What it is, how it differs from the other ways to use Claude, and the basics you need before we start.
Chat, Cowork, and Code are the three tabs in the Claude desktop app, and the ones we use today. Cowork and Code can also run in the cloud, on Anthropic's servers.
A conversation. You ask, Claude answers, and you go back and forth. It works with what you type or upload. It doesn't reach into the folders on your computer.
You give Claude a goal and, if it needs them, a folder of files. It makes a plan, works on its own while you do something else, and hands back finished work like a document, a spreadsheet, or a tidied folder. No terminal needed. When it works on your own files, the desktop app has to stay open.
Your app may show Chat and Cowork in one message box. Anthropic is rolling that out gradually, so your screen might not match a teammate's.
Claude works with you, step by step, inside a project folder. It reads and edits the files, runs commands, and shows you each change as it goes. That makes it the right fit for projects where changes get tracked and checked, which is why we're using it today.
Anthropic makes other Claude products too. Claude Design creates visual work like prototypes, slide decks, and one-pagers. Claude Managed Agents is a hosted service for developers who run Claude as an agent. We won't use either today.
This holds for Cowork too. Both work from real files and real steps.
No pasting and no screenshots. Claude opens the latest version of whatever is in the folder, so it works from what's actually there.
Ask for an outcome, like "update my card and publish it." Claude works out the steps, does them, and checks in with you along the way.
What Claude changes gets saved as real files. Close the window and everything is still there.
A file called CLAUDE.md holds standing instructions, and Claude reads it at the start of every session. In Code it lives in the project itself, so everyone on the team gets the same rules.
Our work is a handful of people touching the same client project. Code is built for that.
Each client engagement, or any topic we work on together, gets its own repo, with notes, working documents, and the updates we share with the client in one place. Pull before you edit and you start from the current version, so nobody works from an old copy.
When Claude changes a file, you get a diff that shows exactly what changed. You read it before anything goes to a client, like a project update. We practice this today with a preview on your own computer.
Every client repo carries a CLAUDE.md with the house rules. Our shared Firefly skills for voice, branding, and the steps we repeat on every engagement come from our plugin. Whoever opens the project gets the same standards, so the update sounds like Firefly.
Every change is kept with who made it and when. If we save each update when it goes out, we can show exactly what a client was told and when. For any change, we can see what changed and who changed it, and go back if we need to.
We'll tour these live on the call. This is the short version.
Pick Local or Cloud, then the folder to work in. Local works on a folder on your computer, and that's what we use today. Cloud runs on Anthropic's servers, keeps going after you close the app, and saves its results on a separate branch for you to review. Next to the send button you'll also see a model picker. We cover that in part two.
Claude works inside the folder you pick. It reads and edits files there and can run commands in it. Next to the send button, the permission mode sets how much Claude asks first. Manual asks before every edit or command and shows you each change. Auto lets Claude run without routine prompts, while a background check blocks risky actions. Plan proposes an approach and edits nothing. Your app may start in any of them, and we'll pick one together on the call.
Type / in the message box to see commands and saved shortcuts. Type @ and a file name to point Claude at one exact file.
Enough for day one.
Every session works inside the folder you pick when you start it. Pick the project folder, like team-manual, so Claude sees the right files and the project's house rules.
Write the way you'd message a colleague. You never type code or git commands. Say what you want and Claude Code does the work.
Try "What are you about to do?" or "Walk me through your plan first." Claude explains, and you decide.
Click the stop button, or type a correction and press Enter to steer without stopping. Depending on your permission mode, Claude may also ask your OK before it edits a file or runs a command.
When Claude changes something, look at it before you accept it. Ask Claude for a local preview, a private copy of the site that only runs on your computer, and check the change there. Hands-on shows you how.
"Update my role in team/kim.md" gets a better result than "fix my profile." The more exactly you point, the better Claude does.
You'll copy the team manual to your computer, edit your card, and watch the live page update.
Go to Hands-on