Google Cloud emphasizes the importance of customer collaboration in developing effective products. In January 2026, their Developer Advocacy team engaged in a technical sprint with a major telecommunications client, yielding valuable insights that significantly improved the Model Armor service, which focuses on runtime security for generative AI.
Driving GenAI Adoption with Empathy
The primary goal of this collaboration was to assist in the production of an advanced GenAI customer support platform utilizing Google Cloud's Agent Development Kit (ADK) and Agent Platform. By working closely with the client's developers and security experts, the team observed real-time interactions with the Gemini Enterprise Agent Platform, gaining insights that traditional documentation could not provide.
This approach fostered a deep understanding of the challenges faced by developers, allowing for immediate identification of friction points in their workflow.
Identifying Key Challenges
Through direct observation, four main friction points were identified:
- Search-First Workflows: Developers preferred searching for specific code examples rather than navigating documentation hierarchies. The absence of comprehensive, ready-to-use code snippets for common tasks, such as PII redaction, was a significant hurdle.
- Balancing Confidence Levels: Striking a balance between effective threat detection and minimizing false positives was difficult. For example, aggressive settings often resulted in excessive false positives, disrupting legitimate customer support processes.
- Need for Granular Guidance: While the fundamentals of Model Armor were understood, developers sought more detailed information on the practical application of various enforcement methods to enhance security without sacrificing usability.
- Integration Roadblocks: Developers frequently encountered 403 PERMISSION_DENIED errors when integrating Model Armor with services like Apigee, highlighting gaps in documentation regarding required IAM roles and permissions.
Implementing Changes Based on Insights
The insights from this collaboration led to a thorough revision of Model Armor’s documentation:
- Code Samples: Numerous tested, copy-pasteable code samples were added to facilitate search-first workflows.
- Confidence Level Matrix: A new technical reference was introduced to clarify the trade-offs between different filter levels, with explicit recommendations to minimize false positives.
- Integration Guides: Updated guides now clearly outline necessary IAM roles for smooth deployments, particularly for Apigee and Gemini Enterprise Agent Platform.
- Enhanced Documentation: The documentation now includes more in-depth explanations of enforcement methods and their real-world applications.
Value of Customer Engagement
Engaging directly with customers enabled Google Cloud to align technical accuracy with operational needs. This collaborative journey is designed to ensure that Model Armor effectively supports users in securing their generative AI workloads.
For those interested, the updated documentation for Model Armor is now available for exploration.
