---
title: 'Jack Dorsey''s Buzz: Does It Beat Hermes Agent?'
source: 'https://youtube.com/watch?v=N_S61LTFsOc'
video_id: 'N_S61LTFsOc'
date: 2026-07-29
duration_sec: 1750
---

# Jack Dorsey's Buzz: Does It Beat Hermes Agent?

> Source: [Jack Dorsey's Buzz: Does It Beat Hermes Agent?](https://youtube.com/watch?v=N_S61LTFsOc)

## Summary

This video demonstrates how to set up Buzz, a free app from Jack Dorsey's Block, which acts like Slack for AI agents, allowing them to collaborate automatically. The presenter shows how to connect various agent models, create custom agents, and orchestrate teams to automate tasks like content creation and image generation. The tutorial highlights the ease of use and integration with existing tools like Claude Code and Obsidian, making it a practical solution for workflow automation.

### Key Points

- **Introducing Buzz** [00:03] — Buzz is a free app that turns AI agents into a team of employees that communicate and work together in a group chat interface similar to Slack.
- **Setup with Goose CLI** [01:01] — Buzz is tested alongside Goose CLI, both from Block. Installation involves signing in, creating an account, and installing agents like Claude Code and Codeex.
- **Creating Default and Custom Agents** [02:54] — Default agents are available, and custom agents can be created for specific tasks. The process includes naming, describing, and assigning an emoji.
- **Using Obsidian Vault for Personalization** [04:44] — The presenter uses an Obsidian vault as a 'second brain' to store memories and context, which agents can reference for personalized suggestions and automation ideas.
- **Image Generation Agent** [06:30] — A custom 'image designer' agent is created to generate image variations based on provided examples and logos, showcasing the flexibility of custom agents.
- **SEO Agent and Content Publishing** [08:12] — An SEO agent is set up to perform keyword research and publish blog posts to WordPress. It leverages existing tools and data from Claude Code.
- **Agent Teams for Orchestration** [12:25] — Multiple agents can be grouped into teams (e.g., 'marketing team') to work together on tasks. The presenter demonstrates how agents collaborate autonomously.
- **Quality Controller Agent** [15:17] — A quality controller agent is added to review and provide feedback on the work of other agents, improving output quality before publication.
- **Inviting Humans and Sharing Access** [20:12] — Human team members can be invited to channels, bringing their own agents. Permissions can be restricted to control access to specific agents.
- **Key Benefits of Buzz** [23:46] — Buzz is free, open-source, intuitive, integrates with Claude Code and Codeex, and allows easy management of automations. The user only pays for the model subscriptions they bring.

### Conclusion

Buzz provides an intuitive and powerful platform for orchestrating multiple AI agents to automate complex workflows, from content creation to quality control. Its similarity to Slack reduces the learning curve, making it accessible for teams to delegate tasks and improve productivity.

