How Block powers collaborative AI agents with Buzz and Unity Gateway

Block’s Buzz is an open source collaboration platform where employees and AI agents work together in channels, threads, and direct messages, enabling group discussions and focused exchanges. Users can create AI agents, select their underlying models or coding harnesses, and invite colleagues to join ongoing work within the same environment. The platform includes a built-in lightweight agent based on Goose and supports additional coding harnesses and tools, ensuring flexibility in agent configuration and deployment.

Databricks Unity Gateway powers the Buzz experience by providing centralized access to frontier models from OpenAI, Anthropic, and Google, as well as select open-weight models, through a single integration. Employees can choose Databricks as their inference provider and select models based on task requirements, performance, and cost, rather than relying on a single model for every workflow.

Buzz enables multi-agent workflows where different models and harnesses can be assigned to specialized agents, allowing teams to optimize performance and cost for specific tasks. Engineers at Block are experimenting with configurations where a more capable model coordinates complex work while smaller models handle simpler steps, all within the same collaborative workspace.

Block Buzz is designed to be model-agnostic and supports self-hosting, giving organizations control over deployment and configuration. The platform reflects a shift toward combining multiple models, harnesses, and agents to complete work effectively, moving beyond the focus on individual model performance.