know messing with my uh leads there. Um uh no, it's just a it's just a an example app. It went in, logged in, tried it out, and yeah, it's live and I can just go ahead

[10:13] and go to it. There it is. Boom. So, one of the things that was a showstopper for me was that it's all desktop based, and a lot of these tools are desktop based, and I don't want to have to keep coming back to my desktop to work. You said they've got a mobile version now. Can I see that?

[10:28] Yeah, sure. Um so, here. You want to bring it up? You just installed it yourself here. Um, yeah, there's your phone. You're mirroring it. Yeah. Let's see. We got too many things open here. Yeah. So, there it is right there. Can you see it? Uh-huh. I see it. Okay. So, this is what it looks like on the phone. I see all

[10:45] the channels. So, the same What are we in channel CRM now? There we go. And I can look at the um the threads and Oh, that's fantastic. So, now you can code remotely. You can work with your agent and ask for it to do things remotely. It does threading which you can't get in Telegram. Um, and it works with subscription clawed. This is so far amazing. Now, one of the problems with just chatting and asking for a project is that things just get lost. And so, they added projects into this. What is the project section of

[11:18] Buzz? Um, so if you've ever used Git or GitHub, this will look very familiar. You've got repositories and um, these are your projects essentially. So um these are your coding projects and uh the interesting thing that is kind of hiding under the surface here that wasn't mentioned on the initial launch is that um they're also offering that the same server uh that your buzz community is hosted on. There is also a um remote git hosting solution. So when you work on code locally on your computer or your agent does, it will actually create a copy or multiple copies in a specific buzz folder on your computer and it will work there so that

[12:05] it doesn't mess with any of your stuff, doesn't mess up your any of your code. And then if you want to, you can configure it with either GitHub, um you can configure it uh with linear issues, things like that. connect it to the things you're used to using maybe or you can push the code to your um your actual community, the server that your community lives on. So, you're getting a sense there that they're really focusing on you being able to control how your agents work, what models it use, as well as where your code goes, who owns your code, who's able to to look at your code

[12:40] or not. Right? Okay. So, it's a very interesting concept. And um besides that, you just get this kind of uh view into what's going on with your projects. So like if I go to overview, uh I was working on a a blog. So I wanted to to build a um Substack alternative so people could pay for you know my my premium blog post or

[13:02] something like that. Okay. So um you know it it created the repository here and um you know just nothing going on really there. So nothing really important but created the repository. Here's the repositories that I have. The Buzz CRM that we just created. Uh the blog. Let's take a look at the the Buzz CRM.

[13:20] Sure. So is that on GitHub? Uh this is this one is on GitHub, I believe. Okay. Um let me see where. Yeah, I pushed it to GitHub. So when if you were to click that link, it would take that's what pushes it to GitHub. And you have to whenever No, that's just showing you that it's remote, right? that the remote repository is hosted on GitHub, not on

[13:38] my uh community relay. Okay, I really like the layout and design of that. So, essentially, you build something, it sits on your computer, you tell it push to GitHub, it's now on GitHub, and it's updated. You want to hit that one box, and you go take a look at it on GitHub, and you see

[13:52] it. Anyone on your team can contribute. I could already see people, and they call it people, but I know that one of them is an agent. Um, but I like how agents are treated like people in there. This all this all makes tremendous sense. Um, to download it, you go to GitHub or to their website, buzz.xyz to get it. And you mentioned Wasp earlier when you were building the CRM. What's Wasp and why did you use that to build

[14:16] your CRM? So, um, I work I'm a founding uh, developer relations engineer for Wasp and it's a full stack web app framework. So, um, if that maybe that doesn't say mean much to you, but it basically, you know, when you're building an app, there's a bunch of different parts and they're all kind of disjointed. You have to connect them

[14:35] together yourself or your agents do. Wasp kind of puts that all together and makes it very easy for human and agents to collaborate on web apps. So it's a great um solution to build uh web apps with um agents because you get things like you know authentication uh one command deployments so these things usually take hundreds of lines of code to do and with Wasp it it's all glued together and it's all available to you or an agent in one command one CLI command or a couple lines of code so you're like freeing up your agent to do more important work than this boilerplate mundane coding tasks and

[15:16] things like that. One one last area we didn't look at, there's workflows top left um fourth from the top. What's workflow? So um you basically can um set up certain tasks or triggers and in in uh Buzz so I have one here. I built a um tweet dashboard. So, it's pulling it's an app that I built with Wasp and it pulls all the tweets from our team and um it makes you know takes some stats like the impressions, our engagement, our top 10 tweets every day. So, we have a bit of like a overview of like how we're doing on marketing, how we can improve our Twitter marketing and stuff

