How to Use Claude Like a Pro in 2026

The most surprising advice in the tutorial is negative: stop using the strongest settings by default. Extended thinking burns message limits on simple tasks, Opus is wasted on emails, and long prompts often cost more than tight ones. Using Claude like a pro mostly means matching the tool to the task, and this guide shows exactly where each setting belongs.

How to use Claude: the model you pick matters more than the prompt

How to use Claude effectively starts with model choice: Claude is Anthropic AI assistant, and it offers three model tiers with different speed, cost, and capability trade-offs. The tutorial's rule is simple: start every task with Sonnet, drop to Haiku for bulk or trivial work, and save Opus for problems Sonnet cannot crack. You can confirm current model names and capabilities in Anthropic model overview.

The video describes the lineup this way, based on the author's first-hand experience (evidence class D):

Setup: account, custom instructions, and memory

Setup is the step most beginners skip, and it is where the tutorial says the biggest gains live. Create an account at Claude, then open the profile settings and write custom instructions that describe your role, audience, and preferred output style. The author's example: 'I run a YouTube channel about AI. Assume I understand technical concepts but appreciate practical examples.'

Memory is the second unlock. In settings, under capabilities, enable generating memory from chat history. The author recommends three intentional first-week conversations: one explaining your work context, one your communication preferences, and one your current projects. After that, Claude carries that context into new chats. The video demonstrates this by stating context in one chat, closing it, and getting a beginner-toned YouTube intro from a fresh chat with no repeated instructions.

The author also recommends opening multiple chats in browser tabs (middle-click on 'new chat') to run parallel conversations without losing context. None of these features require a paid plan according to the tutorial, though paid plans raise usage limits, which Anthropic documents on its pricing page.

Extended thinking: when to turn it on and when it wastes tokens

Extended thinking makes Claude reason step by step before answering, and the toggle sits below the model selector. The video's guidance is to use it for roughly 20% of queries: decisions with real consequences, multi-variable analysis, and any problem where you would normally think 'I need to really think about this.' Keep it off for reformatting, quick summaries, and fast back-and-forth.

The author shows a pricing-strategy question answered twice. Without extended thinking, the answer is a generic pros-and-cons list. With it on, the visible reasoning weighs churn risk, beginner-buyer psychology, and one-time versus recurring revenue for a solo creator. Note that this comparison is the author's own demonstration, not a benchmark; Anthropic does document extended thinking as a feature for complex reasoning in its extended thinking guide.

A practical cue from the video: if a response feels surface-level, regenerate with extended thinking on rather than rephrasing the prompt five times.

Projects and Artifacts: building context and files, not chat text

Projects turn Claude into a persistent workspace. Create one project per major area of work, upload key documents to project knowledge, and every conversation inside it references those files automatically. The author keeps a one-page brief, three or four examples of good work, and a constraints list in every project, and says an hour of setup improves every future conversation in that project.

### What goes into a project knowledge base

Three documents do most of the work, according to the tutorial:

• A one-page brief: what the project is, who it serves, what success looks like.

• Examples of your best work, so Claude calibrates tone and quality to real samples.

• Explicit rules, such as 'always write for beginners' or 'never use jargon without explaining it.'

Artifacts are the output side of the same idea. Instead of chat text, Claude renders spreadsheets with formulas, documents, HTML pages, and code in a side panel you can edit piece by piece and download as a real file. The author's mindset shift: every time you copy text out of the chat to format it elsewhere, ask whether it should have been an artifact. Both features are documented in Claude help center.

Claude Code for non-developers: four projects and five mistakes

Claude Code, Anthropic agentic coding tool that runs in the terminal, is the tutorial's answer for people who need software but not a software engineering career. The setup takes about five minutes: install Node.js, install Claude Code, and open Cursor, an AI-first code editor built on VS Code, so you can watch every file Claude touches. The video then builds four projects with the same pattern:

### The opus-to-sonnet handoff pattern

1. Plan with Opus: type a command to switch models and ask for a full plan before any code is written.

2. Build with Sonnet: switch models again and say 'follow the plan exactly and build it.'

3. Fix by screenshot: paste an image of anything that looks wrong and ask Claude to fix it; the tutorial shows one screenshot of a button producing a matching style update across all buttons.

The four example projects are a personal landing page with Next.js, a Chrome extension that summarizes highlighted text, a Python folder-watcher that generates social captions via a vision API, and a single-file habit tracker that stores data locally in the browser. The author's caution is worth preserving: this workflow does not replace engineers for complex fintech backends or security-critical platforms.