## Transcript

free app called Buzz and it turns your AI agents into a full team of employees that talk to each other and work together all inside one simple group chat that looks like Slack. It's free and today I'm going to show you exactly
how to set it up. Imagine typing one message and watching an entire team spring into action. One agent writes your blog post, another designs your images, and a third one checks all the work and fixes all the mistakes all by
themselves without you lifting a finger. I built this whole system in about 20 minutes. And I'll show you every step today. Plus, this one agent you absolutely have to add to your team that makes everything 10 times better. And
me. By the end of this video, you'll have your own AI team working for you Today we're going to be testing out Buzz, which is a new app from Jack Dorsey, and we're going to be testing it out with Goose CLI as well. So that's a
a model you can plug into Buzz. Now, if you're not familiar with Buzz or how it works, basically it's like Slack for AI agents. It seems a lot more intuitive and easier to use than something like Hermes agent. So I'm just going to
install it right here and then we'll start testing out. Also, you can get it to install. So, I'm just going to go with the normal app, which I think is to sign in here. Just going to get my key in the background. I actually just
create a new account. And now we're going to install each CLI as you can see going to install each CLI as you can see here. So, Claude Code, Codeex, and Goose. Now, probably need to install Goose in the background from GitHub. So,
Code, we've already got Codeex. We don't have the CLI for Goose set up yet. So, I'm just going to grab that from GitHub. And we can actually get the app for that, too. So, we'll open this up. This
that, too. So, we'll open this up. This is Goose that we're going to connect to Buzz in a second. Both of these are from Block, which is Jack Dorsey's company. So, I'm just going to connect to a provider. So, once you've got Buzz set
up, like you can see right here, you can go to your settings. So go to your profile on the bottom left settings and then go to agents. This is once you've created a community, logged in, etc. And so once you've done that, you can select
so once you've done that, you can select which agent you want to use. I did test out with Hermes agent and also for example goose with free models doesn't seem to work so well. But if I select for example claude code here and we just
go with the default model. Save the defaults now. That should work pretty well. So let's just tag one of our agents in and say, "Hey, are you online?" And now you can see it's beginning to reply. Perfect.
All right. So we've got our agent set up. These are the default ones. If you wanted to create a new custom one, you can go to new agent over here. And then can go to new agent over here. And then what you can start doing is creating
new custom automations based on whenever you work on day-to-day. So let's say for example we wanted to have a channel that's just like I want to be able to ask my second brain for new ideas, right? And we're going to click on
create a channel. Second brain looks up my memory and gives me ideas. So this my second brain. It understands everything inside my obsidian vault. And
this is my obsidian vault over here which you can see. And so it's my second brain. So I can use that to just come up with new ideas and that sort of thing. Okay. So now what we can do is add an agent to that channel. So we'll add Fizz
there. And if we tag in fears and we say, "Based on what you know about me from my Obsidian vault, give me some ideas on what I could automate next." If
you're not familiar with Obsidian, basically it's a free app that you can use to store all your memories about you, right? So you can see, for example, with Obsidian, we have all of our memories. All my agents automatically
update this stuff and they know about my projects, what I'm working on, who I'm working with, what sort of tools I use, the people I work with, um, even my knows everything about me and we can use that to look through our Obsidian vault
auto. If you pull in, for example, codeex into here, you can use that to codeex into here, you can use that to basically have personalized ideas on we'll get some ideas first and then we'll start building out the next
minute as well is we'll build out some teams. So, some agent teams. So, if we back to us. As you know, we said based on what you know about me, I just had to
reconfigure it quickly cuz I got Fable 5 to try and help me um before on setting up Goose and it totally messed everything up. So, don't do that. Um, the vault. Here's your automation ideas. So, for example,
let's come up with a bunch of ideas that we could come up with like a a QC a quality control pipeline, some ideas for like SEO indexing. You've got, for example, ideas on on email and that sort of thing. And then also a
dashboard that we could build. So, all sorts of cool ideas right here, right? Including like a a follow-up for AI profit boardroom members that we could set up as a funnel. So, you can get loads of automation ideas like this. And
this is claude code connected to your setup with buzz. Right? So if we now go to a new channel, we can create a new channel here. Um we could call this for
example like image generation. Create a channel like so. And then we can add an agent inside there. And what we can actually do is create a new agent specifically designed for this. So I'm going to call this image designer. And
then we have to describe what the agent should do. So I'm going to say design should do. So I'm going to say design images based on the image I give you and come up with multiple different versions for the image idea. So we've basically
got like a custom trained agent inside here. And then we can give it like a an here. And then we can give it like a an emoji based on what it does dayto-day or we can actually create an image as well. And we just have to set the default
select cl code here and then we'll give it a nice smiley face emoji. So we'll create that agent and now that agent is set up to start generating images. All right. So we've added in the image here. And then what we can do is we could for
example like let's say we are creating an image around the video topic from today just as an example custom workflow. So what I'm going to do is grab an example. Then we're going to tag in the image designer here. And I'm also
going to give it an example image that I really like like this. So we've got the logo for the image we want to create and an example. And then we're going to say to the image designer, see the example attached of an image I like and also a
new logo to insert inside that image. Can you create a few variations? And usually use for image generation as well. So let's make sure that works. And you can see every time you post and ask it to do something, it will look at it
and then it will generate a comment. And then you can see here that you can can see what it's doing in the background there, too. Now, if you hover activity, you can see what it's working on and what it's doing in the
background. So you can see the tools that it's using, what it's doing step by step. So if it's taking a while to reply, you can just have that working in timer here. So you can see how long it takes to respond.
