19:25Alex Finn · Tutorial
100 Hours of Hermes Agent Lessons in 19 Minutes
Nine habits from months of daily use with an AI agent, distilled into a 19-minute checklist.
July 8thThe argument in one line.
Buzz is a chat app that makes AI agents first-class citizens instead of bolted-on plugins, so shared context, agent-to-agent debate, and pooled compute all become possible in one workspace.
Read if. Skip if.
READ IF YOU ARE…
- You lead or work on a small team that uses multiple AI coding agents (Claude Code, Codex, custom agents) and wants them sharing context instead of working in silos.
- You're curious about self-hosting local AI models and want an easy way to pool spare compute across a group of people.
- You want AI agents to review or debate each other's output before you act on it, instead of trusting a single model's first answer.
SKIP IF…
- You're a solo builder doing hardcore vibe coding by yourself — the video's own creator still recommends Codex for that use case.
- You just want a simpler Slack replacement for human-only team chat; Buzz's value is specifically the AI-agent layer.
The full version, fast.
Buzz is a free, open-source chat app built around AI agents instead of humans, letting Claude Code, Codex, and local models like Gemma live in the same channels as teammates and repositories. Every agent automatically gets the full channel history, other agents' past conversations, and project context, so agents can debate each other natively — asking two models to argue out a decision instead of trusting one answer. Repositories are built in too, so agents write and commit code directly inside channels rather than through separate GitHub tooling. The standout feature is pooled compute: community members share idle local hardware so everyone can run large local models for free. Best fit for teams; solo builders should still lean on Codex.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:37
01 · Cold open
Alex Finn calls Buzz revolutionary and previews that the video covers its differentiators, use cases, and features he says he's never seen in any app before.
00:37 – 06:34
02 · Product walkthrough
Tour of the Buzz interface: channels holding both humans and AI agents (his OpenClaw 'Henry', a 'Hermes' agent, and a local 'Billy' agent running Gemma 4), adversarial agent-to-agent debate used to draft a newsletter, and built-in repositories that let agents commit code directly inside channels.
06:34 – 13:22
03 · What makes it different
Recaps five differentiators on a title slide — centralized context, easy adversarial agents, traceability via searchable thinking logs, open source and free, and pooled compute — then covers when to use Buzz versus Codex for solo work.
13:22 – 18:04
04 · Shared / pooled compute
Explains Buzz's shared-compute feature: connecting personal hardware (three Mac Studios, several DGX Sparks, Mac Minis, an AMD Halo) so a whole community can run large local models like Kimi K3 for free, then announces he'll open this to his paid Vibe Coding Academy community.
18:04 – 19:04
05 · Recommendation and sign-off
Says everyone should use Buzz and teams get the most value from it, but solo vibe coders should still lean on Codex; closes with thanks and a sign-off.
Lines worth screenshotting.
- Buzz treats AI agents as first-class citizens in the chat app itself, instead of bolting them on through plugins the way Slack and Discord do.
- Every agent in a Buzz channel automatically gets the full conversation history, other agents' past conversations, and the linked repository's context with no manual setup.
- Agents can debate each other natively inside a Buzz thread, such as asking two agents to argue out what a newsletter should cover before either responds to the human.
- Buzz builds repositories directly into channels, so agents can write and commit code without a separate GitHub integration or MCP server.
- Buzz's search indexes agents' internal thinking logs, not just messages, so you can search for what an agent was reasoning about, not only what it said.
- Buzz automatically detects a user's local compute and lets them spin up local models like Gemma 4 in a few clicks, with no separate tooling like LM Studio required.
- Members of a Buzz community can share idle local hardware so everyone in the group can run large local models, like a roughly 500GB version of Kimi K3, for free.
- One creator estimated he has close to two and a half terabytes of combined local compute across three Mac Studios, several DGX Sparks, and Mac Minis that mostly sits idle.
