BASF Coatings: Scaling Enterprise Multi-Agent Systems with Federated Governance on Databricks
Overview
| Experience | In Person |
|---|---|
| Track | Artificial Intelligence & Agents |
| Industry | Manufacturing |
| Technologies | AI/BI, Databricks Apps, Lakebase |
| Skill Level | Intermediate |
| DOWNLOAD SESSION SLIDES | |
Session Speakers
Bernhard Bartmann
/Product Owner AI Solutions
BASF Coatings
Lars Schrameyer
/Head of Data & AI Office
BASF Coatings
Full Summary
Inside MAUI: How BASF Coatings scaled governed multi-agent AI on the Databricks Platform
Enterprises racing to build AI agents often end up with duplicated stacks, fragmented governance, and rising risk. BASF Coatings chose a different path by creating MAUI, a governed multi-agent interface on the Databricks Platform that turns trusted enterprise data into self-service answers. The strategy shifts focus from building one-off agents to building a shared foundation that makes every future agent cheaper, faster, and safer, echoing patterns in Databricks' guidance on enterprise AI agent trends and governance. For organizations evaluating how to govern and scale these capabilities, Databricks outlines a practical starting point in its executive guide to democratizing access to quality data.
FAQ
Trusted data products already existed on the Databricks Platform, but business users lacked a simple, everyday access layer. Answers remained slow and manual, central teams became bottlenecks, and shadow AI was growing. MAUI adds a governed, self-service interface that turns governed data and documents into usable answers.
Each domain appoints data stewards responsible for quality, availability, and compliance. Domain experts provide business context. Together they own the agent, ensuring accountability lives where data and knowledge reside. The central team supplies the platform, reusable components, guardrails, and onboarding, rather than building every agent.
Structured sources are connected via GenAI spaces over governed tables. Unstructured documents are ingested into a vector index using standard or advanced RAG pipelines. At runtime the agent routes questions to both and merges results, enabling answers that blend metrics with narrative insights.
MAUI captures direct feedback and analyzes conversation traces for indirect signals, such as abbreviation confusion. It compares those signals to prompts, GenAI metadata, vector indices, skills, MCPs, and tools, then proposes concrete fixes for owners to accept. Reporters are notified when improvements are applied.
Priorities include document and report generation, proactive notifications and scheduled prompts, and deeper MCP-based integrations with systems like SAP, Salesforce, and Office. The goal is to move from ask and answer to a proactive enterprise workflow orchestration layer.