Transcription
Most people open Claude, see three different models, and get paralyzed. They don't know whether to use Haiku, Sonnet, or Opus. They miss powerful features hiding in the interface. Let me show you exactly how to use Claude like a pro.
First, let's get you set up properly. Head to claude.ai and create your account. You can start free, and honestly, the free tier is surprisingly generous. You get Claude Sonnet access, web search, memory, and even file creation. That's enough to test everything before upgrading.
The interface looks simple, but there's more here than meets the eye. At the top, you'll see the model selector. This defaults to Sonnet 4.6, which is perfect for most tasks. Don't change this yet. Below that is the extended thinking toggle. Leave this off for now. We'll cover when to use it.
On the left sidebar, you have your chat history and projects. Projects let you organize related conversations, share context between chats, and build up knowledge over time. Create your first project now. Name it something like work-related, like marketing strategy or code reviews.
Custom instructions are hidden in your profile settings. Click your avatar, go to settings, and find custom instructions. This is crucial. It tells Claude your context, so you don't repeat yourself. Mine says, "I run a YouTube channel about AI. Assume I understand technical concepts, but appreciate practical examples. Focus on real-world applications over theory."
The free tier gives you about 30 messages every 8 hours with Sonnet 4.6. That's actually quite a lot for testing. But if you're a serious about using Claude daily, the $20 pro subscription [music] is worth it. You get 5x more usage, access to Opus 4.6, and new features like Claude Cowork.
One interface tip most people miss, you can have multiple chats in tabs. Middle click on new chat to open in a new tab. This lets you run different conversations in parallel without losing context. Game-changer for comparing different approaches.
Claude has three models, and understanding when to use each one will transform your experience. Think of them as different tools for different jobs.
Haiku 4.5 is your quick assistant, lightning fast, perfect for simple tasks. Use it for reformatting text, basic summaries, quick questions. It costs the least tokens, so it's ideal for high-volume tasks. In the web interface, you manually switch to Haiku when you need speed over depth.
Sonnet 4.6 is your workhorse, and honestly, it's incredible. This model is so good that most developers now prefer it over the previous Opus. It handles 90% of tasks brilliantly. Writing, coding, analysis, research. It costs 5x less than Opus.
But let's be clear about complex coding. While Sonnet 4.6 handles most daily programming tasks brilliantly, Opus 4.6 remains the undisputed champion for gnarly debugging, system architecture, and complex refactoring. When you're stuck on a problem that other models can't crack, Opus often solves it faster and cleaner. Think of Sonnet as your daily coding partner and Opus as the senior architect you bring in for the tough calls.
Opus 4.6 is your expert consultant. When you need the absolute deepest reasoning, complex architectural decisions, multi-layered analysis, tasks that require understanding massive amounts of context, that's when you switch to Opus. It also supports agent teams, which we'll cover later.
Here's my decision framework. Start every task with Sonnet 4.6. Only switch to Opus when Sonnet's answer isn't quite deep enough, and drop down to Haiku for bulk processing or when you just need something reformatted quickly. On the pro plan, you get generous limits with Sonnet and some Opus usage. The key is not wasting Opus on simple tasks. I see people burning through their Opus limits asking it to write basic emails. Don't do that. Save Opus for the problems that actually need that level of intelligence.
Speaking of getting more value from your Claude subscription, let me show you one of the most powerful workflows I've discovered. Now, one of the most powerful use cases I found with Claude 4.6 is using it for scaled outreach. Claude can write incredible personalized emails, but here's the problem. You still need to find the right people to send those emails to. And that's where my workflow gets interesting. So, I've been using Lese AI for this. It's the first AI agent built specifically for finding people. And honestly, it's changed how I approach outreach completely. Instead of wasting hours on LinkedIn with those terrible filters, I can just describe exactly who I want to reach. Like, watch this. I'm looking for AI consultants who worked with enterprise clients and actually know what they're talking about. I'll type, "AI consultants who implemented LLMs for Fortune 500 companies and write about prompt engineering." See that? It's not just matching keywords. Lese actually understands what I'm looking for. These are real people who match all my criteria, not random consultants who happened to mention AI once in their bio. I've tested every people search tool out there, and nothing comes close. I export this list, and right here, I can write personalized emails to all these people right inside Lese AI. The whole process takes maybe 10 minutes instead of hours. If you want to try this yourself, check out the link in the description below. They've got integrations with Claude code, too. So, you can build this right into your AI stack.