- Buzz is open source and free, positioned directly against the $10-a-month cost of Slack.
- The creator recommends Buzz for any team-based building, but still prefers Codex for solo, single-player vibe coding.
Why unifying agents, code, and context in one place changes how AI performs
WHAT TO LEARN
Buzz's real lesson isn't the app itself, it's that AI agents perform noticeably better once they share full context and can check each other's work instead of operating in isolated silos.
- Fragmented AI context is a real performance tax: an agent that can't see your other agents' past conversations tends to give worse answers than one with full shared history.
- Letting two AI agents debate a task before either responds to you can produce a stronger final answer than trusting a single model's first response.
- Cross-checking one AI agent's output with a second, different agent is a cheap way to catch mistakes in code or research before you act on it.
- Most people are sitting on idle local compute; pooling it with a small group can unlock free access to large local models that would be too expensive to run alone.
- A tool's category label undersells its actual value: calling Buzz 'Slack for AI agents' misses the point, which is centralizing context so every agent works from the same information.
- Solo builders and small teams get different value from the same tool. Teams gain the most from multi-agent, multiplayer workflows, while solo work may still move faster with one dedicated coding agent.
Terms worth knowing.
- Buzz
- A free, open-source chat app from Jack Dorsey that puts AI agents, code repositories, and human teammates in the same channels, treating agents as native participants rather than bolted-on plugins.
- Adversarial agents
- Two or more AI agents debating or critiquing each other's output inside the same conversation, used to get a stronger result than trusting a single model's answer.
- Pooled / shared compute
- A Buzz feature where community members connect their own local hardware to a shared pool so anyone in the group can run large local AI models for free.
- Gemma 4
- A local AI model shown running on a user's own hardware inside Buzz, spun up automatically once Buzz detects available local compute.
- Kimi K3
- A large open-weight AI model, roughly 500GB in some versions, used as an example of what pooled local compute in Buzz could run for a whole community.
- DGX Spark
- A small AI-focused desktop computer from Nvidia; one of several machines the creator lists as part of his personal compute pool.
Things they pointed at.
02:29toolCodex
02:29toolClaude Code
15:26toolKimi K3
Lines you could clip.
00:06
“Jack Dorsey just released a new app called Buzz that I think is absolutely revolutionary and is a glimpse into the future.”
00:44
“This isn't just a Discord with AI agents just built in. This isn't just Slack. This is a major paradigm shift in how chat apps work.”
04:15
“No other app on planet Earth does this because every other app, AI agents are like second class citizens.”
06:46
“Context is everything. Having good context and having crap context is the difference between getting amazing performance out of AI and awful performance out of AI.”
17:07
“It's kinda like the Bitcoin compute in a way because that's kinda like the idea of crypto, that it's decentralized.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
Jack Dorsey just released a new app called Buzz that I think is absolutely revolutionary and is .
It is a new multi agent chat app that destroys Slack and Discord. But there's so much more to it that no one else in the entire Internet's talking about.
We're gonna talk about it here. In this video, I'm not only gonna show you what makes Buzz so special, but I'll also show you some use cases that will change the way you work and show you some new features I've never seen in any app ever before. By the end of this video, you're gonna be a Buzz master.
Let's lock in and get into it. So this is Buzz, and there is a lot of things going on here. The entire Internet, bad pun intended, has been buzzing the last week over this.
And for good reason, this isn't just a Discord with AI agents just built in.
This isn't just Slack. This is a major paradigm shift in how chat apps work. Quick walk through, and then I'll go into all the differentiators and use cases.
This gets really, really juicy. This is what Buzz looks like as of today. They are shipping fast.
They are shipping really fast. So by the time you see this video, there might be new things on the sidebar. There's also an excellent, excellent mobile app, so make sure to check that out too.
What makes Buzz so special is everything is in one place. Your AI agents, your code, your repository, your teammates.
And the reason why you want all of this in one place is because now context can be shared amongst all these things. Let me show you what we're looking at here. I'm in this channel right now.
