From “chat with AI” to “AI that can do work”
Today you learn the new way many people build software: you describe the goal in plain language, and an AI agent helps create the app. You also meet the main types of tools you will use all week — an agent manager, an IDE, and a quick web builder.
The course’s message: your job is shifting from typing every line of code to explaining what you want and checking that the result is correct.
A chatbot mainly talks back. An agent can also do things: open files, run commands, call APIs, fix errors, and work toward a goal over many steps.
Chatbot = someone giving advice. Agent = someone who can also use your tools to finish the task.
You say what you want (“make a page with two buttons that trigger animations”) and the AI builds or edits the project. You review, test, and steer — like directing a builder instead of laying every brick yourself.
There is a range: casual vibe coding (quick experiments) → agentic engineering (same idea, but with clear rules, tests, and safety).
Model = the LLM brain (e.g. Gemini). Harness = everything around it: permissions, tools, sandboxes, memory, guardrails, and how tasks are split into steps. A smart brain without a good harness is unreliable; a good harness makes the brain useful.
The normal path of building software: plan → design → build → test → deploy → maintain. With vibe coding, the build step can be much faster, so planning and testing become even more important.
A workspace for coding: files, editor, terminal, run/debug tools, and often an AI assistant built in.
Putting your app on a server so other people can open it in a browser. A prototype on your laptop only works for you; deployment makes it public (or available to your team).
Learn the type of tool first. The course picks one example per row; you can swap others later.
| Category | What it does | In this course | Same idea elsewhere |
|---|---|---|---|
| Agent manager / agentic IDE | Run and supervise AI helpers on your projects; chat, approve actions, manage multiple tasks. | Antigravity 2.0 (standalone app) | Cursor, Windsurf, GitHub Copilot Workspace, Devin-style agents |
| Code editor with AI | Write and edit code in files; AI suggests or edits inside the project. | Antigravity IDE | VS Code + extensions, Cursor, JetBrains AI, Xcode |
| LLM / model provider | The language model that powers reasoning and code generation. | Gemini (via AI Studio or API) | OpenAI GPT, Anthropic Claude, Meta Llama, Mistral |
| Prompt-to-app builder | Describe a small app in language; get a runnable prototype fast. | Google AI Studio “Build” | v0, Lovable, Replit Agent, Bolt.new |
| Cloud hosting | Run your web app on the internet without managing your own physical server. | Cloud Run | Vercel, Netlify, AWS Lambda, Azure App Service, Railway |
| Agent framework (introduced more on Day 3) | Code structure for building agents with workflows and tools. | ADK 2.0 (mentioned today) | LangChain, CrewAI, AutoGen, OpenAI Agents SDK |
What you will get from it: A spoken overview of why software building is changing — from typing syntax to stating goals — and how autonomous coding agents fit in. Good “trailer” before the whitepaper.
Main ideas in plain language:
Concept practiced: Using an agent manager — create a project, start conversations, set what the agent is allowed to do, and work with multiple folders.
Transferable skill: Any agentic IDE has “project scope + permissions + chat threads.” Learn that pattern once.
Codelab 2: Build a Web App in AI Studio and Deploy to Cloud Run codelabs.developers.google.com/deploy-from-aistudio-to-runConcept practiced: Prototype → deploy — describe an app, test it, publish to the cloud so it has a public URL.
Transferable skill: Same flow on other stacks: prompt builder → test → host on Vercel/Netlify/etc.
Course hosts and Google guests discuss how education and jobs may change, risks of AI-written code, and demos. Useful if you want real-world context beyond the readings.