One Person, Five Roles: Shipping a Full App with Claude Code on GCP

1 min read
claude-codegoogle-cloudvertex-aimcpsubagentsdevopsci-cd
View as Markdown
Originally from vm.tiktok.com
View source

My notes

Watch on TikTok Tap to open video

Summary

A Google Cloud developer advocate demos how one person, wearing five roles (PM, UI/UX, software engineer, security reviewer, analyst), builds and ships a full feedback app on Google Cloud using Claude Code with Claude models hosted on Vertex AI. The talk maps each Claude Code feature (plan mode, subagents, MCP, skills, plugins) to a stage of the software lifecycle and walks the app from paper sketch to production deploy plus analytics dashboard.

Key Insight

Claude-on-Vertex vs API key

  • Auth via Application Default Credentials plus a wizard that auto-detects project/region and lets you pin available Claude models. No API key to store or rotate.
  • Billing is pay-per-token with no message cap; need more capacity, use Provisioned Throughput.
  • Data stays in your GCP project; models served multi-region/global endpoints for HA.

Feature-to-role mapping (the actual recipe)

  • PM role: CLAUDE.md rule turns a coffee-napkin sketch into a wireframe, auto-commits and opens a PR (PM never touches Git).
  • UI/UX role: plan mode converts wireframe into 4 production pages; Claude proposes spec before coding (could pull from Figma).
  • Engineer role: subagents parallelize, one for API, one for BigQuery pipeline, one for dashboard, simulating a team sprint with different models.
  • Security role: custom plugin runs an input-validation plus service-account permission review, fixes code, re-tests, opens PR.

Two new Google Cloud integrations worth noting

  • Developer Knowledge API plus MCP server: gives Claude fresh GCP docs/implementation guides refreshed every 24h, so it picks the right architecture (Cloud Run plus Filestore plus BigQuery plus Looker) without the dev knowing GCP.
  • Official Google Cloud Skills: handle actual deploy, generating a Cloud Build (CI) plus Cloud Deploy (CD) pipeline that triggers on PR merge, with dev-to-prod promotion gating.
  • BigQuery MCP plus a Looker/dashboard MCP turn raw feedback data into analytics and a shareable dashboard link on the fly.