Inside this channel, we have a whole bunch of things. One, we have me, and we also have a bunch of AI agents. We have my Open Claw Henry who is in here.
We have my Hermes agent Hermes. I just named my Hermes agent Hermes because I like that name. We also have some other things.
We also have an agent named Billy. Billy is a local agent.
This is a model, Gemma four, running on local compute. I was able to very easily spin up these local models and local agents inside of Buzz because Buzz detects all my local compute.
This is a really, really big part of Buzz, this local compute. We'll get into this later in the video, but this is something no other app in history has ever done. But let's talk about the AI agents first.
So for the first time, we truly have a way to bring all our agents in one place. Our OpenClaw, our Hermes, any other agents we have, Codex, Claude Code, I also have those in here as well.
They are all in one single place. And because they're all in one place, that gives us a few different advantages. One, and it's one thing you can see here, adversarial conversations are really, really easy to do now.
What I mean by adversarial conversations is a big unlock for using AI is when you can have different AI models and harnesses and agents debate each other.
Claude thinks one way. ChadGBT thinks another way.
Hermes thinks one way. OpenClaw thinks another way. Before you had a kind of jigsaw it all together and build skills and MCPs and plug ins so they can kinda communicate to each other in a really rough way.
You don't need to do that anymore. They all communicate to each other natively inside Buzz. There's no extra setup required.
So for instance, I send a weekly newsletter link down below. And when I typically write these newsletters, I go to my Open Claw. I say, hey.
What are you thinking for a newsletter this week? Then I go to my Hermes. I say, hey.
What are you thinking? Then sometimes I'll copy and paste them amongst each other. It's a whole lot of work.
Now I can do this. Hey, Henry and Hermes. Can you discuss amongst each other what would be the best newsletter for me to write?
They start a thread amongst each other where they literally start messaging each other back and forth. They're literally debating each other back and forth. I'm not even in the conversation.
I'm not saying anything. They are discussing back and forth what my newsletter should be for that week. Adversarial AI.
This yields a significantly better result than just asking one AI.
So adversarial AI natively supported. No other app on planet Earth does this because every other app, AI agents are like second class citizens.
Discord, Slack, you're building plug ins, which is basically just glue and tape to get your AI agents in there. With Buzz, they are native, so they can natively talk to each other just like they're human beings.
Speaking of human beings, human beings can work side by side in here as well. I'm actually gonna be opening up this community. In in Buzz, they're called communities.
I'm actually gonna open up this community very soon. Details on that later in this video as well. You can invite people in, and they can work alongside you, your Open Claw, your Hermes, your other agents, and they can all discuss with each other and have these kind of adversarial conversations to get way better results.
Many other things built in that are incredible, and we're gonna go into why these things are so special soon. Another reason, built in repositories. Repositories are built in.
They're not just replacing Slack and Discord in here. They're also replacing GitHub. Repositories are built into this.
So you can create projects directly in Buzz, replacing GitHub, and then you can then go into your different channels and have your agents actually write and commit code into these repositories, into these projects.
This is, again, amazing because now for the first time, repositories, AI agents literally living in the same place. You don't need to glue it together with MCP servers to GitHub and this and that all in one place.
And that opens it up even more with the fact now that you can bring human beings in here, and your humans, your AI agents can all work together to write code together and ship code and commit it to the local projects. There's never been, like, a multiplayer AI setup like this before.
There's never been, like, a multiplayer coding setup where you have multiple people working together in the same space to build projects. You know, typically, you'll create your commit separately. They'll the code will step on each other's toes.
You'll have to do merges. You have to figure out what's going on. Here, everyone works alongside each other so you can see who's doing what, who's building what, who's thinking what, whether it's your AI agents, your humans.
In Buzz, they're the same thing, and it makes building code so much nicer. Here is my list I made of, like, the biggest special things about Buzz. This first one is the biggest thing.
