01 / AI · Analysis
Buzz lets Block staff choose models for each coding agent
Buzz brings colleagues’ code review and model selection for individual agents into one workspace, but Databricks disclosed no cost figures for different combinations.

Databricks, a data and AI company, announced that Buzz supports model selection through a single integration when employees at payments company Block create agents.
According to Databricks, Block employees use Buzz for coding and analysis, with colleagues reviewing agents’ work and adding context in the same conversation. This puts human review alongside the agents’ work in a shared workspace.
The Databricks introduction describes model and coding-tool selection for each agent. The unit of choice is therefore the individual agent brought into a shared conversation, allowing different agents to use different configurations.
A shared workspace for coding tools
Databricks said Buzz supports other coding tools alongside its built-in agent. Employees can create an agent and bring it into a conversation with colleagues.
Databricks also described Buzz as open source and model-agnostic: its source code is available, and it is not tied to a particular model. It can be self-hosted on an organisation’s own network or systems.
Experiments with model combinations and costs
The Databricks introduction identifies Unity Gateway as the model access layer. Employees select Databricks as the inference provider—the service that runs the model—and choose from models supplied by leading AI companies and selected open-weight models, whose model weights are publicly available.
Databricks said Block experimented with a more capable model coordinating work across agents powered by smaller models. Block also developed an internal evaluation approach to compare combinations of models and coding tools.
The workflow described by Databricks brings model selection, agent coordination and colleagues’ review into one workspace. Whether particular combinations reduce costs remains unresolved: Databricks disclosed neither processing-time nor cost figures for those configurations.