Extended thinking is the feature that separates Claude from every other AI. When you toggle it on, Claude doesn't just respond. It actually thinks through the problem step by step before answering. Here's when to use it. Complex analysis where nuance matters, multi-step problems with dependencies, anytime you'd normally think, "I need to really think about this." That's when you turn on extended thinking. Watch what happens when I analyze this business strategy. With extended thinking off, Claude gives me a good answer in 5 seconds. Now, watch with it on. See that thinking section? Claude is considering different angles, questioning assumptions, exploring edge cases. The final answer is dramatically more thorough. But here's the key, don't use it for everything. Extended thinking uses more of your message limits and takes longer. For simple questions like, "Reformat this list" or "Summarize this paragraph," keep it off. Save it [music] for the problems that actually need deep reasoning.
You can expand the thinking section to see Claude's actual reasoning process. This is incredibly educational. You start to understand how to structure your own prompts better when you see how Claude approaches problems. Pro tip, if Claude's response seems surface level or misses nuances, that's your cue to regenerate with extended thinking on. I'd estimate I use it for about 20% of my queries. The ones where depth really matters.
Same prompt, extended thinking off. "What pricing model should I use for my online course? One-time payment or subscription?" Claude gives me an answer. Reasonable. Lists some pros and cons, generic advice that could apply to anyone. Now, same prompt, extended thinking on. Read that thinking section. Claude is considering my specific context, AI education, beginner audience, freelancers. It's thinking through churn risk for subscriptions. It's considering the psychology of beginner buyers. It's weighing one-time payment against recurring revenue for a solo creator. The final answer is completely different, more specific, more nuanced, actually useful for my situation. That's the difference. Extended thinking doesn't just give you more words. It [music] gives you better reasoning.
So, when exactly should you turn it on? Decisions with real consequences, pricing strategy, business model questions, choosing between technical approaches, anything where getting it wrong costs you time or money, analysis that requires holding multiple variables at once. "Analyze these five marketing channels and tell me which two to prioritize given my budget and audience." Multivariable problems are exactly where extended thinking shines. When you need speed, leave it off. Quick back-and-forth brainstorming sessions, any situation where momentum matters more than depth. Extended thinking is like using a calculator to count to 10 if the task doesn't need it. Wasteful. Extended thinking is the most underused feature for people who do knowledge work. Consultants, strategists, founders, educators, anyone whose job is to think through complex problems. If that's you, this toggle should be your best friend.
Memory is the feature that transforms Claude from a tool into true assistant. And the best part, it's available even on the free tier. Go to settings, find capabilities, and turn on generate memory from chat history. Now, Claude remembers important details across all your conversations, not just within a project, across everything. Here's what makes this powerful. Tell Claude your preferences once, and it remembers forever. I told Claude I prefer concise bullet points for summaries, that I work in AI education, and that I'm familiar with technical concepts. Now, every response is tailored to that context, but memory goes deeper than preferences. Claude remembers your projects, your goals, the problems you're working on. When I come back a week later and say, "Update the marketing strategy we discussed," Claude knows exactly what I'm talking about.
Here's my memory workflow. Start each major project by explicitly telling Claude the context. "I'm launching a new course about prompt engineering for beginners. The target audience is freelancers who want to add AI services." Claude remembers this and factors it into every related conversation. The compound effect is incredible. After a month of use, Claude knows your work style, your projects, your preferences. It's like having an assistant who's been working with you for years. Watch what I'm doing right now. I'm in a fresh chat, no context, no setup, one message. Remember these things about me, I run an AI education business targeting complete beginners. My students are non-technical, no jargon ever. I publish on YouTube and sell online courses. My tone is direct and practical, not academic. I'm closing this chat entirely, opening a brand new conversation. Starting completely fresh. Write me an intro for a YouTube lesson explaining what an AI model actually is. Read that response. Claude didn't write for a developer, it didn't use technical language, it wrote for a beginner in a direct tone structured for YouTube. I said nothing about any of that in this conversation. Memory carried it over automatically.
