Agentic Coding Workshop

Pre-work, about 15 minutes

Coding agent basics

This reading covers only what the setup and the workshop require. Design, examples and building happen in the workshop itself.

How Claude Code works, Claude channel on YouTube, 3 minutes
The Explore, Plan, Code, Commit workflow, Claude channel on YouTube, 3 minutes

Coding agents and project folders

A chat assistant answers in a conversation window. A coding agent such as Claude Code works inside a folder on your computer: it reads the files there, writes new ones, runs programs, checks the result and tries again when something fails. You describe the task in plain language, and the agent performs the steps. The folder is the project; everything the agent makes is saved there as ordinary files you can open, copy or delete.

At the start of every session Claude Code reads a file named CLAUDE.md in the project folder, if there is one. It holds standing instructions for that project, such as the purpose of the tool, the rules it must follow and the data it must never store. In the workshop you put your own spec in it.

Desktop app, terminal and file paths

In the workshop you use Claude Code in the Code tab of the Claude desktop app: you open a project folder, type requests in a box, and approve changes with buttons. No terminal is needed. The same Claude Code also runs in a terminal, a window where you type commands; there, cd followed by a folder name opens that folder, claude starts Claude Code, and /exit closes it.

A file path is the address of a file, for example Documents/test-site/index.html on a Mac or Documents\test-site\index.html on Windows. When the agent names a path, it is telling you where on the laptop a file is.

Permission requests

Claude Code asks before it takes certain actions, such as editing a file, running a command or using the network. Read each request before approving it. A request to delete files or to send data somewhere unexpected is a reason to decline and ask the agent to explain.

The permission mode controls how often it asks. In the desktop app it is set in the row under the request box: Manual asks before every change, Accept edits applies file changes and shows them for review, and Plan proposes an approach without editing anything, which is useful at the start of a task. In the terminal, Shift+Tab cycles through the modes, and /rewind restores the files Claude edited to an earlier point in the session; changes made by commands it ran, such as deleting or moving files, are not restored that way.

Result checks

Agents make confident mistakes. They can report that a task is finished when part of it failed, drop rows from a data file without saying so, or add features nobody asked for. Open what the agent built and try it yourself, and ask the agent to show the test it used. In the workshop, each spec ends with a line headed DONE WHEN, which states the check that tells you the tool works.

Data rule for the workshop

Use only records you make up; you can ask the agent to make up a few realistic records for testing. No client, field placement or employer data go into the agent.

A personal Claude account is suitable for low-sensitivity data only, and a prototype built in three hours has not been checked for the safeguards real client data require.

Request guidelines

You talk to a coding agent in ordinary sentences. Requests work better when they say:

If the result is wrong, say what you see and what you expected. "The table shows 2,105 rows but the file has 2,103" works better than "it's broken".

Glossary

Coding agent
An AI tool that works in a folder on your computer and can read, write and run files on its own, asking permission for some actions. Claude Code is one; Codex and Cursor are others.
Claude desktop app
Anthropic's app for Mac and Windows. Its Code tab runs Claude Code on a folder on your computer without a terminal.
Terminal
A window where you type commands. Terminal on a Mac, PowerShell on Windows. Optional for this workshop.
Project folder
The folder the agent works in. Everything it builds is saved there.
File path
The address of a file, such as Documents/test-site/index.html.
CLAUDE.md
A file of standing instructions that Claude Code reads at the start of every session in that folder.
Prompt
The request you type to the agent.
Permission request
A question the agent asks before it edits a file, runs a command or uses the network. In the desktop app it appears with buttons to approve or decline.
Plan mode
A mode in which the agent proposes a plan and changes nothing until you approve it.
Spec
A short written description of a tool: purpose, users, what it must do, what it must never do, and the check that shows it works.
Prototype
An early working version, built to test an idea before anyone relies on it.
Made-up records
Invented records with a realistic shape, used for testing in place of real client data. The agent can create them for you.

Videos (optional)

Official short videos from Anthropic's Claude channel, about 16 minutes in total.

Further reading (optional)

For web basics, HTML and CSS, and UI and UX design, see Learn more.