CHAPTERS
Goal of the session: build and deploy with Claude on Google Cloud
Ivan Nardini frames the talk around a gap: many people use AI tools to code, but few use them end-to-end to build and deploy on Google Cloud. The session’s goal is to demonstrate an end-to-end workflow using Claude on Google Cloud and Claude Code.
Enterprise SDLC personas and how Claude Code augments each role
The talk introduces a simplified enterprise software lifecycle with five personas: PM, UI/UX, software engineer, security engineer, and analyst/marketer. Claude Code is positioned as an “augmentation layer” across all these roles, accelerating handoffs and execution.
Setting up Claude Code to use Claude models on Google Cloud (ADC + setup wizard)
Ivan explains the simplest path to run Claude models on GCP from Claude Code using Application Default Credentials (ADC). He highlights a setup wizard that detects project/region, lists available models, and allows pinning models for development.
Why run Claude on Google Cloud: pricing, throughput, governance, and regions
He outlines the motivations for using Claude via Google Cloud rather than other access methods. The benefits emphasize enterprise needs: pay-per-token, optional provisioned throughput, simpler credentialing, policy control, and regional/global endpoints.
PM workflow: turning a napkin sketch into an app prototype with Claude Code
Wearing the PM hat, Ivan shows how a rough drawing can be converted into an interactive prototype quickly. The point is to reduce traditional back-and-forth with design to get a first version stakeholders can react to.
UI/UX workflow: using Plan Mode to create production-ready pages
As the UI/UX persona, Ivan upgrades the prototype into a more polished interface with multiple pages (landing, thank-you, dashboard). Plan Mode is used so Claude proposes steps before writing code, enabling review and alignment with design standards (e.g., via Figma/MCP).
Software engineer workflow: designing a cloud-native backend with fresh GCP docs + skills
Ivan shifts to the engineer role, focusing on deploying to Google Cloud without requiring deep prior GCP expertise. He introduces the Developer Knowledge API (via MCP server) for up-to-date documentation and “Skills” to implement common building blocks reliably.
Reference architecture for the feedback app: Cloud Run, Firestore, BigQuery, Looker
He describes a practical architecture: a serverless API on Cloud Run, operational storage in Firestore, ingestion into BigQuery for analytics, and dashboards in Looker. The emphasis is that Claude Code plus MCP/skills can guide the design and implementation.
Parallel development with Claude Code subagents (API, ingestion, dashboard)
Ivan demonstrates Claude Code subagents to build multiple components in parallel, akin to running a sprint with multiple engineers. The workflow includes architecture drafting, API spec generation, parallel implementation, and automated testing steps.
Security engineer workflow: automated review, OWASP checks, least-privilege service accounts
Before deploying, Ivan introduces a security review stage, including common web security concerns and cloud IAM considerations. He uses a pre-built security review workflow in Claude Code to detect and fix issues and ensure a safer deployment posture.
Deploying to Cloud Run and live demo: collecting feedback + Claude-powered feedback summary
The application is deployed and shown running on Cloud Run with a live endpoint. Audience feedback updates the dashboard in real time, and a “feedback analyzer” generates a summary by calling Claude on Google Cloud based on collected comments.
Analytics and improvement loop: BigQuery + Looker, and how MCP servers help
Ivan returns to the final persona: analyzing usage and performance metrics to improve the product (e.g., response time KPI). While he doesn’t demo querying BigQuery or building Looker dashboards due to time, he points to MCP servers that enable these tasks through Claude Code.
Where to find MCP servers: Agent Platform registry + open-source Looker integration
He shows where Google Cloud catalogs supported MCP servers via the Agent Platform’s Agent Registry. He also mentions the open-source MCP Toolbox for Databases, which includes Looker integration and a documented quickstart.
Wrap-up: combining Claude Code components with Claude on GCP + code release
Ivan summarizes the two main takeaways: Claude Code components (skills, MCP servers, subagents) accelerate the SDLC, and Claude models on Google Cloud provide a seamless enterprise-friendly experience. He notes the code and quickstarts will be released and invites follow-up questions.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome