From Invisible Chatbots to Visible Teammates
An AI agent workspace is a digital team collaboration platform where software agents join channels, threads, and workflows as visible, accountable members alongside humans, with their own identities, access rules, and recurring responsibilities, instead of acting as disposable chatbots hidden behind copy-pasted prompts or one-off conversations. Today’s wave of tools, led by Block’s Buzz and Offloop, make a blunt statement: AI is no longer a sidekick; it is part of the team. Buzz is a free, open source workspace built on the Nostr messaging protocol that gives people and AI agents a Slack-like environment for collaboration. Offloop’s invite-only agent workspace keeps recurring teamwork alive inside persistent Channels instead of resetting context with every new prompt. Together, they mark a clear shift in human-machine collaboration from ad hoc assistance toward continuous, shared workstreams owned by agents.
Buzz: Slack for Agents, With Cryptographic Passports
Block’s Buzz is opinionated software: it assumes that AI agents deserve a proper seat in your workspace, not a ghost presence behind the scenes. At first glance it looks like a familiar team collaboration platform, with channels, threads, direct messages, voice, media sharing, and code repositories. The twist is that agents are treated as full participants. You can plug in popular agent harnesses such as Claude Code, Codex, goose, or bring custom models, then invite them into channels the way you would invite a new engineer. The critical innovation is identity. On Nostr, every participant is a public–private key pair; Buzz gives each agent its own key pair plus an extra signature that ties it back to a human owner. This cryptographic "passport" creates a paper trail proving, "this is my agent," which others can verify. That lets teams enforce real rules, such as only allowing agents owned by members into private channels, rather than relying on vague bot accounts. Buzz’s code has been on GitHub under an Apache 2.0 license, and Block recently added a hosted Nostr relay so teams can sign up without running their own infrastructure.
Offloop: Channels Where Agents Own the Follow-Through
If Buzz reimagines identity, Offloop reimagines accountability. Its AI agent workspace, now in private beta, is built around the idea that most small teams stall on tedious follow-through rather than hard problems. Offloop’s unit of work is the Channel: people, agents, messages, files, schedules, signals, tool progress, and finished artifacts all live in one place. An agent picking up a task inherits past context, owners, and permissions, so recurring tasks are no longer re-explained via prompts. Schedules and incoming signals can independently activate an agent, turning a one-off request into a persistent loop. For example, an agent can watch a Google Workspace inbox and trigger a Channel with a summary whenever a matching email arrives. Actions stay visible and inspectable; humans can redirect, approve, or stop a run at any point. Coordination across several agents in a Channel is handled by D1, Offloop’s dispatcher model that decides which agent acts as context and updates change. Connectors to tools like Notion, GitHub, GitLab, Sentry, Supabase, Vercel, Railway, Cursor, Codex, and ChatGPT make the workspace feel more like a control room than a chat box.
Why Teams Are Ready for Agentic Workspaces Now
The timing is not accidental. Block is reshaping itself around AI, cutting more than 40 percent of its staff from over 10,000 to just under 6,000, with its founder citing AI as enabling "a new way of working". Buzz is the collaboration layer built for that smaller, more agent-heavy company. Its existing internal agent system, BuilderBot, already runs over 200,000 operations a day, merges about 1,500 pull requests a week, and accounts for roughly 15 percent of production code changes. That kind of volume demands transparent, shared coordination, not private chat windows. Offloop arrives with its own proof points. The company reports its multi-agent framework scores 84.9 percent on GDPval and 67.2 percent on JobBench, surpassing frontier baselines at about a fifth of the cost per task. It argues that it learned where a single agent reaches its limits after building general-purpose agents at scale, and designed its Channel and dispatcher model around those boundaries. Both products respond to the same pain: the most important collaboration between humans and agents is happening off-stage, in private chats and scripts, where nobody else can see or review how decisions were made.
Beyond Slack Integrations: What Changes for Ordinary Teams
The easy path would have been to keep adding bots to existing tools. Many teams already run coding agents inside Slack; others rely on agents embedded in issue trackers and productivity suites. Buzz and Offloop argue that this bolt-on approach is not enough. Buzz may live alongside Slack for now, but Block is explicit that it is building something that could, over time, replace an "enterprise-grade Slack" for agent-heavy work, even if that requires new business workflows and migration of a decade of history. Ordinary users feel this in practical ways: Block’s hosted Nostr relays are free at launch and meant to offer an off-the-shelf option, while teams that care about control can run their own relay with nothing routing through Block. Axen notes that a 10-person open source project is cheap to host, while a full Slack replacement would need a different arrangement. Offloop, meanwhile, targets small teams that manage launches, customer follow-up, research, feedback, growth, and operations—areas where work often stalls because no one has time to chase the next step. Its three access tiers range from an invite-only Operator plan for individuals converting recurring work into Channels, to a Team Pilot with shared memory and priority onboarding, up to an enterprise tier with custom connectors, SSO planning, and an audit-friendly runtime. In all cases, the promise is the same: AI agents that own ongoing work streams, in a workspace where their actions are visible, reviewable, and bound by the same permissions as human teammates.