The five beginner mistakes the video names: skipping the Claude.md memory file, writing vague prompts like 'make it look better,' skipping plan mode on tasks touching more than a few files, never running /compact until the context fills, and retrying the same broken fix three times in one session instead of starting fresh.

Prompting techniques Anthropic actually recommends

The strongest part of the tutorial is that its core techniques trace back to Anthropic prompt engineering documentation rather than folklore. Treat Claude the docs put it, like a new employee without context: the more precise the brief, the better the result. The techniques that carry the most weight:

### Five techniques with the best evidence

• Four-block structure. Separate instructions, context, task, and output format. The video's example: a fitness-app welcome email specified with role, audience constraints, timing, and an exact format, which produced a specific hook instead of generic onboarding filler.

• XML tags. Anthropic uses XML in its own internal prompts; wrapping context, examples, and rules in tags removes the model's need to guess what is what. The author reports (class D, self-measured across 40 prompts) better formatting consistency with tags than without.

• One example beats adjectives. 'Friendly but professional' averages into blandness; a single paragraph of your real writing locks voice, rhythm, and length.

• Prefill. Write the first few words of the response yourself so Claude continues in your format instead of opening with 'I'd be happy to help.'

• Prompt chaining and self-critique. Split multi-step work into sequential prompts, then ask Claude to critique its own draft and rewrite. Anthropic console even ships a prompt improver built on this draft-critique-revise loop.

A note on precision: several statistics in this section of the video ('3x better first drafts', '94% vs 61%') are the presenter's own informal tests on small samples, not published Anthropic research. The techniques are documented; the specific multipliers are not.

Tutorials in this genre come from all over, including channels like Dev Doido do canal do youtube and the AI Master channel behind this one, and the same rule applies to all of them: test the technique on your own prompts before trusting the numbers attached to it. For more prompt-writing breakdowns, crazystack.com.br collects similar guides.

Plans, pricing, and what this video gets wrong or cannot verify

Pricing per the video: a free tier with limited Sonnet messages, Pro at $20/month with higher limits and Claude Code access, and Max at $100/month for heavy users. Match this against Anthropic current pricing page before you subscribe, because plans and limits change. The video's cost advice for Claude Code is sound regardless: start on Pro, and only move to pay-per-token API billing once you know your actual usage.

Two claims deserve caution. First, the video cites a 76% retrieval accuracy figure at 1 million tokens of context; that number is presented by the presenter without a linkable source, so treat it as unverified. Second, the tutorial promotes third-party and unnamed tools, including an open-source rendering tool and a paid education platform with bundled model access; 'fully open source' claims and bundled discounts are the creator's commercial pitch, not independent verification. Anthropic itself has published guidance on long-context reliability noting that retrieval quality degrades with context length in model- and task-specific ways, so treat any single accuracy number skeptically.

FAQ

• Which Claude Code should beginners use? Start with Sonnet as the default for nearly everything. Use Haiku for reformatting and bulk work, and switch to Opus only when Sonnet's answer lacks the depth a hard problem needs, since Opus consumes limits faster.

• Is Claude free tier enough to learn on? The tutorial says yes: the free tier includes access to Sonnet, web search, memory, and file creation, with limited messages per rolling window. Upgrade to Pro at $20/month only once daily usage hits the ceiling.

• When should I turn on extended thinking? Turn it on for decisions with real consequences, multi-variable analysis, and problems where depth matters more than speed. Leave it off for quick reformats and brainstorming, because it uses more of your message limits.

• Do I need to code to use Claude Code? No, according to the tutorial: you describe what you want in plain English and Claude writes the files. You should still review what it produces, and the video itself says complex, security-critical systems still need real engineers.

• What is the single highest-impact prompting change? Replace one-line search-style queries with a structured brief: role, context, task, and output format, separated clearly (XML tags help). This is the technique Anthropic own documentation emphasizes most.

Turn your own walkthroughs into written guides

This article exists because a 105-minute video walkthrough contained a complete, teachable system: model choice, setup, prompting structure, and a build pattern anyone can follow. If you have recorded similar knowledge, a tutorial, an interview, a process explanation, that same material can work as a searchable, citable article instead of sitting inside a video timeline.

Skala blog handles that conversion: paste a YouTube URL, get a transcription, and generate a structured draft article you can edit and publish.

Source video