Jack Dorsey co-founded Twitter. Then Block. Now he's shipped something people are calling a Slack killer.
It's called Buzz. And the pitch is simple: it's Slack, except half your teammates are AI agents.
Not bots that post standup reminders. Agents that brainstorm with you, build software in the chat, deploy it to the internet, and send you screenshots before you've even checked what they were doing.
Greg Isenberg just did a full walkthrough with Vinny from Wasp, the guy whose Buzz breakdown got retweeted by Jack Dorsey himself. I watched the whole thing so you don't have to.
Here's what Buzz actually is, the parts nobody's talking about, and my honest take on whether a small bootstrapped team should switch.
What Buzz actually is
Buzz looks like Slack. Channels, DMs, huddles.
The difference: agents are first-class citizens. They're not an integration you bolt on with an API token. They're members of your team, sitting in your channels, reading the same context as everyone else.
You @ an agent the same way you @ a coworker. "Fizz, let's tackle this bug." And it goes and does it.
Vinny's framing stuck with me: the sweet spot right now is solopreneurs and small teams. You use it to brainstorm, keep everything in context between your agents, and ship products faster.
Greg put a number on it: if you're a founder with 1 to 10 people and $0 to $10M in revenue, this is aimed at you.
That's most of the people reading this blog.
The killer feature: swappable harnesses
This is the part that sold Vinny, and honestly it's the part that sold me too.
Every agent in Buzz runs on a harness underneath. Claude Code, Codex, Goose, OpenCode, whatever you already use. And you can swap it.
Mid-project. Without losing anything.
All your chat history, all the apps you're building, all the context moves with the agent to the new harness and the new model.
If you've been building with AI for more than a month, you know why this matters. A new model drops every other week. Vinny calls it model fatigue: "I don't even want to hear the good things about model X cuz I'm still over here using model Y."
Buzz is a layer on top of that churn. Today's best model ships tomorrow? Swap the harness, keep the context, keep moving.
Your globally installed skills carry over too. If you've built up skill files for Claude Code, your Buzz agents can use them.
The demo that made me sit up
Vinny typed one message into a Buzz chat: spin up a simple CRM app using the Wasp full-stack framework and deploy it to Railway.
Then he went and did something else.
The agent created the project, made a feature branch, worked in a parallel worktree (so it never touches your local files), pushed the code to a remote repo, deployed it live, and posted screenshots back into the chat.
"That's all before I even checked out what it was even doing."
A working CRM dashboard, live on the internet, from one chat message.
And here's the workflow that follows: you add your teammates to that chat, argue about which features are missing, reach a conclusion, and tell the agents to go build it. The conversation IS the spec.
Greg's example was an agency pitching the NFL: drop your meeting notes into the channel, and agents spin up a custom proposal app for that one client, on the fly. Proposals are the lifeblood of an agency, and here they're a byproduct of a group chat.
The context flywheel (with real numbers)
My favorite thing Vinny showed wasn't the CRM. It was his tweet leaderboard.
He built a full-stack app in Buzz that tracks his team's Twitter output. Top tweets, impressions, who's posting what.
Then he closed the loop:
→ The app exposes a public API.
→ A daily Buzz workflow pulls the stats into a channel.
→ He replies to the digest and asks Fizz, "what's the common thread between my top tweets this week?"
The agent had the data. His top two tweets: over 1 million impressions. The bottom half: 88,000 combined.
The pattern it found: novelty gets you reach, friction gets you engagement. His winners were "I tried X as a beginner and here's what happened" posts.
He used to do this manually, exporting tweet data and pasting it into ChatGPT in a browser tab. Now it flows into the same place his team already talks, and the agents remember all of it.
The agents aren't the real insight of Buzz. The shared context is. Every chat, every app, every stat lives in one place, and every agent and teammate can see it.
I've written before about why AI agents are the new SaaS. Buzz is what it looks like when someone builds the office those agents work in.
The parts nobody's talking about
Three things from the video that most coverage skipped.
1. It's built on Nostr, and your data lives on your relay.
Buzz runs on an open protocol. Your chats, your code, your context sit on a relay you host (or Block hosts for you). Buzz even has its own Git hosting on that relay, so it's quietly going after GitHub too, not just Slack.
