Alex Kerss tests Buzz, an open-source workspace from Block where humans and agents share channels, threads and messages. A simple request for two agentsA multi-agent system contains multiple AI agents that interact, coordinate, divide work, or influence one another while pursuing tasks. to talk immediately produced a long loop because each mention triggered another responseAn AI feedback loop occurs when AI outputs or their consequences become inputs that influence later model or agent behavior..
Buzz provides identitiesAn AI agent identity is a persistent, verifiable identifier used to attribute messages, actions, permissions, and history to an agent., routing, event storage, agent tools and a shared relay, but it does not automatically enforce manager-worker roles, task dependencies, joins, retries or deterministic completionAgent completion verification checks evidence that an agent achieved the requested outcome rather than merely producing intermediate work.. A simple manager-and-two-workers prompt succeeded, yet the language models, not an orchestration engineAgent orchestration coordinates AI agents, tools, people, tasks, state, and control flow so a larger workflow reaches a verified outcome., were responsible for following the workflow.
Complex workflows can therefore fail through missing mentions, silent worker errors, premature manager replies or endless conversations. Alex Kerss also notes that business adoption depends on clearer hosted-service maturity, administration, compliance, storage and recovery guarantees.
Watch on YouTube




