Make the Most of Claude AI: From First Chat to Full Autopilot
Claude is the most powerful AI tool for work right now. This 2026 guide covers all features and how to use them to transform how you work. With steps and screenshots.
Claude is probably the most capable AI tool available right now for work. I've written about it a lot. Individual deep dives on Cowork, Skills, Plugins, Scheduled Tasks, Claude Code, computer use, subagents. Each one covering a specific feature in detail.
But I’ve never put it all in one place. It’s not that easy when the pace it’s evolving at is hard to keep up with, even for someone writing about it every week.
This week alone, Anthropic shipped two major updates: a 1 million token context window (that’s roughly 750,000 words in a single conversation) and inline visualizations, where Claude now draws interactive charts and diagrams right inside the chat.
And because it has so much to offer, it’s easy to miss a lot of it. Just start chatting and never explore what else is in there. Projects, skills, agentic work, connectors to your tools.
Features that can change how you work and it’s a waste not to use them.
So I put together this complete beginner guide. Not a feature list, but a path organized in levels, from getting started all the way to letting Claude work for you.
Whether you’re just getting started with Claude, switching from ChatGPT, or you’ve been using it for months, there are features in here you probably didn’t know existed. And if you want to make Claude even smarter, I also built an MCP that works right inside it, giving you access to all my prompts, image generation, and a place to save your own.
I'll walk you through all of them with steps and screenshots, so you can start using Claude like the top 10% of users.
Here’s what we’re covering:
Level 3: Let Claude work for you
Computer use + Dispatch: Claude works on your computer while you're away
Claude Managed Agents: how to build AI agents into your products
Let’s get into it.
Getting started with Claude
Step 1: Get Claude and pick your plan
First things first: you need to decide where you’ll use Claude and which plan makes sense for you.
Where to use Claude: browser, desktop, or mobile
The browser at claude.ai is the simplest way in. You get the full chat experience, projects, artifacts, connectors, and web search. If you're just trying Claude out, you can start here.
The desktop app (Mac and Windows) is what I’d recommend. It includes everything the browser has, plus features you can't get there: Cowork, Claude Code, and desktop extensions. It’s also more stable, no tab refreshing that kills your conversation mid-flow. Download it here.
The mobile app (iOS and Android) is for quick questions, voice input, and catching up on conversations. Not where you’d do heavy work, but useful on the go.
My recommendation: install the desktop app. It’s free, and it’s where the most powerful features live.
Once you’ve got Claude installed, you’ll need to pick a plan.
Which Claude plan to choose: free, pro, or max
Claude has three main plans for individual users.
Free gives you access to Sonnet and Haiku (not Opus 4.6, the most powerful model), plus web search, projects, artifacts, memory, connectors, file creation. You get limited messages per day and you can’t use Claude Code or Cowork. It's a good option for getting started and exploring what Claude can do.
Pro ($20/month) is the sweet spot for most people. You get access to all models including Opus 4.6, roughly 5x the usage of the free tier, Claude Code, Cowork, unlimited projects, Research mode, memory across conversations, Claude in Excel, and Claude in PowerPoint. If you're using Claude for work, this is where you want to be.
Max ($100-$200/month) is for heavy users. Choose between 5x and 20x more usage than Pro. You get higher output limits, early access to new features, and priority access during peak times. This is the plan for people who live inside Claude all day.
My recommendation: start with Pro. The free plan is fine for a quick look, but Pro is where Claude becomes useful for your work.
Step 2: Bring your context to Claude from ChatGPT (or any other AI)
Now that you have Claude installed and your plan set up, the next thing to do is bring over the context from whatever AI you’ve been using. If you haven’t been using another AI, skip to Step 3. But if you have, you don’t have to start from scratch.
I built a tool for exactly this. Here’s how it works in three steps:
Export your ChatGPT data from the Privacy Portal
Generate your complete profile using the AI Migrator. It processes your entire conversation history and extracts everything about you: your preferences, your work context, your writing style, the projects you’ve been working on
Upload your profile to Claude. Open a new conversation, paste your profile, and ask Claude to store it in its memory (make sure you’ve toggled on memory in your settings first, which we’ll cover in Step 3)
That’s it. Claude now knows you the way ChatGPT used to know you. You don’t start from zero.
Read my step-by-step guide on how to do all three.
Step 3: Set up your Claude account
You’re in, you have your plan, and if you’re switching, your context is imported. Now let’s go through your settings.
There are two reasons to go through your settings: to make sure everything is toggled on so you're not missing features, and to personalize how Claude works for you. People don’t usually open them, but there's a lot of useful stuff in here.
