Free guide · Claude 2.0 series, 4 parts
Claude changed a lot this autumn: new models, documents and designs inside every chat, dashboards, a new agent app and new ways to let it work on its own. People call it "Claude 2.0". It's a nickname, not a new model.
This page explains every idea in plain words, in four parts that match our four Reels. You don't need to code for any of it. At the end there's a short glossary and a checklist for where to start.
Accurate as of the end of October 2026. Plans, prices and features change, and some features depend on your plan, your country or a waitlist. If you can't see something, that's usually why.
Claude is a family of models. Haiku is the small, fast, cheap one. Sonnet is the everyday workhorse. Opus takes the long, messy jobs. Right now they are all version 5.5. Think of them as a team: you wouldn't send your most senior person to answer every quick email. A simple rule: a clear, well-defined job goes to Sonnet; an open, vague one, like a 90-day plan, goes to Opus. The free plan has Sonnet and Haiku, but not Opus.
Above Opus sits Fable (currently Fable 5.1), built for demanding reasoning and very long tasks that Claude does on its own. It's expensive: pay-per-use costs about two and a half times what Opus costs, and on the Max and Ultra plans it can use up to half your weekly limit. For most business work, Opus is plenty. There's also a command, /advisor, that lets Claude check in with a stronger model like Fable at key moments without running the whole job on it.
A token is a small chunk of a word. Everything is measured in tokens: what you pay and how fast you reach your limits. The context window is how much Claude can hold in mind at once. In Claude Code, the current models hold 1 million tokens, roughly 555,000 words. Most tokens go on reading, not writing, so the files and instructions you give Claude matter more than how long your prompt is.
Effort is how hard Claude thinks before it answers: low, medium, high, extra high or max. The default is medium. Change it with /effort or in the bottom-right corner of the desktop app. More isn't always better: max costs more, takes longer and can overthink. Keep medium for everyday work and go higher only for genuinely hard jobs.
Two limits run at the same time: a 5-hour session limit and a weekly limit. Chats in the app and work in Claude Code share the same allowance. The plans: Free ($0), Pro ($20 a month), Max ($100, five times Pro's usage) and Ultra ($200, ten times Pro's weekly usage; verified Ultra accounts no longer have the 5-hour limit). Max and Ultra also include API credits, which are for building your own apps on Claude, not for working in the Claude app. To see where you stand, type /usage or click the small circle in the bottom-right corner of the desktop app.
Try it today: type /usage to see your limits, then /effort to check which level you're on.
The desktop app has three tabs: Chat, Code and Cowork. Here's what each one is for, plus the other places Claude now works.
The Claude most people know: you type, it answers. Chat can now also make documents, slides and designs inside the conversation, and even the free plan has web search, memory, connectors and artifacts. Good for looking something up, thinking an idea through or a quick draft.
The Code tab is Claude Code. Don't let the name scare you: it works inside a folder on your computer, so it can read your files, create and edit them, move them around and run commands. It's less of a chatbot and more of an assistant that does the work, and plenty of people with no coding background run their business from it. Included from the Pro plan up.
Anthropic describes Cowork as Claude Code for the rest of your work, everything that isn't code. You give it a goal, it works across your files and tools, and you come back to finished work that only needs checking. For example: going through a folder of contracts and checking them against your standard terms, or building one spreadsheet from several regional reports.
Each conversation in the Code tab is a session with its own history and its own folder, and you can run several at once: one researching, one writing, one building. Cmd+N opens a new session, and split view shows two side by side. With remote control, your computer shows up on your phone, so you can check on your sessions from anywhere.
In the middle of a big task and want to ask something without throwing it off course? Press Cmd+; or type /btw (short for "by the way"). The side chat can see everything in the session, but nothing you say there goes back into the main conversation. In the desktop app, side chats aren't saved, so once you close one, it's gone.
Anthropic's new agent, in an app of its own. You message it the way you'd message a person, by typing or talking, and it acts in the apps you've connected. Each morning it can scan your email and calendar, flag what needs you and start drafting replies, and during the day it follows up on threads that went quiet. It asks before it sends a message, makes a payment, signs a document or deletes anything. For now it's a research preview with a waitlist, US only, on the Pro, Max and Ultra plans.
Anything Claude makes that opens next to the chat: a document, a deck, a design, a dashboard or a small interactive tool. Artifacts start private, and nobody sees them until you share them. Useful for a team presentation, an internal tool or a trip plan you send to friends.
The three most common kinds of artifact. Docs are documents, slides are presentations, and design is for visual work like landing pages and app mock-ups. All three are out of beta and on every plan, including free. Design used to be a separate site; it now lives inside your chats, and the old standalone site (claude.ai/design) closes in December 2026, so move anything you keep there.
Ask a question about your business in plain words, and Claude pulls the answer from your data, builds a dashboard and keeps it updated as the data changes. It connects to sources like BigQuery, Snowflake, Databricks and Salesforce, and you can click any number to see the exact query behind it. In beta, on paid plans.
Turns a prompt into a short animated explainer that plays like a video and downloads as an MP4. The animation is built with code, not with a video-generation model, so there's no AI footage and no fake people. In beta, on the Team and Enterprise plans.
Make me a one-page [price list / weekly schedule / checklist for onboarding a new client] for my business, as an artifact. My business: [what you sell, and to whom]. Keep it to one screen and easy to read on a phone. If you need something you don't have, ask me instead of making it up.
