From Chatbots to Colleagues: What an AI Agent Workspace Really Means
An AI agent workspace is a collaboration platform where humans and AI agents share the same channels, threads, files, and workflows, with agents treated as persistent team members that hold identities, context, and responsibilities across ongoing work instead of responding to isolated prompts like traditional chatbots. The shift is not cosmetic. It challenges the idea that AI belongs in side panels and pop-up assistants, and insists that if software is going to help make decisions, it should sit where decisions are made: inside the team’s main workspace, visible and accountable. Block’s new Buzz platform and Offloop’s private beta both embody this view, and together they signal a quiet but important redesign of human-machine collaboration that will matter far more than yet another chatbot release.
Buzz: Open-Source Collaboration Where AI Agents Get a Passport
Buzz is Block’s answer to the question, “If AI is part of the team, why doesn’t it sit in the same workspace?” Buzz is a free, open source collaboration platform that looks like a familiar Slack-style interface, with channels, threads, direct messages, voice, media sharing, and code repositories. The difference is that AI agents join those spaces as full members, not hidden tools invoked through one-off commands. Buzz is built on Nostr, a decentralized messaging protocol where every participant is represented by a public–private keypair. Buzz gives each agent its own cryptographic identity and then adds a second signature tying that identity back to a specific human owner, creating a verifiable paper trail of action. This encrypted AI identity makes agents auditable: workspaces can enforce rules such as allowing only agents owned by a channel member to enter private spaces or trigger workflows. In a world of opaque SaaS bots, Buzz’s insistence on clear, signed accountability feels less like a technical detail and more like a moral stance.
Offloop: Recurring Work and the Rise of AI Dispatchers
If Buzz asks how AI joins the conversation, Offloop asks how AI keeps the work moving after the meeting. Offloop’s AI agent workspace, now in private beta, organizes collaboration around persistent Channels that hold people, agents, messages, files, schedules, incoming signals, tool progress, and finished artifacts in one place. Agents don’t wait for a fresh prompt; they inherit full context and can be activated by schedules or external events, turning one-off requests into loops that continue running in the background. Offloop coordinates several AI team members per Channel using D1, a dispatcher model that decides which agent should act at each step as context shifts. The platform connects to tools like Notion, GitHub, GitLab, Sentry, Supabase, Vercel, Railway, Cursor, Codex, and ChatGPT and is aimed squarely at small teams whose launches, customer follow-up, research, feedback, growth, and operations stall because no one has time to chase loose ends. In Offloop’s design, AI is not a helper; it is the colleague whose job is to care about the follow-through that humans are too busy to own.
Why Persistent AI Team Members Matter Now
The move toward AI agents as real team members is not happening in a vacuum. Buzz arrives as Block reshapes itself around AI, cutting more than 40 percent of its staff—from more than 10,000 employees to just under 6,000—after its founder described AI as enabling “a new way of working.” That “new way” clearly expects software to take on recurring, economically valuable tasks rather than remain a novelty chatbot. Offloop is explicit about the stakes. It reports that its multi-agent harness reached state-of-the-art performance on GDPval, OpenAI’s benchmark of real-world work spanning 44 occupations across nine sectors, across jobs that pay USD 2.4 trillion (approx. RM11.0 trillion) a year in the US. It also scores 67.2 percent on JobBench and claims higher performance than other frontier multi-agent baselines at roughly a fifth of the cost per task. Those numbers are not marketing fluff; they are a challenge to knowledge workers who assume AI remains a toy and to managers who still treat AI licenses like optional perks instead of core infrastructure.
Open Protocols vs. Closed SaaS: Choosing the Future of Collaboration
The most radical part of Buzz is not that agents join channels; it is that the whole AI agent workspace rides atop open protocols instead of another locked SaaS silo. Buzz’s code has been on GitHub under an Apache 2.0 license, and it runs on Nostr relays that teams can host themselves, with nothing routing through Block if they prefer to own their infrastructure. Teams can stand up their own relay or use Block’s free hosted option while it is in beta. Offloop, while invite-only and more traditional in its delivery, still treats agents as named participants that operate inside shared Channels, and offers access tiers from individual operators up through enterprise with custom connectors and audit-friendly runtimes. This divergence matters. Open source collaboration platforms like Buzz lower barriers to adoption and help ensure that encrypted AI identities and human-machine collaboration standards can evolve in public rather than being locked inside proprietary stacks. Offloop’s channel-based design shows how AI team members can drive recurring workflows today, but open, protocol-based systems hint at a future where your agents are not owned by any one vendor. The teams that plan for that future now will have more control—and more credible accountability—when AI is no longer an add-on but a central part of how work gets done.