Here's the workflow I recommend for anyone starting out. In your first week, have three intentional conversations. One where you explain your work context, one where you explain your communication preferences, and one where you explain your current projects and goals. After those three conversations, Claude has a solid foundation. Everything after that builds on top of it. When I open Claude in the morning, it's not a blank slate, it's a continuation. That's a fundamentally different experience and it's available on the free tier. Turn it on today.
This might be the most underrated upgrade in Claude 4.6. Both Opus and Sonnet now support 1 million token context window. Let me put that in perspective. That's roughly 750,000 words. You can upload an entire book series, a year of emails, or a complete code base. But here's what's revolutionary. Claude maintains 76% accuracy finding specific information even at the far edges of that million token context. The previous models, they'd hit 18% accuracy, basically forgetting everything in the middle. Watch this real example. I'm uploading 10 quarterly reports, each one 80 pages. In the old world, I'd have to split these up, analyze separately, lose connections between them. Now, compare revenue trends across all reports and identify seasonal patterns. Claude reads all 800 pages, maintains perfect context, and finds patterns I missed. For developers, this is game-changing. Upload your entire code base, every file, every function. Find all instances where user authentication might fail. Claude doesn't just grep for keywords, it understands the code flow across thousands of files. The best part, context compaction works behind the scenes. As you approach token limits in long conversations, Claude intelligently summarizes older parts while maintaining the thread. Your chat can run for hours without degrading. Pro workflow, create a project, upload all your reference materials at once, contracts, documentation, historical data, whatever defines your work. Now, every conversation in that project has perfect context. No more as I mentioned before, Claude remembers everything.
Claude 4.6 is incredible for text analysis and coding. There's one massive gap, it can't generate images or videos. Zero visual creation capabilities. This is actually why I use both Claude and AI Master daily. They are perfect complements. Claude is my thinking partner when I need deep analysis, complex writing, or code architecture. Nothing beats it. But when I need to create visual content, that's where AI Master comes in. Watch my actual workflow. I use Claude to write a detailed image prompt. It's brilliant at understanding visual concepts and crafting descriptions. Then, I take that prompt to AI Master Studio where I have access to all the latest image models, Nano Banana 2, Pro, even GPT image. Same thing with video. Claude writes the script, AI Master generates the visuals with VEO, Kling, etc. And you learn prompt engineering that works across both text [music] and image models. This combination approach is what separates professionals from hobbyists. Use Claude for what it's best at, thinking and writing, and try AI Master for what Claude can't do, structured learning and visual creation. Link in the description below, 30% off annual right now.
Claude's web search isn't just Google built in, it's an intelligent research assistant. And unlike ChatGPT, Claude shows you exactly which sources it's using and lets you verify everything. For basic web search, just ask normally. What are the latest updates to Google's algorithm? Claude searches, synthesizes multiple sources, and gives you current information with citations. Click any citation to see the exact source. But here's where it gets powerful, deep research mode. This is pro only and it's worth the upgrade alone. Instead of one quick search, Claude runs extended research sessions diving deep into topics and building comprehensive reports. Watch this. [music] Do deep research on how small businesses are using AI for customer service in 2026. Claude doesn't just search once, it runs multiple queries, follows interesting [music] threads, synthesizes dozens of sources. The result is a complete research report that would take hours to compile manually. The web search is smart about filtering, too. It knows to prioritize recent information events, but looks for authoritative sources for established topics. When researching tools, it finds actual user reviews, not just marketing pages. Pro tip, for deep research, be specific about what you want. Don't just say research AI trends, say research how retail businesses under 50 employees are using AI for inventory management, focusing on cost savings and implementation challenges.
