Learn Claude Code, one small step at a time

Claude Code writes CLAUDE.md, your daily plan and your notes in Markdown, which looks like text cluttered with special characters when you open it in the terminal or a plain text editor. In Obsidian you open your working folder, read those files with proper

In one conversation an AI agent can work through a spreadsheet that usually costs you hours, and in the next it hands you a month's worth of tasks when you asked for two for today. Find out which first-week stumbles are normal and what

The new Sonnet is already in Claude Code, but Opus stays the default model. The past week also brought mods, along with changes to /effort, the VS Code extension and MCP server handling in the terminal.

A few weeks into working with Claude Code, its memory and your CLAUDE.md file fill up with notes that nobody checks any more. Once a month you can have Claude Code check them and get a short list to decide on after every review.

Before an AI agent drafts a LinkedIn post or a reply to a client's enquiry for you, it has to find out somewhere who you are, who you're writing to and what you sell. Here's what goes into each of the three files it

A new default model isn't the only change this week that touches your limit – Anthropic also added something you won't want to waste. See what changed in pricing, in effort and in the limit, and what's new in the terminal, VS Code and

A skill that puts together your plan for the day each morning checks the inbox and the calendar one after the other, in a single conversation and on a single model. Part of that work can go to subagents – which leaves the question

You start a new conversation, and Claude Code asks again what you do and who you work for, even though memory is switched on. We show you where to keep the facts that have to be there every time, not only when the AI

Claude Code can now read AGENTS.md instead of CLAUDE.md, skills and plugins switched on in your claude.ai account turn up in terminal sessions by themselves, a click changes a value in the /config panel, and at the end we show a scientific paper you

The Gmail MCP connector from Anthropic does exactly what its creators designed it to do, and no more. MCP uses a selection of the API's calls. Today we show you how to give AI access to the Google API, so that it can do

Claude Code tells you it has no tool for an operation, and the matter looks closed. Usually it isn't: several other routes lead into the same service, and a single sentence asks the AI to check them.

Anthropic announced that from today it is raising the weekly limit in Claude Code by 25 percent, and yet the limit you have today is lower than last week's. Both numbers are true. They just count from different starting points, and we explain where

You answer the same kind of message for the hundredth time, and what the AI suggests reads like a support-desk template. Fix that – teach Claude Code your own voice, so the replies sound like you wrote them.

An AI assistant running on your computer does not mean your client data stays there. Find out what can reach the model provider, what is stored locally and how data masking works.

Two top-shelf models in one week: Anthropic first, then OpenAI. We are not settling which one is better, but two launches like that side by side show how fast AI is moving.

You start every day with the same questions to Claude Code: what's waiting in the inbox, what's in the calendar, and what's left over from yesterday. That set of instructions can be written once and called with one command from then on.

Your service isn't in the connector directory? That still doesn't mean Claude Code can't reach it. Two routes are still open, and one of them holds up where a connector gives in.

Copying rows out of one panel into another, filling the same form for the tenth time – those are jobs Claude Code can now click through in your own browser. It also shows you the rules auto mode goes by, and writes its own

Claude doesn't look into your inbox or your calendar on its own. You switch both on in the browser, in a few clicks, and switching them back off is just as quick.

You ask Claude Code for a change in a document and something else comes back? Or you find yourself explaining the same thing over again inside one conversation? We show you what a request like that leaves to guesswork, and how to close the

Your usage limit ran out halfway through a task, and getting going again meant coming back to the computer and telling Claude Code to carry on. It does that by itself now, and one setting turns it off if you would rather stay in

You start every conversation explaining who you are and how you work. Answer once, in an interview: Claude Code will write a profile.md file from it and reach for it whenever the conversation is about your work.

You've just made a new folder and someone tells you to start with /init? Or you've been working in the same folder for months and wonder what that command would even do there? We'll show you when /init saves you the work, when it

You might be wondering whether anyone can tell that Claude had a hand in the text you just wrote. Anthropic has announced it will start marking that text – just not visibly, and not yet.

You explain the same things to Claude Code in every new chat: what you call your files, what nobody should touch, what matters to you. Some of that the AI can write down on its own. This lesson shows you whether it already has.

Somewhere around the fourth prompt of the morning, "yes" stops being a decision and becomes a keystroke. Some people want it that way and some want the opposite, every step in front of them – one setting decides which of the two you get,

From 14 August, Claude Code will stop asking your permission before every command – in newly opened sessions, on Pro, Max and Team. If you have been hitting yes without reading, this change is for you; if you would rather watch every step, here

Single, simple requests to Claude Code are behind you. This lesson walks one task through the whole cycle: a plan before anything moves, the work itself, and checking the result with your own eyes.

Every morning you type the same path before the actual work can start? Or maybe you've launched Claude Code in the wrong folder without noticing. We'll show you how to set up your terminal so it opens straight in the folder you keep for

Anthropic published its own investigation into three incidents where Claude models reached out of the test lab, connectors move to a new MCP spec, and Dario Amodei explains why he doesn't want open-weights models banned.

Starting Claude Code in a folder that already holds your files: /init reads what is there and writes a CLAUDE.md out of it, and you read the result and dictate the corrections.

Do you type the same thing at the start of every conversation with Claude Code? Or has a long session ever drifted away from something you asked for at the beginning? Here is which requests belong in a file and which can happily stay

Claude Opus 5 becomes the default Opus model, /code-review moves to the background, the subagent nesting limit goes back up, and a new connector lets you ask Claude how its own users are actually using AI, occupation by occupation.

Creating your first CLAUDE.md file: how to have Claude Code write it, what to tell it in the interview, how to check that it really gets read, and what to do when an instruction is ignored.

You open Claude Code's configuration and find a column of switches with not one sentence of explanation? Or maybe one of them has been flipped for months and nobody remembers why? We go through the ones that matter, in groups.

This week in Claude Code: manual skill invocation via /verify and /code-review, filesystem isolation controls, and Anthropic grants for rare genetic disease research.

A pass through Claude Code's settings on day one: model and effort, thinking, automatic conversation compaction, permission mode – plus the place on disk where all of it ends up.

Picking a model looks like a detail, right up until the week you start bouncing off your plan's limit. In this guide we'll try to help you match a model to the work you do, and show what jumping between them costs.

This week in Claude Code: auto mode without opt-in on Bedrock/Vertex/Foundry, a screen reader mode, new /fork and /subtask, and stronger abuse safeguards.

Installing Claude Code step by step: where to find the terminal, one install command, logging in through the browser, and a first instruction given inside your own folder.

Sooner or later you have to pick a plan, and you may be wondering whether Pro is enough or whether to go straight for Max. We'll try to help you make that call.

This week: Sonnet 5 becomes the default model, Claude Science opens up to research labs, and Claude Code picks up a handful of small but useful fixes.

Setting up an account on claude.ai step by step: logging in without a password, the code from your inbox, phone verification, and a first conversation. Including two points worth pausing at, because those choices can't be undone later.

When you are starting out with AI, the first question is usually which model to pick. The honest answer is that ChatGPT wins outright on a few things, and Claude earns its place somewhere else.