Claude 101: Chat vs. Cowork vs. Claude Code Explained

Claude is the AI I use for about 80% of everything I do: writing, thinking, operations, client work, the automations that run my agency. And the thing that confuses new people most is that Claude is really three tools wearing one name. Same brain, three surfaces, and knowing which surface to open for which job is most of the learning curve.

This is the full map: what each surface is, what to actually do in your first week on each, and the setup steps that separate "eh, it's fine" from "how did I work without this."

Before anything: the two setup steps that change everything

Do these before judging Claude at all. Skipping them is why most people's first impression is mediocre.

1. Connect your tools. Out of the box, Claude knows the general internet and nothing about you. Go to Settings → Connectors and plug in your email, calendar, Google Drive, and meeting notetaker. These connections run on MCP (Model Context Protocol, the standard plug that lets an AI read your other apps), and each takes about a minute. Connected, Claude stops being a clever stranger and starts being a colleague who knows your business. The full walkthrough is in the MCP starter guide.

2. Give it your rules. Claude can't know things you never told it: how you write, what you never say, what "done" looks like to you. Write those down once, somewhere it reads every time (project instructions, or a rules file in your file setup). The 20-minute method, dictated out loud, is in the rules doc guide.

Fifteen minutes, total. Now the three surfaces.

Surface 1: Chat (the conversation)

The one everyone knows: you ask, it answers. This is where I think out loud, draft, process a hard meeting, pressure-test a decision, and ask the questions I'd otherwise sit on.

What your first week in chat should include:

Talk instead of typing. Hit the voice icon and brief it like a person. Spoken prompts carry more context, and context is what makes output good.
Run one connected ask that would be impossible without your tools plugged in:

Look at my calendar for this week. For each external meeting, check my
email and meeting notes for context and brief me: who they are, where
we left off, what I owe them.

Process something real. After your next frustrating meeting, open a chat and talk it out: what happened, what you wish had gone differently, what you actually want. Then ask it to turn the vent into a plan. This is the use nobody markets and everybody keeps.
Use projects (or a dedicated chat per client/area) so context accumulates instead of starting from zero every conversation.

Open chat when: you're thinking, drafting, deciding, processing, or asking. Anything conversational.

Surface 2: Cowork (the coworker)

Cowork is where Claude stops answering and starts doing. You hand it a task and access to your stuff, and it works through the steps: reading sources, producing files, and (the killer feature) running on a schedule without you.

The flagship example is my daily meeting briefing. Every morning it reads my calendar, pulls context from email and past meeting notes, writes an agenda for every meeting, saves them to Drive, and messages me the digest. I set it up once, by voice, describing the output I wanted. Claude asked a few follow-up questions and built the thing itself. That's the posture shift Cowork teaches: you don't have to be the architect. You're the designer and the director.

Your first Cowork move, concretely:

Open Claude → Cowork → Scheduled, and create a new task.
Describe a recurring chore you already do manually. Start with the meeting briefing (full copy-paste prompt in the daily briefing guide) or an end-of-day meeting-notes sweep (in the meeting-notes automation guide).
Run it once manually, correct what's off, tell it to remember the corrections, then trust the schedule.

Open Cowork when: the job has steps, produces files, or should happen on a schedule while you sleep.

Surface 3: Claude Code (the builder)

Claude Code is the most powerful surface and the most misunderstood, because the name scares off the exact people who'd love it. Yes, developers use it to write software. At its heart, though, it's Claude with full access to a folder of files: able to read, organize, create, and edit them in bulk, and to remember how your whole system fits together.

That's why it's what I point at my second brain. Asks like:

Read the last six months of this client's meeting notes and pull every
story they've told about why they started the business.

Draft next month's content plan from this client's strategy doc and
last month's top performers, following our content rules.

are Claude Code jobs: they read across dozens of files at once and follow rules stored in the folder itself. It also does the housekeeping (filing notes, creating client folders with the standard structure, keeping everything committed to GitHub with clean messages), so the brain maintains itself.

And the part that matters for non-technical readers: Claude Code now runs inside the Claude app. No terminal, no developer setup. If you can point it at a folder, you can use it.

Open Claude Code when: the job is about many files at once: your second brain, a content library, a reorganization, anything where "read all of it first" is part of the task.

The one-line cheat sheet

Talking? Chat. A task with steps or a schedule? Cowork. A folder full of files? Claude Code.

And the honest adoption path: week one, chat with connectors. Week two, your first scheduled task in Cowork. The month your files are worth reading in bulk, Claude Code. Don't rush the ladder; each rung makes the next one obvious.

Mistakes to avoid

Judging Claude from a blank, disconnected chat. That's judging a new hire you refused to onboard.
One giant chat for everything, forever. Context gets muddy. One project or chat per client or area of work.
Correcting instead of training. When output misses, say why and tell it to remember. Silent fixes teach it nothing, and you'll give the same note forever.
Automating a chore you've never done manually with it once. Run it live first, fix the misses, then schedule it.
Ignoring Claude Code because of the name. It's the surface built for exactly the second-brain system everything on this site describes.

FAQ

Is Claude better than ChatGPT? I use both. Claude carries about 80% of my work (writing, thinking, operations, everything connected to my business data), and ChatGPT covers the rest, mostly image generation. Different jobs, not a rivalry. The ChatGPT 101 covers its side.

What's the difference between Claude Cowork and Claude Code? Cowork is task-shaped: give it a job with steps, possibly on a schedule, get a deliverable. Claude Code is folder-shaped: it lives inside your files and works across them in bulk. My daily briefing is Cowork; my second brain is Claude Code.

Do I need Claude Code if I don't code? If you build a second brain, yes, eventually. It runs inside the Claude app now, no terminal required, and it's the surface designed for many-files work.

Which should a beginner start with? Chat, with your calendar and email connected and your rules written down. One week of that teaches you more than any tutorial, and it makes the other two surfaces make sense.

How much does Claude cost? Plans change too fast for this page; check claude.com for current pricing. The free tier is enough to evaluate chat; the paid tiers are where Cowork scheduling and heavy Claude Code use live.