Jack Dorsey’s Block released Block Buzz on 21 July 2026, a free, open-source workspace where people and AI agents work in the same channels. Instead of treating agents as chatbots waiting for prompts, Buzz gives each one its own identity, permissions, channel membership, and audit trail, and lets it contribute alongside human teammates.
Dorsey pitched it plainly: a place for teams of people and agents, built to reduce Block’s dependency on Slack and GitHub. It is available now at buzz.xyz, as a hosted service or as software teams run themselves.
For marketers, the interesting part is not another team chat app. It is the bet that human and AI agent collaboration becomes a normal operating model, with shared context replacing the usual cycle of prompting, copying, pasting, and explaining the same campaign again. Whether that bet pays off for marketing teams specifically is a different question, and worth being honest about.
What Block announced
Block Buzz combines channels, threads, direct messages, voice, media sharing, code repositories, and automated workflows. The interface is built to feel familiar to anyone who has used a modern team communication tool. Desktop apps for macOS, Windows, and Linux were available from launch, which is worth noting against Mac-only releases like OpenWorker the same week.
The difference is how agents participate. Each agent joins as a member with its own permissions and can post, review code, run approved automations, and contribute to conversations the way a colleague would, rather than appearing as a single assistant bolted onto one human’s account.
Buzz is built on Nostr, the decentralised protocol Dorsey has funded for years. It gives every human and agent a cryptographic keypair that belongs to them, not to the platform. Block calls identity the most fundamental problem in multi-agent work, and this is its answer: an agent’s identity and history travel with it across any Nostr-compatible system rather than disappearing when a vendor revokes an account.
There is a second mechanism worth understanding, because it is the one that matters for accountability. A second signature ties each agent back to its human owner. Block describes the result as a cryptographic paper trail that neither the human nor the agent could produce alone. In plain terms: every action an agent takes is signed, attributed, and traceable to a person. That is a real answer to “who approved this,” and most agent platforms do not have one.
Buzz is model-agnostic and agent-agnostic. Teams can run agents powered by Anthropic’s Claude Code, OpenAI’s Codex, or Block’s own goose framework, released in January 2025 and now past 50,000 GitHub stars. The software ships under an Apache 2.0 licence, so companies can self-host with full control over their relay, agents, and data, or use Block’s hosted version.
Block is its own first customer. Its internal coding agent, BuilderBot, handles more than 200,000 operations a day and accounts for roughly 15% of Block’s production code changes, and it ran inside Buzz throughout development.
Read the “engineering tool” signal before you plan a marketing use
Here is the part most launch coverage soft-pedals. Buzz is pitched as a replacement for Slack and GitHub. Git hosting is built in through a software forge, feature branches become channels, and patches, continuous-integration results, and review comments are stored alongside the discussion that produced them. Its three native agent frameworks are all coding agents.
The self-hosted path means standing up a Nostr relay and managing Rust and Node version requirements before the first channel exists. That narrows the immediate audience to software engineering teams already coordinating agents.
No B2B SaaS marketing team is installing this next quarter, and pretending otherwise would be selling you a use case that does not exist yet. The value of watching Buzz is not the product. It is the operating model, which is portable to marketing whatever tool eventually delivers it.
Why the shared-context idea matters for marketers
Most marketing teams already have AI. Their harder problem is coordination.
A campaign brief sits in a document. Customer objections live in Slack. Performance data is in three dashboards. Sales feedback arrives on a call. The AI tool sees one fragment, returns a polished answer, and misses the context that would have made it useful.
Buzz is built around the opposite idea: keep humans, agents, conversations, files, and approved actions in one shared space. Applied to marketing, that is what would make AI marketing automation more than a prompt box. An agent could read a launch channel, pull approved product claims, draft assets, flag missing information, and route work to a human reviewer without losing the discussion that produced the request.
For content teams, content marketing automation stops being about generating more drafts and starts being about holding context. One agent tracks customer language, another compares competitor messaging, a third prepares a brief, and the review happens in the same channel where product and sales made the decisions.
The same shape applies to paid media automation. A reporting agent watches campaign data, a second checks CRM quality, a third compares landing page claims against the current product brief. The useful output is not “CPL increased.” It is a traceable account of what changed, which evidence was used, and which action still needs sign-off. Deciding which of those calls an agent should make alone is the same judgement covered in AI marketing automation tools.