This is the most important. When it comes to AI, context is everything.
Having good context and having crap context is the difference between getting amazing performance out of AI and awful performance out of AI. The more context, the more relevant context you can have, the better your AI is going to perform.
Now what's amazing about Buzz is now for the first time, all your context is in one place. All your agents get many pieces of context with every prompt.
They're getting the entire conversation history of the channel they're in. They're getting any sort of agent prompt description you gave when you put them in here. They're getting all the context from their past conversations.
Like, if you have your Open Claw, your memories from your Open Claw from your other conversations, they're getting context of the other people you have in your Buzz. They're getting context around the projects and repositories inside your Buzz, and they're getting context from all the other agents in the Buzz.
Previously, when you're using codex or you're using Claw I don't know. You just saw fly across It's my codex doing work off screen. I don't know what the hell is going on there.
Um, but previously, when you were doing work in clawed code or codex or Hermes or OpenClaw, all of your context is separate. Your Open Claw can't see the context from Hermes.
Your Hermes can't see your conversations from Claude Code. Your Claude Code can't see the work you've been doing in Codex. Codex is just clicking on my screen right now.
This is what happens. You have too many AI agents that could be they just click around and do things, and you're not even looking. Now all that context is in one single place, all in one place.
That's amazing because that gives so much more power to the work all your other AI agents are doing. So when I go to Henry, my open claw, and I say, hey. Work on our project in here, this on screen notes app I'm building.
It's gonna have context around all the conversations I'm having with my Hermes, and it's gonna have context around conversations I've had in other channels discussing the project. I don't need to update it on the conversations from my other agents. That is really, really unbelievably powerful.
At its absolute core, I think that's truly what the value buzz is is centralized context for all your work. That is incredibly powerful.
There's way more to it, though. When we get to the shared compute part, I think it'll blow your socks off, but there's way more to it, though. Having all this shared context makes other things more powerful.
For instance, search. I can easily search for anything here. Maybe I'm working on an app, and maybe I've worked on it.
If you're anything like me across Claude Code, Codex, other agents, Hermes, OpenCloud, they're all contributing different ways. If I'm searching for a specific thing, I'm gonna have to go to each one of those agents and search to see, okay.
Who added this feature? Who added this feature? Here, I can just search in one centralized place for everything.
Like, if I'm searching for my agent Henry, I'm not only seeing the agent Henry, things they've said, things I've I'm seeing every single thing anyone said to Henry. I'm not only see I'm seeing all the thinking logs of anyone who's thought about Henry in the search. The search is incredible.
It also gets thinking logs in here. So you can search amongst AI agents' thoughts to see if anyone's thought of Henry.
That's how unbelievable the context in this app is. And, in the second, we're gonna get to use case. We're gonna get to the shared compute stuff.
I just wanna cover what makes Buzz so special and different first. We talked about the adversarial agents. This is also huge for coding, by the way.
Adversarial agents are amazing for coding. So if you build something out, sending that to another agent.
For instance, if you build something in codex, having Claude code check it over usually makes the code better and vice versa. That's now super easy in here. Instead of having to write skills where Claude code calls Codex or Codex calls Claude code, you just have it all in one place where you can just tag Claude and Claude and hey.
Check this out. What do you think? And they'll work together and improve the code.
No, again, taping things together through MCP or plugins or skills or anything like that. So it's not just for writing newsletters. It's also for coding.
Anything you do, it's just super easy to have adversarial agents. Traceability, I just kinda showed you that in the search.
You can trace everything, every thought, every message, anything that happens, easily search through. It's open source.
It's free. This is what Jack Dorsey's been about ever since he sold Twitter. I think everything's open source.
It's amazing. Very, very amazing. I'm very pro open source, so you can customize this any way you want.
No more paying, like, $10 a month. I I pay $10 a month for Slack. I don't even what the hell I get.
