Module 3 · Page 1 of 7 · 10 minutes

Why move beyond the browser?

As we have seen do far, a browser is a great place to talk with AI. But to let an agent work on a real project, it often needs a workspace where it can inspect files, use tools, create outputs, and test its work.

An agent needs a workspace

For the kinds of projects we will build in this course, an agent may need to:

inspect a folder and understand how its files fit together,
read several documents, spreadsheets, or data files,
create and edit files directly in the project,
run tools or commands to perform work,
check whether its output actually works, and
keep project instructions and memory alongside the work.

That usually means giving the agent controlled access to a local workspace. This does not mean giving it unrestricted access to your entire computer. You choose the project, tools, and permissions it can use.

Two common ways to work locally

There are two common ways to put an AI agent closer to your files and tools: use a desktop app, or use a terminal-based agent.

Ecosystem	App	Terminal / local agent
Anthropic	Claude desktop app	Claude Code
OpenAI	ChatGPT and Codex app	Codex CLI
Google	Antigravity IDE	Antigravity CLI
OpenCode	—	OpenCode

Definition IDE (Integrated Development Environment): an app where you can see and edit your project files, run your work, and use related tools in one place.
CLI (Command-Line Interface): a text-based way to control a program by typing commands instead of clicking buttons.

Why learn the terminal?

Desktop apps are often easier to start with. They give you familiar buttons, menus, previews, and permission prompts. But terminal-based agents offer several important advantages once you know how to use them.

Advantage	What it means in practice
More portability	Only a handful of major AI companies offer polished desktop agent apps. The terminal gives you access to a much broader ecosystem of models and providers. Once you learn the workflow, you can switch among Claude Code, Codex CLI, OpenCode, local models, and other providers without relearning a new app each time. You are less dependent on one company’s interface, subscription, or product decisions.
More transparency	You can often see details that graphical apps hide or summarize, such as token usage, commands being run, and other low-level activity.
More flexibility	Terminal agents expose more of the configuration directly: model choice, providers, tools, MCP servers, permissions, and other settings can often be changed without rebuilding your workflow.
A clearer work environment	You can more easily inspect exactly which folder the agent is working in, what files it can access, what tools are available, and which instruction, memory, and configuration files are active. Desktop apps often deliberately hide or abstract more of this setup behind the interface, making the system easier to use but less transparent.

Tip This does not mean terminal tools are always better. Depending on the task, a desktop app — or even the browser — may be easier and more convenient. If you are brainstorming, reviewing a document, working with a small number of files, or doing a one-off task, the extra setup and visibility of the terminal may not add much. The terminal becomes more valuable when the work is more complex, repeated, or tool-heavy, and when you want more control over the agent's environment.

Required Setup for this course

In this course, we will primarily use Claude Code through the terminal. To make file access easier, we will use the terminal inside an IDE called VS Code. This may look less familiar than a browser or desktop app at first, but it makes the agent's workspace easier to understand, and is a more transferable skill.

If, for technical reasons, you cannot get this setup to work, you may use the Claude app instead. The main ideas and workflows in this course will still apply, with the caveats described in the table above.

Next · Page 2 of 7

Setup overview

→