Claude can do a lot, but it's only as good as what it knows about you and your business. Every new conversation starts from zero, so this part is about information that stays with it.
Memory is how Claude carries things from one conversation to the next. In the Claude app it's on every plan, even free. In Claude Code there's auto memory: Claude writes its own notes from your corrections and preferences and reads them at the start of every session. The more you work with it, the better it knows how you work.
If memory is what Claude writes down, CLAUDE.md is what you write down: a plain text file with your instructions and rules that Claude Code loads at the start of every session. A good one says who Claude works for and where everything lives (contacts in this folder, brand files in that one). Keep it short, under about 200 lines, because it's read every single time.
A skill is a saved set of instructions for one specific job, in a file called SKILL.md (.md just means a simple text file with headings and bullet points). Claude uses a skill when it sees that it's relevant, or you run it yourself by typing / and its name. Skills cost almost nothing to keep, because Claude only reads a short description of each one until it needs it. The rule: if you've typed the same request twice, make it a skill.
Connectors plug Claude into the apps you already use, like Gmail, Google Calendar, Drive or Slack, so it can read and act there itself instead of you copying and pasting. Many of them run on MCP (Model Context Protocol), a standard that lets two pieces of software talk to each other. Connectors are on the free plan too.
A plugin is a bundle you install in one go: skills, commands, connectors and agents, packed together. Anthropic has a directory of plugins, companies release their own, and a team can share one plugin so everyone's documents come out consistent and on brand.
A mod is a plugin that changes how Claude Code itself works: new commands, new tools, a panel next to the chat or small changes to the interface. You don't need to code one. Describe the mod you want, and Claude builds it.
# Who you work for You work for [your name], who runs [business name], selling [what] to [whom]. Your job: save time on [the two or three jobs that come back every week]. # How to work - When something is unclear, ask one short question. Don't guess. - Never send, publish, pay for or delete anything without my OK. - Write in plain words. [Any other tone rules] # Where things live - Clients and contacts: [folder] - Brand files (logo, voice, offers): [folder] - Invoices and money: [folder] - Finished work: [folder] # Main skills - [skill name]: [when to use it]
These decide how much Claude asks before it acts. There are five. Manual asks before it edits files or runs commands. Accept edits lets it change files but still asks before most commands. Plan has it look around and propose a plan without changing anything. Auto works without the routine questions, while a background check makes sure every action matches what you asked for. Bypass permissions skips almost every question; leave it alone unless you know exactly why you need it. Since August, Auto has been the default for new sessions on the Pro, Max and Team plans. Nervous? Start in Plan, read the plan, then let it run.
A subagent is a helper Claude sends off to do one specific job. It works in its own context window and reports back only a short summary, so the messy research doesn't crowd your main conversation or burn through your tokens.
The word means two things. In the Claude app, a project is a folder for related chats, with its own files and instructions (our brand brain guide is built on one). In Claude Code, a project coordinates a whole piece of work: you say what needs doing, and it starts a separate thread for each task. Those threads usually run in the cloud, in parallel, so the work carries on after you close your laptop and you can check on it from your phone. The difference from a subagent: a subagent is a quick helper inside one session, while a project can run a full session for every task. Claude Code projects are in beta on the Pro, Max and Ultra plans and still rolling out.
Normally Claude does one turn and waits for you. With /goal you give it a finish line, and it keeps working, turn after turn, until that condition is met. Make the finish line impossible to misread, or it can get stuck going round in circles.
/goal Go through every invoice PDF in the invoices folder and build one spreadsheet with client, amount, date and paid or unpaid. Done when every PDF appears in the spreadsheet and the totals match the PDFs.
This is how Claude works when you're not there. In the desktop app you can set a task to run on a schedule, like every Monday or every morning. Routines are the same idea, but they run in the cloud and can start on a schedule or when something happens in another tool. You can set one up in plain words.
Set up a scheduled task for every Monday at 8:00: read last week's emails from customers, list every one I haven't answered yet, and draft a short reply to each. Don't send anything. Leave the drafts for me.
The "brain" Claude runs on. Haiku is fast and cheap, Sonnet is for every day, Opus is for big jobs.
A small chunk of a word. It's how price and limits are measured.
How much Claude can hold in mind at once.
How hard Claude thinks before it answers. Medium is enough for most work.
One conversation in Claude Code, with its own history and folder.
Something Claude makes that opens next to the chat: a document, deck, design, dashboard or tool.
A link between Claude and an app like Gmail or Google Calendar.
The shared language Claude uses to talk to other software.
Saved instructions for one job, which Claude follows the same way every time.
Your instructions file, which Claude Code reads at the start of every session.
A bundle of skills, commands, connectors and agents you install in one go.
A plugin that changes Claude Code itself, like a new panel or command.
A helper that works separately on one job and reports back a short summary.
How much Claude asks you before it acts.
A finish line. Claude keeps working until it gets there.
A scheduled task that runs in the cloud.
You don't need all 26 on day one. If you only do three things today, make them a short CLAUDE.md, one skill for the job you repeat most and a few tries with /goal. The full list, in order:
Learn one part a week.
That's how our Reels run: one part each week in November. Come back to this page after each one and tick off what you've done.
Next step
Ready to put this to work? Build your first AI agent with a single file.