Building the App
Plan approved, now we build. Rather than letting a single agent grind through the whole thing, the main agent acts as an orchestrator and hands work out to sub-agents running in parallel. This lesson covers how to set that up, how to keep an eye on it while it runs, and how to properly check the app yourself once the agents go quiet.
What you'll walk away with:
- Handing the implementation plan to the agent and instructing it to build without stopping
- Using the goal command in Claude Code and Codex to keep the agent running through to completion
- Delegating to parallel sub-agents instead of one agent doing everything in sequence
- Pinning the model your sub-agents use, so a weaker one doesn't get picked automatically
- Why orchestration protects the context window, and the token cost that comes with it
- Watching the main agent scaffold the project and hand off tasks
- Reviewing the file changes and reading the diffs each agent produced
- Checking in with the main agent mid-build for a status update or a course correction
- Confirming the Postgres database is running in Docker Desktop
- Watching the agent open a browser and test its own work
- Starting the development server and opening the app
- Testing it by hand: sign-up, password strength, QR generation, colour and logo customisation, export sizes and formats
- Verifying the dashboard, plus search, edit and delete
- Signing in with a second account to confirm users can't see each other's data