Set up your Claude profile
These preferences apply to every conversation, so it’s good to have them set up early.
Go to Settings → General → Profile. Here you’ll set your name, your work function (from the dropdown), and your personal preferences for responses.
If you used the AI Migrator, your preferences are already extracted and you can drop them right in here.
If not, write them yourself. Think about:
How you like to approach your work and what methods you rely on
The terms, tools, or concepts that come up often in your field
The kinds of tasks you’ll typically ask Claude to help with
Whether you prefer short, direct answers or longer, detailed ones
How you want Claude to behave: should it ask questions when something is unclear? Push back on your ideas? Think things through before responding?
Claude capabilities and settings to turn on
This is the section that unlocks most of what Claude can do. Go to Settings → Capabilities.
Turn all of these ON: Search and reference chats (so you can ask Claude "find our chat about redesigning my website"), Generate memory from chat history (so Claude remembers you and your preferences over time), Artifacts (documents, code, and tools Claude creates in a side panel), AI-powered artifacts (artifacts that use AI themselves), Inline visualizations (charts and diagrams right inside the chat), Cloud code execution and file creation (Word docs, spreadsheets, presentations, PDFs).
We’ll cover what each of these does throughout the guide. For now, just make sure they’re all toggled on.
A few that need a note:
Tool access → select “Load tools when needed”. It keeps your conversations lighter.
Allow network egress → ON, but monitor if you’re working with sensitive data.
Domain allowlist → “All domains” works fine for most people. If you work with sensitive company data, you can restrict it, but Claude might fail at some tasks if it can’t access what it needs.
How to check your Claude usage and limits
I get asked about this a lot. You hit your limit, Claude stops responding, and you don't know why or when it resets.
Go to Settings → Usage. You'll see two things: your current session limit (which resets every few hours) and your weekly limit. Both tell you how much you've used and when they reset.
You can also check before and after a task to see how much it consumed, which is especially useful if you're on a Pro plan and using Claude extensively. It gives you a feel for what's eating your quota so you can pace yourself.
Your Claude is ready. Now let’s talk about what you can do with it.
I organized the rest of the guide into three levels that take you from making the most of every conversation to having Claude work for you on its own.
Each one unlocks a new way of using it, and by the end, you’ll wonder how you got anything done without it.
Level 1: Use Claude in conversations
This level is about getting comfortable with Claude: how you talk to it, what you can ask for, and what it can do in a single conversation.
By the end of this, you'll know your way around.
Claude models: Opus, Sonnet, and Haiku, and when to use each
Claude has three models, and you switch between them using the model picker in the chat. Knowing when to use each model saves you both time and quota.
Sonnet is the default. It’s fast, smart, and handles the vast majority of tasks well: writing, analysis, coding, research, summarizing. Think of it as the workhorse.
Opus is the most powerful model. Deeper reasoning, better at complex multi-step tasks. Slower and uses more quota, so save it for when Sonnet isn’t enough.
Haiku is the lightweight option. Very fast, good for simple tasks and quick questions.
Both Opus and Sonnet also have extended thinking, a toggle in the model picker that lets Claude think through a problem before responding. I personally always keep it on.
The chat: how you interact with Claude
Most of these features are accessible from the + button in the chat input bar. Here’s what each one does.
Dictation: talk to Claude instead of typing
The microphone button on the far right lets you talk to Claude instead of typing. Great for brainstorming, brain dumps, or when you just don’t feel like typing. It transcribes in real time and you can edit before sending.
Upload files and pull from your tools
You can upload almost anything into a conversation: PDFs, images, spreadsheets, documents, code. Claude reads them and works with them.
If you’ve connected tools like Google Drive or GitHub, you can pull files directly from there without downloading and re-uploading. Same goes for Gmail and Calendar.
Web search
Claude can search the internet for current information. It does this automatically when you ask about recent events or data, but you can also toggle it on or off from the + menu.
Research mode
If you've used deep research in ChatGPT or Gemini, this is Claude's version. It searches multiple sources, cross-references what it finds, and produces a detailed report with citations. It runs in the background and can take 5 to 45 minutes, so you can keep working while it does its thing.
Writing styles
Styles let you tell Claude how to write for a specific type of work. You create them in Settings → Styles and then pick one from the + menu before or during a conversation.
Think of them as presets: one for formal client emails, one for casual social posts, one for technical documentation.
A note: if you try uploading long documents for it to create a style from, it fails. Building a custom writing skill (Level 2) gives better results. You can still use styles for everyday stuff and smaller pieces of writing.