background, you can come back to it later and we're good to go on that. Now if we create a second channel here, uh we could for example come up with a new automation that we work on dayto-day. So let's call this SEO blogs. This is for
creating content. We'll create a channel here. And then we're going to create a new agent for this as well. So we're going to call this SEO agent. And this is using Claude code as well. Your job is to use your SEO skill to come up with
keywords and content ideas and also publish content straight to WordPress for it. I'm just going to add an icon for that. And we'll use like a chart emoji, I think, for this one. There we go. We'll save that. Using my Obsidian
Vault, can you find some new keywords to create content around? Ideally, keywords around Hermes agent. Does Buzz self-improve like Hermes? That is a good question. Well, I guess it's using clawed code, right? And clawed code is
always taking notes and then improving its memory based on how you prefer to do its memory based on how you prefer to do stuff. So, it's basically loaded in with the context from clawed code. So, if you're wondering, okay, does Buzz
which agent you're using, but if you're using something like Claude, well, then other thing that you could do is when you actually add in the agent. So if we have a look at SEO agent for example, we can actually say based on your feedback
from me, self-improve and analyze your skills so that you get better every time skills so that you get better every time you are used. So you can see this is now replied to us with a top pick. For example, Hermes workspace. So if we
click on the reply here, it's actually come up with a bunch of ideas. So it says Google search console pulled fresh Hermes agent keyword picks is the biggest one first. So it's actually looked at the keywords we could rank for
based on the Google search console data that I've already given to Claude in the past before we even started using Buzz. So the great thing about this is when context and all the training and all the improvements that we've given to Claude
in the past. So we can see for example, we've got a bunch of keywords. Now, already have. So, I'm actually going to say, can you come up with new ideas, not
existing keywords already ranking for? And now you can see it's actually come what we wanted to rank for. So, if I have a look at these keywords, some of them are good, some of them like I wouldn't use. Let's have a look here.
So, I think this could be a good one. Install Hermes agent on Hostinger. So, I'm going to tag in. It's important that you tag in the agent as well. If you need to know how to do that, you just type in at and then the agent. So,
we can put SEO agent install Hermes agent on Hostesser as a keyword. Can you agent on Hostesser as a keyword. Can you publish the article to julianoldie.com? Let's see if it can do that. And then it's actually using it skills for SEO to
back to in a second. It'll probably take a few minutes to start creating the content and then publishing it to the website. But you can see we're gradually building out all the automations that we want to use for this. And then the other
it's fairly intuitive. Like most people are familiar with Slack and it's designed to be like really user friendly. But also we can add and invite other people to this as well. Now also we were generating that image a second
we were generating that image a second ago. So we gave it the logo and we gave it an example thumbnail that we like. Right now, what it's actually done from there is it's created these images that are actually pretty nice. Like, it's
are actually pretty nice. Like, it's done a good job on that. So, you can see some examples of the type of style that is used. So, it's basically taken this example plus this and then use that to
generate this and this, right? And the good thing about this is like you can want. It's all stored inside your workspace. And we've already set up two automations in the space of what like 10 to 20 minutes so far it's taken me. So
we've already got for example the SEO blog set up here which is working on in the background. And then we have the image generation tool here as well. Now what happens if you wanted a team of agents working together so that you can
orchestrate all of them at once. Let's try that. So, we're going to go to agent teams over here and we'll create a new team. So, we'll create this team and we'll add the image designer and the SEO agent and then we can add optional
instructions applied to every team member and we can give a description for member and we can give a description for this team. So, I'm going to say SEO blog creator and I guess the benefit of this is you have like customtrained experts
working together as a team to automate and build anything. So if we have a look at for example at this we could call this the marketing team and then we could say inside here work together with your specialist skills so that you can
basically automate and work together as a team to automate my marketing for me. So we'll create the team like so. Now one thing I want to be like 100% transparent with you is that this sometimes times out I think. So if we
the image generation workflow we actually asked it to do something, right? So, you can see we gave it the prompt there and I think it timed out because it didn't reply to us and we actually have to say
actually replied to us and gave the outputs and I'm assuming it's the same with this like so sometimes if it's a big task or if it takes a few minutes it seems to like time out and take a while to reply. So, just something to bear in
mind right there might depend what model you're using as well. So, for example, like something like Opus 5 might be really slow to respond and that's why there's an issue. All right. So, we now have the blog live, right? So, just to
recap on this workflow, we said using my Obsidian Vault, can you find some new keywords to go for? And then we said, can you go for completely new ideas that looked at my Google Search Console data. I didn't have to plug in any new APIs
in. and it already took the APIs that it's got installed into cloud code on my desktop app. And then over here you can see that it's now created the blog. So if we go over here, we have the blog fully automated from Buzz. So didn't
have to log into WordPress. It targeted the keyword that we wanted. It's even, for example, pulled in some case studies. It's formatted it nicely. And have uh set up ready to go with minimum
uh set up ready to go with minimum hassle. So if you look at Buzz, I think that the main benefit of this is just very very easy to orchestrate your AI taught your agents in the past. So when you're using this like you don't have to
plug in new API gates, you pull them in from Claude. you can pull in the context whether that's inside Claude or Codeex or inside Obsidian, whatever you prefer to use. And it's just pretty simple and easy to use, right? So, it's it's a
great place. And then also, you've got everything stored in here, so you can come back to it for later, which is really useful as well. Be crazy. So, create a quality controller agent here to check the work. So your job is to
check the work of your team to make sure it's actually good. And if it's not good enough, then make sure that you give feedback to the AI agents to improve it. So we'll create that agent right here. Give it a little emoji. There we go.
Save the changes. And then if we go over to, for example, our team that we created earlier, we'll add in the quality controller as well. Save that now. And if we go over to a new channel here and we'll just call this like
marketing or something like that just to test it out. Marketing team. Create the channel. Then we're going to add our agents. So we'll select the marketing team. So a good thing about this is you don't have to select individual agents.
here, you can add the team inside there. So add that team in and then we're going to say okay marketing team. This is the other benefit as well is like you can tag the whole team in instead of selecting the individual agents. So
click each one of these, we can just select marketing team. And then by the way, if you're wondering, I'm using Whisper Flow to speak to the AI. So from here, we're going to say create a blog and an image about Hermes agent and
quality controller. Just check the work and give them feedback if it's not good enough. So let's see how that goes. So you can see all three have marked it and all three are beginning to comment and then we can see them working below here.
