Gemini Enterprise Agent Builder: The Architectural Standard

The Intent: From Fragmentation to Collaboration

Enterprise AI development often begins in fragmented environments like Google AI Studio, where designers create powerful but isolated “vibe-based” prototypes. While useful for exploration, these one-off experiments suffer from scaling bottlenecks, behavior assumptions, and extreme handoff friction. The intent of the Gemini Enterprise Agent Builder was to break these silos and move the team from “Mockup Guesswork” to Exact Functionality via a shared, high-governance engineering ecosystem.

The Human Multiplier: Bridging the Designer Gap

When the design team reached their collaboration ceiling, I leveraged my separate initiative—spearheading the migration of the Cloud UXE organization to Internal GitHub Enterprise (GHE)—to create a dedicated home for the Agent Builder.

• The Migration: I transitioned the existing AI Studio prototypes into a structured repository powered by Jetski.
• The Onboarding: I authored the definitive onboarding protocol (go/cloud-ai-ux-workflow) and acted as the human bridge, teaching non-technical designers how to branch, commit, and open Pull Requests.
• The Culture Shift: This transformed designers from “isolated prompters” into “collaborative architects” capable of high-velocity iteration on a single source of truth.

The Infrastructure Multiplier: Velocity via CI/CD

To restore the “one-click deploy” simplicity designers loved in AI Studio, I built a custom infrastructure stack that de-risked complex behaviors for PM and Eng partners.

• Automatic Deployment: Created a Google Cloud Build pipeline where any merge to the main branch automatically triggers a container build and deployment to Cloud Run, making changes live instantly.
• The Preview Sandbox: I architected a system where suffixing a branch with -preview automatically generates a unique, isolated URL.
• UXR & Stakeholder Loops: These isolated environments allowed UX Researchers to conduct user studies on specific unreleased features and provided stakeholders with “frozen” demo URLs for Cloud Next, physically isolated from ongoing development.

Figure 1.0: Automatic Preview Branch Deployment & Sandbox URL Generation

The Mandate: Repository Governance & Oversights

My Director established a clear mandate: I was responsible for the architectural integrity of the Agent Builder repository. I implemented a strict Governance Layer:

• Branch Protection: Enforced invariants where direct commits to main were blocked, mandating a formal PR workflow.
• Ruleset Hardening: Managed and updated the project’s behavioral rulesets to keep agentic reasoning aligned with evolving standards.

High-Fidelity Agentic Skills & Embedded Tooling

I developed specialized Jetski Skills and embedded a custom Tools Menu directly into the prototype to simplify engineering overhead:

• pr-review Skill: An automated workflow where a single command pushes changes and opens a PR with a high-fidelity summary of the diff posted as a comment for instant observability.
• Visual Inspector: A custom-built HTML inspector that allows designers to see color/typography tokens, edit properties in real-time, and compare components against Google Material standards.
• Annotate (Draw): A built-in tool for drawing over the prototype, with screenshots automatically saved to the clipboard for instant pasting into Jetski—because a picture says more.
• Commenting: Implemented a Figma-style commenting system (powered by Firebase) with @mentions, allowing for direct feedback on specific UI elements in specific states.

Strategic Interventions: Marketplace Demo & A2UI

While managing the infrastructure, I engineered the Cloud Next Marketplace Demo feature using A2UI (Agent-to-UI) protocols. This allowed the autonomous agent to dynamically render and update UI components based on its internal reasoning state, demonstrating the future of deep agent-UI integration.

Figure 1.1: Cloud Next Marketplace Demo — A2UI Orchestration

The Outcome: Scaling Beyond the Team

The success of this collaborative environment caught the attention of CoreUX, Corp Eng, and Google Material. Their requests for replication led to my next mandate: distilling this entire workflow into the Cloud UX AI Blueprint to scale high-fidelity agentic development across Google.

Artifact Proof: Enterprise Agent Orchestration

Figure 1.2: Enterprise Agent Orchestration Preview (Forensic Trace)

Figure 1.3: Visual Inspector — Real-time Material Token Auditing

Lead Architect: Ilteris Kaplan
Status: Platform Standard (Deployed)
Trace ID: GE-2026-ALPHA-01