Connectors in the chat
Claude can connect to tools you already use, like Google Drive, Gmail, Calendar, Notion, Slack, Canva, and more. (We’ll cover how to set them up in Level 3.)
Once they’re connected, you can turn them on and off per conversation from the + menu, so Claude only accesses the tools you need for that specific task.
What Claude can produce
Everything above is about how you guide the conversation: what you give Claude, how you search, how you set the tone.
But Claude doesn't just reply with text. Here's what else it can create.
The inline visualizer: when Claude explains things visually
This just launched this week. Claude can now build interactive visuals right inside the conversation: charts, diagrams, flowcharts, timelines.
Sometimes it does this on its own when a visual would explain something better than text, but you can also ask for it directly: "draw this as a diagram", "chart this data", "explain this to me visually".
The visuals are temporary and part of the conversation flow. If you want to keep one, you can save it as an image or convert it into an artifact.
Artifacts: when Claude builds things for you
Artifacts appear in a side panel next to your conversation. They’re for anything Claude creates that’s more substantial than a chat response: documents, code, interactive apps, websites, dashboards, visualizations.
When Claude creates an artifact, you can copy it, download it, share it with a link, iterate on it, or fork it into variations. Some artifacts are interactive. With “AI-powered artifacts” turned on, Claude can even build apps that use AI themselves.
File creation
Claude can create actual files you can download: Word documents, spreadsheets, presentations, PDFs, and data reports. Polished, formatted, ready to send to clients or share with your team.
Level 2: Make Claude personal to you and your work
Level 1 was about individual conversations. This level is about making Claude personal.
You’ll set up projects so Claude always has the context for your work, teach it how you do things with skills, connect the tools you already use, and bring Claude into the apps you work in every day, whether that’s Excel, PowerPoint, or your browser with Claude in Chrome.
By the end of this, Claude stops feeling like a tool and starts feeling like a collaborator who knows how you work.
If you're not sure whether something should be a project, a skill, or a Cowork thing, I wrote a decision framework that helps you pick the right one.
Projects: your persistent workspace
If you've used projects in ChatGPT, this works the same way. You set your context once, and every conversation inside the project starts with it.
Claude also builds memory across all conversations in the same project, so the more you work inside one, the smarter it gets about that work.
How to create a project in Claude
Click “Projects” in the sidebar
Create a new project
Give it a name (something clear: “Newsletter”, “Client Work”, “Product Development”, “Book writing”)
Add custom instructions: tell Claude what this project is about, how to approach it, and what context it needs
Add knowledge files: upload documents, style guides, reference materials, any context that helps Claude produce better work on that project
What to put in your Project’s custom instructions
This is the key part. Your custom instructions are the context Claude reads before every conversation in this project.
Use this prompt in a new Claude chat to guide you through creating them:
You are a Claude Project setup assistant. Your job is to interview me, then generate custom instructions for a new Claude Project.
RULES:
- Ask ONE question at a time. Wait for my answer before moving on.
- Do NOT generate instructions until the interview is complete.
- If my answers are vague, push back and ask me to be more specific.
- Use plain language, no jargon.
INTERVIEW FLOW:
1. What specific task or workflow is this project for?
2. Who is the audience for the output?
3. What role should Claude play? (e.g., ghostwriter, editor, strategist)
4. What tone and format preferences matter?
5. What should Claude ALWAYS do in this project?
6. What should Claude NEVER do?
7. How should Claude approach work? (ask questions first? draft options? go straight to output?)
8. Do you have examples of good output you can describe?
Skip any questions that have already been answered. When done, generate instructions using this structure:
**Role** — What Claude does in this project (one sentence)
**Context** — Who the audience is and what matters to them
**Rules** — Always/never rules (max 8-10)
**Process** — How Claude should approach tasks step by step
**Output format** — Formatting preferences
End by suggesting what documents to upload to the project knowledge base.
By the time you’re done, you’ll have custom instructions ready to paste into your project.
If you used the AI Migrator, you can go one step further: drop the project context from your generated profile into this same conversation so Claude uses it as a starting point alongside the interview. The result will be custom instructions that already know what you’re working on.
One more thing: if you start a conversation with Claude outside a project and realize it belongs in one, you can add it anytime. Either from the + button inside the conversation, or from your chat history by clicking the three dots on any chat and selecting "Add to project".
Skills: teach Claude how you work
Skills are pre-built instruction sets that give Claude specialized expertise for specific tasks.
Once you set them up, they work everywhere in the Claude desktop app: in the chat, in projects, in Cowork, in Claude Code.