Why care? Vinny's point: Slack controls your data. Try leaving and taking years of message history with you. Good luck.
Same with models: your $200/month unlimited plan could be $2,000/month next year. Nobody knows. Owning your context means you can walk.
2. Bitcoin payments are coming.
Nostr is tightly coupled with Bitcoin Lightning. It's not live yet, but the settings already hint at agents paying for compute and people tipping each other for tasks. Imagine a designer in a public Buzz community dropping you a logo and you tip them on the spot.
3. Shared compute.
Flip one setting and you can run a local model on one machine and share it with your whole team. Vinny's example: a couple of college students split the cost of one beefy Mac Studio and everyone gets an AI teammate without paying for four max plans.
The honest downsides
Vinny was straight about this, which I appreciated.
It's early. Beta, maybe alpha. Workflows and recurring tasks "weren't really landing great" when he tried them.
It's also slower than working directly in Claude Code or Codex, because everything round-trips through your relay.
His verdict: great for solopreneurs and small teams iterating on small products. If you're doing genuinely complex software engineering, this is not your tool yet.
How to set it up without overthinking it
Vinny's setup advice was refreshingly lazy. Here's the whole playbook:
→ Don't build a fleet of custom agents on day one. The defaults are fine.
→ Pin agents to models by task weight. Vinny runs Fizz on a frontier model for heavy work and Honey on a cheaper one, so he's not burning premium tokens on small tasks.
→ Once you do add specialists (copywriter, code reviewer, brainstormer), create one "Chief Agent Officer" agent and delegate the delegation. You tell it the task, it picks the right agent.
→ Talk to agents like people. No prompt wizardry. Vinny's actual prompts have typos in them.
That last one matters. Prompt tricks stopped being the edge once agents got this much context, and context is Buzz's whole design.
If you're managing more than a couple of agents, the skill itself is worth studying. I broke down Greg's earlier video on exactly that in the most valuable skill of 2026: managing AI agents.
Should you ditch Slack for it?
Here's my take as someone who runs a content pipeline on 11 AI agents.
Not yet. But you should absolutely try it this week.
Greg said something at the end that I'd frame and put on a wall: even if you hate Buzz, you'll come back to Slack knowing exactly what it's missing. "There is no losing in this situation."
The founders who win the next few years are the ones getting their hands dirty with these tools while they're still rough. Vinny is already pushing his team at Wasp to move onto it. Not because it's polished. Because whoever learns to work this way first gets a real edge.
And it's open source, built on an open protocol. If it's missing something, someone will add it. That's how these things compound.
The future of work looks like a group chat where half the members are agents, everything you've ever discussed is shared context, and software gets built as a side effect of conversation.
Buzz isn't that future yet. But it's the clearest glimpse of it I've seen.
FAQ
What is Buzz by Jack Dorsey?
Buzz is an open-source team chat app from Jack Dorsey (co-founder of Twitter and Block) where AI agents are full team members, not add-ons. Agents share your channels' context, build and deploy software from chat messages, and run on swappable harnesses like Claude Code, Codex, or Goose. It's built on the Nostr protocol, with your data stored on a relay you control.
Is Buzz really a Slack killer?
Not today. It's early software (beta at best), workflows are flaky, and it's slower than working directly in a coding agent. But it does things Slack can't: agents as native teammates, shared context across your whole team, built-in Git, and no vendor lock-in on your data. For solopreneurs and teams under 10 people, it's worth testing now.
Do I need to be technical to use Buzz?
Less than you'd think. Vinny deployed a full CRM by typing one sentence into a chat. The agents handle repos, branches, and deployment behind the scenes. Knowing your way around tools like Railway or GitHub helps you ask for better things, but the agents do the actual work.
What does Buzz cost?
The app is open source. Your real cost is the AI models your agents run on, and Buzz lets you control that: pin cheap models to light tasks, use frontier models only where it counts, or share a local model across your team from one machine using the shared compute feature.
Steal how real founders are using this stuff
Every week on the Profitable Founder Podcast I sit down with bootstrapped SaaS founders making $100K to $10M a year and pull out exactly how they build, ship, and grow. The tools change fast. The playbooks compound.