I think I just try to join someone else's channel one day. Like, you have to pay for it. Like, what am I paying for?
To join a chan I I didn't understand it. It's free. No more Slack $10 a month for god knows what.
And the pooled compute part, which again, this is gonna be exciting. I'm covering this a little later as well. So use cases, when would you use Buzz as opposed to Claude code or Codex?
If you work with people, right, if you have teammates or things like that, you should be using Buzz for pretty much everything. You should be bringing your teammates in, bring your agents in, and you all can have this kind of multiplayer building experience, which is incredible.
If you're building anything with teammates, anything with other people, you should be using Buzz for your vibe coding. If you're a solopreneur and you're doing kinda your own hardcore vibe coding, I still lean towards Codex at this moment.
I'm sure there's gonna be way more features that might change my mind. I still lean towards codex for kinda solo hardcore vibe coding. But for adversarial code checks, I would bring it into Buzz and have the adversarial code checks there.
Multi agent adversarial reviews and research is also amazing here. Like, I showed you how I built this newsletter by having my Open Claw and Hermes actually debate with each other.
Any type of kind of deep research you're doing, I would bring that into Buzz, and I would have it be adversarial. I would say, hey, Hermes, open claw, clawed code, codex. Go research this for me.
Everyone get your responses, present them to each other, debate, and bring it into one concise response that you deliver to me. I've gotten really, really good results from that.
So I would take advantage of Buzz for kind of adversarial research as well. Any kind of big decision you're doing from your business perspective, I'm really liking Buzz and the conversations here for that.
Local AI, if you have any sort of decent hardware, Local AI here. And, again, we're going deep into this in a second here, but it is so easy to spin up Local AI agents in here.
If you're doing LM Studio or other apps, I mean, it's not difficult, but it's still a little complex. Here, it's super easy.
You just go into your settings. You go into compute, and you can just choose a model. It even actually recommends models based on which computer it's on.
You just click it and boom. You have a local model running. I think really everyone should be running, no matter what hardware you have, because it's gonna find a model you can actually run.
By the way, if learned anything so far, make sure to leave a like down below. Subscribe, turn on notifications. I'm always just using the most cutting edge tools, and I tell you about it, and I tell you how to get the most out of it with the most actionable AI channel on planet Earth.
You got any value out of this, make sure to subscribe and turn on notifications. Also, Vibe Coding Academy, I'm doing a live boot camp on this in the Vibe Coding Academy next week. Make sure you sign up for that below.
Number one AI community on planet Earth. Link for that down below. Now let's get to a part here.
I am super, super excited about, and I think it's gonna change a lot of things. That is the screen we're on right now, shared compute. What is shared compute inside of Buzz?
Shared compute is the ability for everyone in your community to pool together your compute and then spin up models and agents on that compute that you all can talk to. Why is that so amazing and so different? First of no one else is doing that.
There's there's literally just no other app on planet earth that does anything remotely like this. But second, this allows you to basically have, like, your own where you all can share your intelligence and just use free local intelligence with everyone in your community.
So for instance, if you watch any of my previous videos, you know I have bought an absurd amount of computers over the last few months. I have three Mac Studio five twelve gigabytes.
I have a few DGX Sparks, a few Mac Minis, and AMD Halo. I have, like, two and a half terabytes of compute ready to go at all times. I do a bunch of things with those, but a lot of the time, one or two of those computers will sit idle.
I can now connect all those computers to this community and have basically two and a half terabytes of compute ready to go that anyone in this Buzz community can use, and that's a lot.
You can get versions of Kimi k three right now. They're, like, 500 gigabytes. That means I can have, like, five Kimi k threes in here ready to go that anyone can just talk to and use completely for free.
And in fact, that's what I'm going to do. Over the next few weeks, I'm going to open up this Buzz community with all my compute connected to it to my Vibe Coding Academy community.