What skills look like
A skill takes something you do repeatedly and turns it into instructions Claude follows every time. Some examples:
Writing voice: analyze your best writing, define your style, and Claude matches it across articles, posts, and emails (check my writing skills here and here)
Meeting action items: feed in your format, and any transcript gets turned into action items the way you’d write them
Presentations: your preferred structure, slide flow, tone, and design, applied every time you build a deck (see my Claude Brand Skill for creating on-brand docs)
Sales offers: your framework, pricing logic, structure, so proposals sound like you
How to get your own Claude skills: download or create your own
Downloading — there are community-built skills you can upload. This works well for generally applicable skills where customization doesn’t matter much.
Creating your own — this is where the real value is. My usual approach is to feed Claude examples of the output I want: past work, documents I’ve created, formats I use. Let it analyze how I do things, then turn those patterns into a skill. From there, I tweak and update it until I’m happy with the results. See how I built my own writing skills step by step.
Keep in mind there’s a 50-skill limit in the desktop app. If you fill it with other people’s skills, you won’t have room for your own once you start creating them.
Connectors: give Claude access to your tools
Connectors let Claude access the tools you already use: Google Drive, Gmail, Calendar, Slack, Notion, Canva, Gamma, and more.
Instead of jumping between apps to grab what you need, copy-pasting stuff from one place to another, connectors let Claude go straight into your tools: extracting information, writing files, making changes for you.
Go to Customize → Connectors, find the tools you already use, and connect them. Each one takes about 30 seconds: you authorize it by logging into that platform, and it’s ready.
The AI Blew My Mind MCP: prompts and image generation inside Claude
Speaking of extending what Claude can do, I built my own MCP that works right inside Claude. It gives you access to my curated prompt library, the ability to save your own prompts so you can access them easily across conversations, plus image generation powered by Google Gemini, and .
You can search prompts, fill them in through a guided interview, and generate images without ever leaving the conversation. Free subscribers get access to the MCP with a selection of prompts. Premium subscribers unlock everything, including the prompts from my deep dives and ones I don’t publish anywhere else. Here’s how to set it up.
Claude in Excel and PowerPoint: where you already work
Claude also works directly inside Excel and PowerPoint. Both are available on all paid plans.
In Excel, it reads your workbooks, explains calculations, suggests formulas, creates charts, and cleans messy data.
In PowerPoint, it reads your slide layouts, fonts, and templates, then creates or edits presentations that stay on brand.
You don't switch to Claude to do this work. Claude comes to you, right inside the app.
Claude in Chrome: AI inside your browser
Claude in Chrome is a browser extension that lets Claude see and interact with the page you’re looking at. It can read content, extract information, fill forms, and take actions on websites.
Cowork also uses it when a task involves browsing, so if you ask Cowork to research something or interact with a web app, Claude in Chrome is what makes that possible.
Level 3: Let Claude work for you
Everything up to here has been about working WITH Claude. This level is about Claude working independently, on your behalf.
Claude Cowork: from chatbot to co-worker
Cowork is Claude’s autonomous mode, and the best way for non-technical people to experience what agentic AI can do for their work.
How it works
You access it by switching from the Chat tab to the Cowork tab at the top. Give it a task and be clear about what you want, like you would when delegating to a person. Then come back when it's done.
In the meantime, Cowork:
Breaks the work into steps and tracks its own progress
Reads and creates files on your computer
Automatically pulls in your skills and connectors
Handles multiple things at once
Here's an example: I asked Cowork to create a promotional content calendar by writing 3 Substack Notes for each of my 20 most recent articles. It read my voice and substack-notes skills, analyzed all 20 articles, and created an Excel file with 60 notes in my writing style, all on its own.
Read my full deep dive on Cowork, where I covered everything about it and shared 77 use cases.
And once you're in deep, give it a permanent home with a Cowork operating system. That’s how you take it to the next level.
Plugins in Cowork
Plugins extend what Cowork can do. They bundle skills, connectors, and commands into specialized packages for specific roles and workflows.
Think of them as turning Cowork into a specialist for marketing, sales, finance, or whatever your work requires.
To get started, go to Customize → Plugins, click the + button, and browse the plugins created by Anthropic and install the ones that are relevant to your work. You can also create your own with Claude or upload one someone else built.
Scheduled tasks in Cowork
Claude can run tasks on a schedule, automatically. Think of it like a recurring calendar event, but instead of reminding you to do something, Claude does it for you.
To set one up, type /schedule in any Cowork chat and describe what you want repeated. You set the recurrence and Claude runs it on its own from that point on. You can also go to the Scheduled section in Cowork to create new tasks or edit existing ones.
