Model vs. Harness: Where AI Agent Controls Belong
A comparison of what AI models should learn versus what agent infrastructure must enforce. See how reasoning, permissions, isolation, budgets, and observability split across the boundary.
DatabricksBradley Axen, Brandon Kvarda, Julia Powell, Kelly AlbanoAnnouncement
Block developed Buzz, an open source collaboration workspace where employees and AI agents work together in channels, threads, and direct messages. Users create agents, choose a model or coding harness per agent, and collaborate with multiple agents and colleagues in one environment. Databricks Unity Gateway provides centralized access to frontier and open-weight models from OpenAI, Anthropic, and Google through a single integration. Block experiments with multi-agent workflows, using a more capable model to coordinate agents powered by smaller, more efficient models, and has an internal benchmarking approach for different configurations. Buzz is model-agnostic, self-hostable, and supports multiple agent harnesses including a lightweight version of Goose.
As AI agents became part of everyday work, Block needed a way for employees to work with agents in the same place they work with each other, rather than switching between different AI tools. The company developed Buzz, an open source collaboration space where people and AI agents can work together in group discussions, focused conversations, and one-on-one exchanges. Databricks Unity Gateway provides the model access layer, giving employees flexibility to experiment with different models and match them to specific tasks.
Block developed Buzz, an open source collaboration workspace, and integrated it with Databricks Unity Gateway to provide centralized access to frontier and open-weight models from providers such as OpenAI, Anthropic, and Google. Users create agents within Buzz, select a model or coding harness per agent, and collaborate with multiple agents and colleagues in channels, threads, and direct messages. Block also experiments with multi-agent workflows where a more capable model coordinates agents powered by smaller, more efficient models, and has an internal benchmarking approach to evaluate different configurations.
Buzz is an open source collaboration workspace where employees and AI agents can work together in channels, threads, and direct messages, with support for multiple agents and colleagues in the same environment.
Databricks Unity Gateway provides centralized access to frontier and open-weight models from OpenAI, Anthropic, and Google through a single integration, allowing users to select different models per agent.
Block experiments with multi-agent workflows where a more capable model coordinates agents powered by smaller, more efficient models, and has an internal benchmarking approach to evaluate different configurations.
Block Buzz is an open source collaboration workspace where employees and AI agents can work together in channels, threads, and direct messages. Users can create agents, choose their models or coding harnesses, and collaborate with multiple agents and colleagues in the same environment.
Block uses Databricks Unity Gateway to provide centralized access to frontier and open-weight AI models within Buzz. Employees can select Databricks as their inference provider and choose from available models, including those from OpenAI, Anthropic, and Google, through a single integration.
Multi-agent workflows allow teams to use different models for different parts of a task. For example, a more capable model can coordinate complex work while smaller, more efficient models handle simpler tasks. Block is experimenting with these configurations to understand how model and harness selection affects performance and cost.
Summaries are generated from the original by a language model. Check the original before relying on it.
See the idea
A comparison of what AI models should learn versus what agent infrastructure must enforce. See how reasoning, permissions, isolation, budgets, and observability split across the boundary.
Keep reading
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
One useful email. Unsubscribe anytime.