CHAPTERS
- 0:16 – 0:47
Audience poll & goal: coding with AI all the way to Google Cloud deployment
The speaker opens by gauging how many attendees used AI coding tools recently—and how many went further to build and deploy on Google Cloud. He sets the talk’s goal: show a practical path to improve that end-to-end workflow using Claude on Google Cloud.
- •Quick audience poll on AI coding tool usage
- •Second poll: who also built/deployed on Google Cloud
- •Framing the session as an end-to-end “do better” walkthrough
- 0:47 – 2:17
Enterprise software delivery needs many roles (and where AI can help)
He outlines a typical enterprise team involved in shipping a feature: PM, UI/UX, software engineer, security engineer, and analytics/insights. This establishes the lifecycle Claude Code can augment across roles and phases.
- •PM defines the feature idea and direction
- •UI/UX turns ideas into prototypes and screens
- •Engineers implement backend/core logic
- •Security review is required before shipping
- •Analytics closes the loop with usage insights
- 2:17 – 3:18
Demo plan: one feedback app, five hats, end-to-end lifecycle
The speaker introduces the running example: a simple feedback app used at the end of the session to rate the talk. He explains he’ll “wear five hats” to demonstrate how Claude models on Google Cloud support each stage.
- •Build a feedback app used live for session feedback
- •Demonstrate multiple roles with “five hats” approach
- •Show Claude Code + Google Cloud through the full lifecycle
- •Brief personal intro and partnership context
- 3:18 – 4:19
Getting started: Claude Code setup with Google Cloud credentials
He explains how to configure Claude Code to use Claude models hosted on Google Cloud, emphasizing a straightforward setup. Application Default Credentials and an onboarding wizard detect project/region, check model availability, and pin models for use.
- •Use Application Default Credentials (no manual key handling)
- •Wizard detects project and region automatically
- •Verifies which Claude models are available in the project
- •Pin models for the coding session
- 4:19 – 5:49
Why run Claude on Google Cloud: pricing, governance, regions, reliability
The talk lists practical reasons to choose Claude via Google Cloud: pay-per-token pricing, optional provisioned throughput, simplified auth, and strong enterprise controls. He also highlights regional/multi-region serving and platform-grade availability standards.
- •Pay per token; no message caps
- •Provisioned throughput available for extra capacity
- •No API keys to store/rotate; uses your GCP project controls
- •Data stays in your project; apply your org policies
- •Global/multi-region endpoints and high availability
- 5:49 – 7:54
PM hat demo: from napkin sketch to wireframe + GitHub PR
As the product manager, he shows generating a wireframe from a rough sketch using Claude Code. Even without Git expertise, the workflow can create and submit the output as a pull request to a GitHub repo.
- •Turn a rough sketch into a usable wireframe quickly
- •Use a role-guided prompt via a CLAUDE.md setup
- •Automate repo updates: create a PR from the generated work
- •Accelerate handoff from PM to UI/UX
- 7:54 – 10:28
UI/UX hat demo: planning mode to produce production-ready screens
Switching to UI/UX, he builds a more complete interface: multiple pages including landing, thank-you, and a dashboard view. He uses Claude Code’s planning mode to propose a spec first, then implements and submits changes via PR.
- •Create a multi-page UI (including a dashboard view)
- •Use planning mode to draft a spec before coding
- •Optionally integrate design sources (e.g., Figma) in real scenarios
- •Generate code quickly and submit another PR for review/merge
- 10:28 – 15:06
Software engineer hat: architecture + fresh docs via MCP, then parallel build with subagents
He introduces Google Cloud’s integration with the Claude ecosystem: official skills and a Developer Knowledge API exposed via an MCP server with frequently refreshed docs. Using this, Claude helps design the app architecture (Cloud Run, Firestore, BigQuery, Looker) and then implements components in parallel with subagents.
- •Use Developer Knowledge API (MCP server) for up-to-date docs/guides
- •Draft an architecture: Cloud Run API, Firestore DB, BigQuery ingestion, Looker visualization
- •Use official Google Cloud skills to drive implementation steps
- •Parallelize work with subagents (API, data pipeline, dashboard)
- •Generate API paths/spec and validate against documentation
- 15:06 – 17:07
CI/CD on Google Cloud: Cloud Build + Cloud Deploy to ship to dev (then promote)
After implementation and tests, Claude generates deployment components and pipelines. The PR merge triggers Cloud Build (CI) and Cloud Deploy (CD) to release into a development environment, setting up a controlled path toward production.
- •Automate tests as part of the build process
- •Generate CI/CD config for Cloud Build (CI) and Cloud Deploy (CD)
- •Merge triggers a pipeline with defined build and release steps
- •Deploy first to dev environment to validate before promotion
- •Result: app running on serverless infrastructure (Cloud Run)
- 17:07 – 20:10
Security engineer hat: custom review plugin for validation & least-privilege permissions
Before production, he runs a security review using a custom plugin tailored to organizational requirements. The plugin checks input validation and tightens service account permissions, produces fixes quickly, reruns tests, and opens a new PR for approval and promotion.
- •Use a custom plugin to encode company security requirements
- •Check for missing input validation and remediate issues
- •Review and reduce service account permissions (least privilege)
- •Run tests after changes; open PR with fixes
- •Promote release through Cloud Deploy after review/approval
- 20:10 – 24:37
Analytics loop: BigQuery + Looker dashboards via MCP servers (and wrap-up)
With the app in production, he shows how to analyze usage—like time-to-submit feedback—using a BigQuery MCP server, then publish results as a Looker dashboard link via a second MCP server. He closes by summarizing the Claude Code capabilities used and points to resources for learning more.
- •Collect product metrics from submitted feedback (UX improvement signals)
- •Query/analyze raw data in BigQuery using an MCP server
- •Turn terminal stats into a shareable Looker dashboard via MCP
- •Demonstrated Claude Code tools: plan mode, subagents, MCP, skills, plugins
- •Closing pointers: repo, quickstarts, and documentation