If you wanna take advantage of, like, all the Mac studios I have sitting in this office right now, I'd highly encourage joining the Vibe Coding Academy right now, because I'll probably limit the amount of people that can join the Buzz community just so my models don't get overloaded. So if you join now, you'll have first dibs on it.
But I'm able to connect all these computers, run a bunch of models on them, and then everyone in here cannot only talk and use this free compute that's being hosted on my side, but what's gonna be really amazing is everyone in your Buzz communities that's sharing compute can actually team up together and build apps together, multiplayer app building.
You can come in here and leverage the local compute that's pulled together, and everyone kinda builds projects together and adds features. Again, there's never been anything like this before.
Nothing even remotely close to it. And now Buzz unlocks that. I wanna be on the forefront and cutting edge of this stuff, so I'm going to connect all of these computers to my Buzz community and open it up to the Vibe Coding Academy over the next few weeks.
Just wanna make sure I set up so it's scalable and everyone can use it. So make sure you sign up for that. That's amazing.
But this idea of pooled compute is incredible, especially as compute gets more scarce and more expensive and harder to get.
Being able to enter kinda almost like a compute intelligence commune, and everyone contributes their compute and shares it together and can use it to get free tokens and intelligence. That is a really, really amazing concept.
It's kinda like the Bitcoin compute in a way because that's kinda like the idea of crypto is that it's, like, decentralized, which is very, I guess, alongside, uh, what Jack Dorsey is all about.
He's very Bitcoin, uh, decentralized. All of that stuff is just very Jack Dorsey.
But it's really cool when it comes to this compute stuff. It's really amazing. Should you be using Buzz?
I think everyone universally should be using Buzz. You should download it and use it. Now depending on your use cases and what you do, I think that you'll get different amounts of value out of it.
I think if you have a team and you have multiple people you build with every day, you'll get tons of value out of it. Really, really amazing. If you're solopreneur, you're hardcore hacking by yourself, I still think you'll get value out of it.
I still think you need to use it at the very least. Like, you need to just get used to it because I think this is the future. I am very convinced this is the future.
But I think for vibe coding, Codex is gonna be the way to go. Abel, amazing model, but I think Codex is by far the best harness for AI agents and vibe coding right now.
By way, if you missed my, uh, Chad GBT voice video from a few days ago, it's one of the best videos I ever put out. Make sure to check out my channel and see that. I think it's one of the most amazing features I ever used.
Anyway, Buzz, you need to download it. It's really amazing.
I really believe it's the future. Way more is probably gonna come from this. They have shipped over the last week so many new features.
I mean, this project stuff just dropped. I hope this was helpful. I am so excited to get back into this and keep using it again and connect all my computers.
So I'm gonna get to that. I appreciate you all. I love you all.
Thank you for watching my videos. It really means the world you sit here and listen to me for however how long this video was. Hope you have amazing weekends.
I'll see you in the next video.
The Hook
The bait, then the rug-pull.
Alex Finn spends nineteen minutes walking through Buzz, a new multi-agent chat app from Jack Dorsey that treats AI agents as native participants instead of bolted-on plugins, and closes by offering to open his own idle compute to anyone who joins his paid community.
Named ideas worth stealing.
06:36list
What Makes Buzz Different
- All context in one place
- Adversarial agents are easy
- Traceability
- Open source / free
- Pooled compute
A five-point recap slide the creator holds on screen for several minutes, summarizing Buzz's core differentiators against Slack, Discord, and standalone AI-agent tooling.
Steal forany demo video that needs a clean on-screen recap of a product's differentiators
How they asked for the click.
VERBAL ASK
13:52product
“I'm doing a live boot camp on this in the Vibe Coding Academy next week. Make sure you sign up for that below. Number one AI community on planet Earth.”
Soft-pitched mid-video right after the feature walkthrough, paired with a like/subscribe ask just before it, then reinforced later when he says he'll open his own pooled compute to Academy members first — turning a scarcity hook (limited spots before models get 'overloaded') into the close.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