[15:54] like that. Great part about it is I get it every day in this um channel here. I can reply to it and then I can ask one of the agents like yo um you know what's what are some I asked here give me some topics to tweet about today something safe based on what I'm doing something in the middle and something wild and creative right so like I'm getting it to help me come up with tw uh tweet ideas things like that and um uh the workflow is basically you can get it to do stuff based on that when this comes in it triggers a a um workflow that tells me,

[16:31] hey, new tweet stats just landed and then I know, oh, cool. I can go and and um kind of start brainstorming with my agent on that. So, you can set up recurring tasks, triggers, and we can look at them for a second if you want, right? Sure. Let's take a look. Right. So, this is how you do it. Uh create a workflow. Um let's see. We've got let's just say in the general channel, um we could these are the things we can do. A trigger can be a message is posted. Let's say like

[16:58] Andrew, you're coming in to to help us. We're going to brainstorm a podcast um idea and whenever I get a message posted from you um because I know that you you know you like messages right away, I'm going to say, "Hey, hold on 15 minutes. I'll be with you in 15 minutes." So, hit that message posted. So whenever there's a message posted right now in that one channel so only if it says urgent for example or whatever you want to screen for you could then add a step what's the step that then happens where's the where's the action on the trigger right here you have a filter expression

[17:34] so I can filter for it gives you an example here contains text and let's say you're saying the the example I have is urgent if I use the word urgent then this triggers off it's not just if messages just use But let's let's click on add a step then and see what happens next. Right. So you contain the text uh it contains the text urgent and then I can

[17:56] add an action like send a DM. Got it. Or a reaction, you know, maybe just a reaction like uh like something like that. Anything like that. Or call a web hook to take real action outside of the app. Sure. Like we could say um what's a good example? We could say we could hook up the X API with an agent here um and say every time I just call uh every time the text contains this you like post that to X whatever we just talked about I'll add this this signifier post to X. So, after I'm done brainstorming with my agent on all the tweet stats I got um from from my app,

[18:39] I can be like, "Oh, that's a great tweet idea. Go ahead, tweet it." And there's a trigger for that. I like that. You just create it. The automation works. This is beautiful. I actually was prepared for this to be a lot more beta than it is. Have you had a lot of bugs, issues? Um, so I will say with workflows, I just updated to the newest version, but when I was trying workflows, yes, I was having problems getting them set up. Um, but uh they as you see it is working here. I just wanted like a a test this notification when new new tweet stats drop. Notify me, tag me so that I can

[19:14] start working on brainstorming with you. Um, the other thing that's cool is you saw it's a bit technical in there. You might need to know these expressions. You can just tell your agent like fizz set up a workflow that does right and it will do it. Yeah. Oh, that's great. Yeah. You can delegate you can tell um another thing I do in in this where it's like you can delegate multiple like a

[19:39] complex task to multiple agents at once. So I can say fizz I know fizz is fable right? Fable is a very powerful model. Um, I'll say honey, you're sonnet, so you're a lower level model. You do this very lower level task of maybe like sorting through and or just fetching data or sifting through data. Fable, I want you to analyze it and pull out the

[20:01] valuable insights from that. Then hand it back off to Sonnet and have Sonnet create a small table in Excel for me of it. And this is great because you have multi-step multi- aent workflows that are taking advantage of each each one's strengths and you know and weaknesses. So using less tokens for example when you don't need the power of fable you pass things off to a lower

[20:27] level model you know things like that. You know what I think that they're making a mistake or maybe we're making a mistake by saying this is the Slack killer Slack replacement. That's probably the long-term vision and I think it's a good one and I'd like to see it because I don't like Slack at all. I think they should just talk about it as the Telegram replacement. We're all using Telegram to talk to our agents. Not because it's the best model, but it's because the easiest one to add agents into, but it doesn't have proper threading. It doesn't have lots of features like this. And you can't bring

[20:56] easily regular human beings into it. I know I can, but it's not in it's not meant for normies to do. This allows that. And by the way, maybe we start to shift more of our conversations with humans into it. But the first step I think should be to say this is this looks like the best way to communicate with your agent. I I'm excited that I got to see it. Does that seem like a

[21:14] fair analysis to you? That's exactly how I describe it. I say the Slack killer is almost like not doing it justice because you have so much agent orchestration. You have, you know, the GitHub replacement underneath. You have lots of other cool features like um, you know, being able to share agents and share compute even between other people in your community. So like you could really go very far with some of these more technical features and um it's almost like a global context window, right? So we're used we're before we we're doing stuff between lots of different apps, maybe copy and pasting stuff from Twitter, pasting it

[21:52] over into chat GPT, bringing it back into cloud code, whatever to get a task done. here everything is centered and you can build tools that then integrate with it and keep everything in one central place so that the agents and you build this like crazy knowledge set and context window, you know. All right, if you're listening to this and you want to see what you can have your agents build, I've got a video for you right here and I'll see you in that

[22:16] video next.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt