Agents at Work: Shipping Agentic Apps at Scale

Scale Without Sprawl: Watch the Full Webinar on Running Fleets of Enterprise AI Agents

Executive Summary

At Agents at Work: Shipping Agentic Apps at Scale, Databricks announced the general availability of Unity Gateway, a single governed AI control plane for every model, agent, coding agent, and MCP a team uses. Where AI governance has meant separate tools for cost, security, and observability, Unity Gateway brings guardrails, budget controls, and full usage visibility into one place, whether a team is running natively hosted frontier models or the latest open-source ones. The session also showed how Agent Bricks and Unity Gateway work together in production, and how Stellantis is applying the same governance pattern to a live supply-chain agent.

Key Takeaways

• Unity Gateway is now generally available as a single governance layer for multi-AI (ranging from natively hosted frontier models to the latest open-source models and frameworks) without rebuilding controls for each new provider.
• Centralized AI agent registry  now covers foundational models and managed MCP servers (Google Drive, GitHub, SharePoint, Atlassian, Slack), with built-in cost controls and full lineage and audit trails.
• Guardrails can be applied as input policies before the model runs or output policies after it responds, including PII blocking and custom classifiers, enforced consistently across every provider a team connects.
• Admins can set AI cost management controls such as budget thresholds by workspace, team, agent or user that alert or block usage, giving teams real-time visibility into which models and agents are driving spend.
• Agent Bricks and Unity Gateway work together in production: agents built and connected through Agent Bricks are governed end-to-end by the gateway, with every session traced in MLflow.