Projects are Claude's secret weapon for serious work. Think of them as persistent workspaces where context builds up over time. Create a project for each major area of work. I have projects for YouTube scripts, course development, business strategy, and code reviews. [music] Each project maintains its own context, knowledge, and conversation history. Here's the power move, project [music] knowledge. Upload key documents to project and Claude references them in every conversation. I uploaded my brand guidelines to my YouTube scripts project. script follows my style guide automatically. You can share projects with team members on team plans, everyone sees the same context, the same conversations, the same knowledge base. It's like having a shared AI assistant that knows your entire project history. The project sidebar shows all related chats. Click any to resume exactly where you left off. No more searching through chat history trying to find that conversation about the pricing strategy. Within projects, use naming conventions for chats. I prefix with dates and topics, 2026 0330 Q2 marketing plan or 2026 0330 bug fix for login flow. Makes finding things [music] instant. Pro workflow, start each project with a project brief chat where you lay out all the context, goals, constraints. Refer back to this chat when starting new work. Keeps everything aligned. Watch the difference, same prompt two places. Regular chat, no project, write a hook for a YouTube video about AI memory features. Fine, generic YouTube energy, could be anyone's channel. Now, inside my YouTube channel project, same prompt, write a hook for a YouTube video about AI memory features. Completely different. The tone matches my style, the complexity matches my audience level, the structure fits my format. Claude pulled all of that from the style guide I uploaded. I said nothing about any of it in this conversation.
Three things [music] I keep in every project's knowledge base. First, one-page brief, what this project is, who it's for, what success looks like. Second, examples of good work, three or four pieces I'm proud of. Claude uses these to calibrate tone and quality. Third, constraints and rules. For my YouTube project, always write for beginners, never use jargon without explaining it, always end with a clear call to action. Three documents, maybe an hour of work to create them. Every conversation in that project dramatically better, forever.
Artifacts completely change what's possible with Claude. Instead of just chatting, you can create actual downloadable files directly in the interface. Now, when you ask Claude to create substantial content, it appears in a side panel as an actual artifact you can work with. Watch this. Create a complete sales page for my prompt engineering course. Claude doesn't just write text in the chat, it creates a formatted document in the artifacts panel. I can edit it directly, ask for revisions, and download it as a proper file. But it's not just documents. Claude creates Excel spreadsheets with formulas, PowerPoint presentations with proper slides, even interactive HTML pages. Build me a customer survey form with conditional logic. Boom, we're going HTML form ready to deploy. For developers, artifacts handles code brilliantly, complete files with syntax highlighting, proper structure, even basic testing. You can copy directly or download as proper code files. The revision workflow is smooth. Click any part of the artifact, tell Claude what to change, and it updates just that section. No need to regenerate everything. This iterative process is how you build complex documents efficiently. My favorite artifact workflow, start with an outline, expand sections one by one, then polish the final result. Way faster than traditional writing, but you maintain complete control. Watch this prompt. Create a weekly content calendar for an AI education YouTube channel. Build it as a spreadsheet with five columns: day, topic, hook idea, call to action, and estimated length. That panel on the right, that's not text in a chat. That's a structured artifact. Rows, columns, actual content in each cell, ready to use. I didn't get a wall of text I have to manually format. I got a file. Click into any cell. Tell Claude, "Change Thursday's topic to something about AI image generation." And watch it update just that cell. Not the whole thing, just that one piece. You're not regenerating everything every time you want a small change. You're sculpting. When you're happy with it, hit download. A real Excel file. Opens in Excel, Google Sheets, wherever you work. Not a screenshot, an actual working spreadsheet.
Three more artifact types I use constantly: HTML pages. Turn this content calendar into a simple web page I can share with my team. Same data, now it's shareable with anyone, no Claude account needed. Documents. Write a complete onboarding guide for new students in my AI course. Sections, headers, table of contents, structured, clean, ready to download. No formatting work on my end. Code. Build me a Python script that takes a CSV of YouTube video ideas and outputs them sorted by estimated search volume. Complete file, proper structure, download directly, no copy-paste from a chat window.
The mindset shift. Stop thinking of Claude as a place where you get text. Start thinking of it as a place where you build things. Every time you catch yourself copying text out of Claude to paste somewhere else, ask, "Could this be an artifact instead?" Usually, yes, and the artifact version saves you 20 minutes of formatting work.
All right, we've covered the essentials, but honestly, that's the foundation. I've put the complete advanced section inside AI Master as a hands-on masterclass. In part two, on the platform, you learn how to build skills that automate your repetitive tasks, setting up Claude co-work for read file management, using computer use safely without breaking anything, common mistakes that waste your Claude limits. So, head to aimaster.me to continue the masterclass and start practicing in the studio. Link below, and see you inside part two.