Five Personas Across the Software Development Lifecycle

▶ Watch (1:13)

Nardini frames the talk around a standard enterprise product team: PM, UI/UX designer, software engineer, security engineer, and data analyst. Each persona contributes at a different stage of the software development lifecycle. Claude Code’s role is to augment all five. The demo builds a live feedback app from scratch, with the audience rating the session in real time by the end. Every step from wireframe to deployed API runs inside Claude Code.

Per-Token Billing and No API Keys to Rotate

▶ Watch (5:00)

Running Claude models on Google Cloud charges per token with no message cap. Production workloads can reserve throughput via provisioned capacity. Setup uses application default credentials, so there are no API keys to rotate. Data stays inside your Google Cloud project. Models run on global and regional endpoints. Claude Code recently shipped a wizard that detects your project and region, lists which Claude models are available, and lets you pin the ones you want.

From Coffee Sketch to Three-Page Interface with Plan Mode

▶ Watch (7:37)

The PM step starts with a hand-drawn sketch fed to Claude Code. In minutes, Claude renders an HTML wireframe from the image. The UI/UX step extends that wireframe using plan mode, where Claude proposes all components before writing any code. The developer reviews the plan, accepts, then implementation runs. The demo simulates a Figma design spec arriving via MCP server and produces a three-page app: a landing page, a thank-you page, and a live dashboard.

Developer Knowledge API, Google Cloud Skills, and Parallel Sub-agents

▶ Watch (12:09)

Google Cloud released two components here: the Developer Knowledge API with an MCP server that feeds fresh GCP documentation into Claude Code, and Google Cloud Skills that cover individual implementation blocks like deploying to Cloud Run or connecting Cloud Run to Firestore. With both configured, three sub-agents ran in parallel: one built the API, one built the Firestore ingestion pipeline, and one built the BigQuery dashboard. Each sub-agent also ran tests on its component after finishing.

Security Review, Live Deployment, and Real-Time Audience Feedback

▶ Watch (17:08)

Before deploying, Nardini ran Claude Code’s built-in security review. It found a potential issue, fixed it automatically, then deployed the backend API to Cloud Run. The app went live on GCP with Firestore collecting raw responses and BigQuery storing them for analytics. A feedback analyzer button called Claude on Google Cloud to summarize audience comments in real time. During the demo, one audience member gave the session a score of five.

Notable Quotes

We look at them. We are happy. Ivan Nardini · ▶ 10:33

prior knowledge of how to do that. Ivan Nardini · ▶ 14:01

you don’t need to know how to use Ivan Nardini · ▶ 22:04

Key Takeaways

  • Claude Code on Google Cloud uses per-token billing and application default credentials, no API keys needed.
  • Plan mode makes Claude propose a full implementation plan before writing any code.
  • Three sub-agents ran in parallel to build the API, ingestion pipeline, and dashboard.
  • The built-in security review found and fixed a vulnerability automatically before deployment.
  • MCP servers for BigQuery and Looker let you query and visualize data without prior platform knowledge.