So we can see the activity of each agent what it's doing how it's working etc. So if we have a look here it's starting the SEO work which is the keyword research and then the WordPress draft and then we have the image designer and that's
beginning to start coming up with images as well. Now we have a reply as well and it's asking for a reference image so we'll just add that in. So, we've added a reference image here for the image designer. And then you can see the
yet. I think that's because it needs to wait for the image designer and the SEO agent to finish off. And then once that's done, it can come back to us and uh give feedback to the other agents and communicate with them directly. Now, in
the meantime, you might be saying, okay, what can you do with Buzz? Like what are you a few already, but for example, you can ask the project a question and get an answer. And the other good thing about this is agent search 6 months of
history and post the threats, right? So you can store everything that you've done directly. You can also let an agent triage a bug without giving it access to own keys, their own channel memberships, their own audit trail. So they can
history, but they're only limited to what you silo away for each agent. So not each so every agent has its own permissions and can only look at agent for each project like we were doing for example with image generation
and SEO marketing and everything else. Now you can see here the the SEO agent has come up with the keyword research then drafted the blog post given us the
details here and it's also tagged in the image designer. Right? So now the agents are working autonomously without me. So they're just going off and working as a team in kind of like a group chat to come up with new ideas and they they
image designer is great at image design. The SEO agent is great at coming up with keywords and blog posts. And then finally the quality controller can QC
good. So they're really working as a team and you could build like a massive be super useful. So now you can see the image designer is coming up with the image designer is coming up with the image first and really you can let an
agent run the workspace as well right. So for example if we look at this system here the SEO agent is the one taking charge coming up the keywords everything designer when it needs it. So there's very little orchestration from me
with everything along the way. Now I'm going to tag in the quality controller and I'm going to say okay check this post for me and see if it's good. If not
post for me and see if it's good. If not give feedback to SEO agent. Let's see if that works. By the way this is like very experimental because you know at this point Buzz is still in the early stage. It's only launched about 7 days ago. So,
thinking, why have I never heard of Buzz before? That's why. But, as you can see, it's very useful and it actually has like real use cases, which is great. your settings, just whilst we're waiting for that to load over here inside the
marketing team chat, if we go to the settings over here, you can also invite humans to your channel and they can bring their own agents as well. So, we can invite people to the community as you can see right here.
And then we can give them this link and invite them directly, right? So if you wanted to add some friends to a community or for example, you had a team pretty cool as well. And you can see how quickly this could grow. And also you
can limit what people have access to so that they don't just use the agents that you have access to locally, right? You could, for example, make sure that they right? So, for example, let's say you have an editor for videos. Well, they
could use the image generator instead of, for example, having a a designer and they could use the agent instead if you've given it a really good workflow. Now, if we have a look here as well, you can see that the quality controller has
given feedback to the SEO agent and it's reviewed the work and then tagged me, but also tagged the agent, right? So the agents are just going back and forth which is pretty nice as well. So you can see here the quality controller reviewed
see here the quality controller reviewed the draft verdict content is solid but not ship ready. One real fix is needed. And so what it's actually found is that there's some bugs in terms of the grammar and also the word count was
wrong which is pretty crazy. So, it looked at the word count from the SEO agent and was like, mate, like this isn't 1,970 words as reported. This is 1,580. And then it said everything else checks
out. FAQ is good or vers new way, etc. That's all good. And so, it says SEO agent fix the capitalization instances and reping me. And then it says Julian Goldie blog copy is good. Once that's fixed, still holding on the image,
right? And then the SEO agent says I fixed all three. We're good to go. And the quality controller says confirmed. All seven instances are now fixed. Good to go. The blog is ship ready. So what we can do from here is say to the SEO
agent, okay, get this published now. And that should be able to run this. Now in the meantime, you might be wondering, okay, like how can I add my other agents have Hermes agent. Could you connect
have Hermes agent. Could you connect Hermes agent to a setup like this? Now, this would be super interesting. So, what I think you could actually do is you could get the GitHub from Buzz, get for example Fable 5 to install that or
for example Fable 5 to install that or Hermes agent and then ask it to also add Hermes agent and then ask it to also add Hermes agent as a option when you set up as a separate install. I don't know if that's going to work, but we can try it
out. So I'm going to go into Fable 5 inside Claude and say install this inside Claude and say install this separate instance but edit it so that we can use Hermes on the setup and we'll get Fable 5 working on that in the
background. So if we come back to this workflow now we've got our agents orchestrating each other which is really really cool and then I've said get this published to the SEO agent and it's published it as a full blog post right
here. Right. So, we've got a fully edited blog post, quality controlled, improved massively. It's got the CTAs on the page. It's nicely organized, nicely formatted, and that's fully published from just a simple back and forth
from just a simple back and forth message on Buzz, which is really, really cool. So, some people say, okay, you know, what do you use it for? You use it for whatever you are working on and you need to automate. Why would you use it?
Well, the benefits just to be very clear here. Number one, you can use clawed code with it and you can use codeex. It's free and open source as well. The only thing you would pay for is the model that you bring into it. So, for
example, if you bring in claude or you bring in for example codeex, those are subscriptions that you would add in. It's very intuitive and easy to set up as well. Much easier than setting up for example Hermes. And also you can add
your team very easily to the setup so that they can work with your agents. And also it's much easier to come back to and organize everything that you've automated. So for example, if you look at the workflows that we've created,
we've got history inside the chats of everything that we've done for each automation. So if we go back to Buzz here for example, we can look at the look at the previous images we've generated. It's very easy to just have a
look through and see what we're doing and see what's been done in the past. Also, it's really easy to just add a new agent, a new specialist agent, far it feels a lot more visual and a lot more customizable than something like
Hermes. Now, Hermes still has a place. I mean, you could probably use and we'll test this out, but Hermes could probably work with Buzz directly, which would be that working, that's going to be awesome. So you can see here the Fable 5
would always check that before you actually launch it. But it's going to take about 20 to 35 minutes to set up the workspace. So I think we can run Hermes with Buzz, but it's going to take like 20 to 35 minutes. So it might be
like a separate tutorial that I do. If people want to see it, just let me know in the comments. I'd be interested to create that as well and test it out. Now let's have a look what else we've got inside the settings here. So you can
also change the appearance and how it looks. You can edit your profile. You can change your notifications. And then there's a bunch of shortcuts here. So you can use all these different hotkeys to just switch between things. You can
you can see also we have codecs available here. We've got gro build, claude code, goose, and buzz agent plugged in directly. And we can see communities as well. So we can actually stop these if we want to. And we can
choose between our default harness plus change any environmental variables over there too. Inside this section, we can share this machine, right? Which means that other members can run their agents here as well. So the way this
works is like you can select a model and then it will use remote models for example, Gemma 4. I've not tested that out, but that'll be interesting. And features as well like workflows, projects, polls, form channels, agent
manage profiles. And one thing they're beginning to start working on is the mobile updates. You can see that right here. Basically, I I did test this available just yet, but you can connect Buzz on mobile to your desktop and then
you can control your agents inside Buzz from your mobile as well, which would be great. So, this is really in the early stages, but super good. Lots of great stuff inside there. Really easy to use, really intuitive, and you can see how
by day and just keep improving it over time. So, thanks so much for watching. That is my full tutorial on Buzz, how to use it, how to get goose CLI set up as
there. If you're wondering how that works step by step, so when you're using Goose, make sure you've downloaded it, right? So you've got goose open over here and then you can go to your settings
go to agents and then select goose and then if you want to use goose as your default you can select goose inside the default select your provider for that you can actually add a custom provider as well
that's how you can switch between different models and providers depending much for watching. If you want to get more training like this, feel free to in the comments description or just go to the profitroom.com. This is my AI
learn, grow, and scale with AI automation. We've got over 200 pages of testimonials and wins from people inside the community using it. So, a lot of great place full of positivity, great people to meet as well inside there. And
where we can share and we can learn and we can grow together, which is awesome post questions, get help and support in real time, and I personally answer the classroom, you can also get access to all of our best trainings. So, if you're
a complete beginner, you can go from beginner to expert inside this section. If you want to get our uh courses for each tool, you can get that over here. And then also, we have our agent operating system inside this section.
And we add new master classes and tutorials based on what's just dropped the map, you can meet people in your local area who are building out with AI agents like you. And that's all
description or just go to the aiprofitworm.com aiprofitworm.com to get access. Thanks for watching.