Fifteen Claude Tips, Beginner to Expert

Most people are using Claude at a fraction of what it does. Not because they are prompting badly. Because the things that matter most are configuration decisions, and nobody walks you through configuration.

There are no clever prompts in this guide. Every item below is a setting, a surface or a habit, and each one is either something Anthropic documents or something clearly labelled as a judgement call. Where the video says something the documentation does not back, this guide says so in plain words rather than repeating it.

Six of the fifteen are in the video. The other nine are here.

These five are about not making Claude's job harder than it needs to be.

The current line-up is Sonnet 5, Haiku 4.5, Opus 5 and Fable 5. They are not a quality ladder where the last one is always right. They are sized for different jobs.

• Sonnet 5 is the daily driver. Most of what you do belongs here.
• Haiku 4.5 is the fast one. Reach for it on high volume, low ambiguity work where speed is what you are actually buying.
• Opus 5 and Fable 5 are for the tasks with many moving parts, where the model has to hold a lot at once and reason across it.

The honest version of the video's joke: Haiku is not a worse model, it is a faster one, and speed is a real trade you make on purpose. Anthropic publishes no benchmark that says "use this one for wrong answers" and this guide is not going to invent one. What is true is that running everything on the largest model spends your limits on work that did not need them.

The practical habit: pick the model when you start the conversation, not after it disappoints you.

This is the one that surprises people, and it comes straight from Anthropic's help centre. On memory and Projects, the wording is exact:

"Each project has its own separate memory space and dedicated project summary, so the context within each of your projects is focused, relevant, and separate from other projects or non-project chats."

Read that twice. A Project is not a folder that adds context. It is a boundary that removes it. Inside a Project, Claude is working from that Project's memory, not the general memory it has built up with you everywhere else.

That isolation is the feature. It is exactly what you want when a client's work should not leak into another client's, or when confidential planning should stay separate from general operations. It is exactly what you do not want when you open a Project out of tidiness and then wonder why Claude has forgotten how you like things done.

• Use a Project when separation is the point: a client, a confidential workstream, a body of work with its own reference files.
• Stay in ordinary chats when continuity is the point, which is most day to day work.

Memory is optional and it is off until you switch it on. It lives in Settings, under Capabilities, and it comes with granular controls over what is kept.

The reason to do this on purpose rather than by accident is that memory only pays off once it has something in it. Turning it on today and forgetting about it is the entire strategy. In a month, the difference between a Claude that knows how you work and one that does not is a single toggle you flipped and never thought about again.

Chat search lets Claude find and build on previous conversations rather than making you re-explain. If you find yourself pasting the same background into a new chat for the third time, that is the signal.

People summarise a document into a prompt and then ask about the summary. Upload the actual source. Everything downstream, including how much Claude can check itself against, depends on whether the real thing is in the conversation or a paraphrase of it is.

These five are about the browser, which is where most people's work already lives.

Claude in Chrome sees the page you are on and takes action in it. Anthropic's description of what that means is specific: clicking links, typing text, navigating between pages, and filling out forms, using your existing logins.

That last clause is the part that matters. It is not calling an API on your behalf. It is working the same pages, signed in as you, the way a new hire on their first week would.

So brief it like one. Tell it the outcome, tell it where to start, and check the first few runs before you trust it with anything that submits.

Claude can work across multiple tabs at once. Drag the tabs you want it to see into Claude's designated tab group and it can view and interact with all of them together, instead of you switching back and forth and pasting between them.

This is the difference between "read this page" and "compare these four things", and almost nobody who has the extension installed is using it.

It ships with built-in knowledge of how to navigate Slack, Google Calendar, Gmail, Google Docs and GitHub. On those, plain instructions like "schedule a meeting" or "update the doc" work without step by step directions.

Save the detailed instructions for the tools it has never seen.

Before it acts on a page, a separate safety check reviews the action for risk and for instructions hidden in the site itself, and surfaces anything that looks wrong.

That second half is worth understanding. A page can contain text written to be read by an AI agent rather than by you. The check exists because that is a real category of attack, and the correct response when it flags something is to look, not to click through.

The extension is generally available on all paid plans, and it works alongside Claude Code, Cowork and the desktop app rather than instead of them. The browser is one surface of the same assistant, not a separate product.

These five are where the ceiling actually moves.

The Code tab in the desktop app is an interactive coding surface with access to your local files. It reads the codebase, modifies files, runs tests, makes commits and works across multiple files at once.

The part that changes who this is for: people who have never written a line of code build working tools with it by describing what they want in plain language. It shows you the changes and asks for approval before applying them.

It requires a Pro, Max, Team or Enterprise subscription.

The Code tab can run your dev server and render a live preview in the app. Claude sees the running application, tests endpoints, reads logs and iterates on what it observes.

This is the difference between an assistant that writes code and one that checks its own work. Give it the ability to run what it built and the number of round trips through you drops sharply.