The idea overlaps with approaches already in market. Claude Tag puts Claude inside selected Slack channels. ChatGPT workspace agents let teams build persistent agents across connected tools. Buzz pushes one direction further: it designs the workspace itself for humans and agents together, rather than adding agents to a platform built for people.
For B2B SaaS marketing, shared context matters because work moves across product, sales, marketing, and customer success. A useful agent needs more than a prompt. It needs the right information, a narrow job, and a clear point where a person takes over.
How Block Buzz compares with other agent platforms
Buzz is entering a crowded category, and each rival solves a different problem.
| Option | Main approach | Marketing relevance |
|---|---|---|
| Block Buzz | Open-source workspace where agents and people are first-class members, on portable Nostr identities | Best for teams that want model choice, self-hosting, shared context, and audit trails, and have the engineering capacity to run it |
| Slack with Claude Tag | Adds a capable agent to existing Slack channels | Easiest for teams already committed to Slack, with the least operational change |
| ChatGPT workspace agents | Persistent, admin-governed agents connected to common business tools | Strong for scheduled reporting, research, routing, and repeatable internal workflows |
| OpenAI Presence | Managed deployment for customer and employee voice or chat workflows | Better matched to production service interactions than open internal collaboration |
| Sakana Fugu | Coordinates multiple models behind one API | An orchestration layer, not a workspace; solves model routing, not team coordination |
The Sakana Fugu distinction is the one to keep straight, because it is easy to conflate. Fugu coordinates models behind a single API. Buzz is the shared place where agents and their human teammates communicate, review work, and build on each other’s output. One routes queries; the other hosts a team.
The difference from OpenAI Presence matters too. Presence handles managed customer-facing and employee-facing workflows with testing and escalation controls. Buzz is broader, more open, and earlier: it gives you infrastructure, and leaves the operating rules to you.
What marketing teams should actually watch
The first test is not whether an agent can write a decent campaign brief. Most leading models can. The real test is whether a team can trust the surrounding system, and Buzz gives concrete, checkable answers rather than promises.
Which agent accessed which source? The dual-signature audit trail records it. Who approved an action? The same paper trail, tied to a named human. Can a mistake be traced and reversed? The signed event history is designed for exactly that.
Where the answer is less comfortable is scope. Buzz’s access control is coarse today. Per Block’s own security documentation, channel membership is the only access control mechanism: there are no separate permission lists or capability tiers, so any member of a channel can read and write to it. You limit an agent by limiting the channels it joins, not by fine-grained task permissions. Buzz is also pre-1.0, and its relay does not enforce TLS by default, a deliberate choice to allow deployment behind load balancers. Plan around what exists, not the roadmap.
Security deserves direct attention. TechRadar reports that messages and media on Block-hosted relays are not end-to-end encrypted, and connected model providers such as Anthropic and OpenAI may receive prompts and related channel content. Companies handling customer records, unreleased product details, campaign budgets, or regulated data should review the current hosting and data rules, and consider self-hosting, before putting anything sensitive through Buzz.
The open architecture gives teams more control and, in the same breath, more responsibility. Self-hosting, model choice, permissions, monitoring, and integration upkeep all require technical ownership. Teams already fighting a messy stack of AI marketing tools should not add five agents before fixing the workflow underneath them.
The next three to six months will show whether Buzz attracts a real ecosystem or stays a promising experiment. Distribution decides it. Slack, Microsoft, Google, and OpenAI already sit inside most company workflows. Openness only wins when the surrounding tools are reliable enough for ordinary teams to adopt.
OneMetrik Takeaway
Block Buzz gets one idea right: AI work is becoming a coordination problem, not a prompt-writing contest. The identity-and-audit model, where every agent action is signed and traceable to a person, is the piece worth stealing regardless of which tool your team eventually uses.
The wrong response is to let agents run marketing. The right one is to give narrowly scoped agents enough shared context to handle research, routing, reporting, and preparation, while people keep control of strategy, claims, budgets, and final approval.
If OneMetrik were testing Buzz, we would pick one repeated internal workflow with low customer risk, and define the inputs, permissions, expected output, human checkpoint, and failure condition before connecting a single agent. For a SaaS GTM strategy, that is the real lesson. The winning system will not be the one with the most autonomous agents. It will be the one where every agent has a clear job, every action has an owner, and a person can still see how the work was done.