And once you have scheduled tasks running, you can connect them so they work as a chain. I wrote about how to stitch your workflows together in Cowork so each task’s output feeds into the next and runs without you in the middle.
Projects in Cowork: organize your tasks into workspaces
Cowork now has its own projects, separate from the ones in Claude’s chat. Each project gets its own files, instructions, memory, and tasks, so you can keep different areas of your work organized without everything bleeding into one conversation.
You can create a project from scratch, import one from an existing Claude chat project, or point it to a folder already on your computer. It’s like having a dedicated desk for each area of your work.
If you want help setting up your Cowork project properly, the Cowork Project Setup Amplifier in the AI Blew My Mind MCP walks you through writing your project instructions, designing your folder architecture, and creating scheduled task prompts that work when nobody's watching.
Live Artifacts in Cowork
Live Artifacts are persistent, interactive dashboards that Claude builds for you inside Cowork. Unlike regular artifacts, they save to their own tab, refresh with current data from your connectors every time you open them, and keep version history. Think a morning brief pulling from Slack and your calendar, or a competitive tracker that updates itself. Learn more.
Claude Code: from user to builder
Don’t let the name fool you. Claude Code is not just for developers (anymore).
Yes, it can build things: websites, tools, apps, products. I used it to go from a prototype to a fully live product with payments, email capture, SEO, and analytics.
But it’s also becoming the tool I turn to for work that has nothing to do with coding. Analyzing large datasets. Automating repetitive processes. Processing hundreds of files at once.
I've even built an AI agent in Claude Code that runs my whole content distribution on autopilot, in the cloud, while my laptop is closed.
It’s vibe working: using a coding tool for non-coding work, because it gets things done faster and more thoroughly than doing it manually.
You access it from the Code tab in the desktop app or from the terminal. You describe what you want in plain language and Claude figures out how to make it happen.
Computer use + Dispatch: Claude works on your computer while you’re away
Until now, Cowork and Claude Code still needed you at your desk. Computer use changes that. Claude can now open your apps, click buttons, navigate your screen, and do work on your desktop the way you would.
You can work side by side, tackling different things at the same time. Or you can leave your laptop, text Claude a task from your phone with Dispatch, and come back to a finished result.
Think: content research while you’re at the gym, sponsor outreach while you’re making lunch, a subscriber analytics dashboard that refreshes itself every Monday.
You can use it in three ways:
The desktop app (Cowork + Claude Code tabs): You give Claude a task and let it work on your screen. You can watch, go do something else, or work side by side on different things.
Dispatch (from your phone): You pair your phone with your desktop through the Claude app. Then you text Claude tasks from anywhere. It works on your computer while you’re at a coffee shop, on the train, wherever. Your computer just needs to stay on.
Claude Code CLI (terminal): You type plain English into the terminal and Claude builds it. Now it can also step out of the terminal and onto your screen to test what it built, click through it, and verify it works.
I tested computer use on 6 real workflows. Here’s my deep dive with how to set everything up and the workflows with prompts, video demos, and honest verdicts.
Claude Managed Agents: build AI agents into your products
Everything above runs on your computer, for you. Managed Agents is different. It’s infrastructure for building AI agents into products that other people use.
You define the agent (system prompt, model, tools), Anthropic runs it on their servers, and each user gets their own isolated sandbox. If you have a SaaS tool, an internal platform, or anything where users interact with something you’ve built, Managed Agents lets you add AI capabilities without building server infrastructure from scratch.
Your turn
You don't need to do everything today. Pick one thing from this guide you haven't set up yet and start there.
If you’re brand new to Claude: work through the Getting Started steps. Get everything configured properly. It takes 15 minutes and changes everything about how Claude works for you.
If you’ve been using Claude but it still feels generic: Level 2 is your unlock. Set up a project, create a skill, and watch the difference.
If you’re already comfortable: explore Level 3. Cowork and Claude Code will change how you think about what AI can do for your work.
Once you see the difference, you’ll want to come back for the next one.
I hope this guide helped you discover something new that you can apply in your work today. And if it did, share it with someone who’s just getting started with Claude.
Let me know in the comments what you’re going to try first.
This post is free. If you found it useful and want access to more of what I’m building - prompts, automations, step-by-step guides - paid subscribers get all of it.
Daria, this was awesome! I've read so many guides about Claude this week and watched so many YouTube tutorials, and this is right there at the top. Thank you!
Thanks for the guide! I just commited to try Claude, and I’m pretty much amazed. In my first project using Claude I could set up a Minecraft server and run simulations. There’s a video of it here: https://substack.com/@scientistmom1/note/c-227641979