Claude Code in 2027: What It Is, Pricing and How to Use It (Even If You’re Not a Developer)

39 min read

Claude Code is the AI tool that has changed the way we work at DigiAcción more than any other. Not because it writes code faster than anyone, but because it does things: it opens your files, changes them, runs programs, checks whether they work and fixes whatever breaks.

That makes it something very different from a chatbot. And it opens a door that, until recently, was closed to anyone who doesn't code: building your own tools. A report that writes itself, an internal form, a landing page prototype, a script that cleans up the Excel file nobody wants to touch.

At DigiAcción we use it every day to build internal tools, content generators and website prototypes. And precisely because we use it so much, I know where it shines and where it can get you into trouble: security, personal data, maintenance and SEO.

In this guide I'll walk you through what Claude Code is, how much it costs, how to install it and how to use it well even if you're not a developer, and when an AI-built prototype is enough versus when you need a professional, with 2027 already in mind. Every price and every feature has been checked against each tool's official documentation.

Definition

What is Claude Code?

Claude Code is Anthropic's AI coding agent. It understands the project you're working on, reads and edits files, runs commands and checks whether the result works. You use it by describing what you want in plain language, from the terminal, a desktop app, your code editor or the browser.

The key word is agent. A chatbot answers you, and then you copy and paste. Claude Code acts: you give it a goal and it decides which files to touch, makes the changes, tests that they work and shows you what it did. You approve, correct or undo.

That's why so many people search for it as “Claude CLI”: it started out as a command line interface (CLI) tool for the terminal, and that's still its most complete version. Although its home turf is code, Anthropic's documentation points out that it helps with any task you can do from the command line: writing documentation, searching through files or putting together reports.

What's the difference between Claude and Claude Code?

Claude is Anthropic's conversational assistant: the claude.ai website and the desktop and mobile apps. Claude Code is its coding agent. They use the same models, but Claude Code works inside your project, with access to its files and permission to take actions.

	Claude (chat)	Claude Code
What it does	Answers, writes and analyzes whatever you give it	Works inside your project: reads, creates and edits files
Access to your files	Only the ones you upload to the conversation	The project folder you open for it, plus whatever you allow outside of it
Running programs	In an isolated Anthropic environment, not on your computer	On your machine or in a cloud session, with a permission system
Checking its own work	You test it yourself	It can run tests, read the errors and fix them
Project memory	Projects and chat memory	A CLAUDE.md file and automatic project memory
Plan required	Free or paid	Paid plans or an API account only

How does Claude Code work?

It works in a loop. According to the official documentation, every task goes through three phases that repeat as many times as needed: gathering context (finding and reading the files that matter), taking action (editing, creating, running) and verifying (testing the result). You can step in at any moment to steer it in a different direction.

• Your request
• Gathers context
• Takes action
• Verifies
• Repeats until done

The brains behind it are Claude's models. On the Pro, Max, Team and Enterprise plans and on the API, the default model is Opus 5.5. With the /model command you can switch to Sonnet 5.5 for everyday work, to Haiku 5.5 for simple, fast tasks, or to Fable 5.1, the most capable one, built for long, autonomous jobs. Fable isn't turned on by default: you have to choose it, and on Pro it's paid for with usage credits.

Platforms

Where Claude Code runs: terminal, desktop, IDE, web and mobile

Claude Code runs in the terminal, in the Claude desktop app, in VS Code and JetBrains, in the browser and from the mobile app, and they all share the same engine. Your project instructions (the CLAUDE.md file), your settings and your connections work across all of them.

Where	What it is	Who it's for	What you need
Terminal (the CLI)	The most complete version: you type claude in the project folder	People comfortable with the command line; automations and scripts	Install it with one command
Desktop app (Claude Desktop)	The Code tab in the Claude app for Windows and macOS (in beta on Linux)	People who'd rather not use the terminal and prefer to review changes visually	Download the app and have a paid plan
VS Code (and Cursor)	An extension with inline changes, file mentions and plan review	People who already work in that editor	Install the extension
JetBrains	A plugin for IntelliJ IDEA, PyCharm, WebStorm and others	Development teams	The plugin, plus the CLI installed separately
Claude Code on the web (claude.ai/code)	Cloud sessions that keep working even if you close your laptop	Long tasks on GitHub repositories	A Pro, Max or Team plan (premium seats on Enterprise) and a connected GitHub account
Mobile (Claude app for iOS and Android)	A remote control: you start, follow and steer sessions running in the cloud or on your computer	Reviewing or assigning tasks when you're away from the office	The Claude app

It also integrates with other tools: with Slack (on Pro and Max, you mention @Claude in a channel with a bug report and it comes back with a proposed fix), with GitHub Actions and GitLab for automated code review, and with Chrome to test websites in your own browser using the sessions you're already logged into.

Which option should you pick if you're not a developer?

The desktop app. It gives you the same engine as the terminal with a graphical interface: you pick a folder, type what you want and see exactly what changes in each file before you accept it. It also has a browser panel that opens the website or app you're building, so you can see the result without leaving the app.

Once you lose your fear of it, the terminal is worth it: that's where every feature lives and where you automate things from. Anthropic even has a guide for people who have never opened a terminal.

Use cases

What Claude Code is used for if you're not a developer

You can use Claude Code for any task that gets solved with files and programs: building internal tools, automating repetitive processes, transforming data, generating reports and putting together website prototypes. You don't need to know how to code; you need to know how to explain what you want and how to check whether what it delivers is right.

At DigiAcción it doesn't replace anyone on the team: it takes the mechanical work off their plate and frees up more time for what needs judgment, as you can see in our 13 real websites designed with AI and reviewed by the team.

Can you use Claude Code with no coding experience?

Yes, and Claude Code is one of the most direct ways to do it. Anthropic offers a desktop app so you don't have to use the terminal, plus a step-by-step guide for people who have never opened one. What you do need is something else: knowing how to explain clearly what you want, reviewing what it does and being clear about what you won't let it touch.

My caveat: not knowing how to code doesn't stop you from building, but it does stop you from judging whether what you built is safe. For a tool you use yourself, on your own computer and with test data, the risk is low: think of the front desk at a hotel in Andorra putting together a script to sort the day's arrivals. For anything that will live on the internet or touch customer data, you need someone who knows how to review it.

Claude Code removes the barrier of writing code. It doesn't remove your responsibility for what you publish.

Pricing

Claude Code pricing: how much Pro, Max, Team and the API cost

Claude Code is included in every paid Claude plan, but not in the free one. The cheapest way in is Pro, at $20 a month, or $17 a month if you pay for the whole year up front ($200). You can also use it without a subscription, paying for what you use through an API account.

These are the plans published on Anthropic's official website:

Plan	Price	Includes Claude Code?	Usage included
Free	$0	No	Chat only
Pro	$20/month, or $17/month billed annually	Yes	At least 5x the usage of Free per five-hour session
Max 5x	$100/month	Yes	5x the usage of Pro, priority access and $100 a month in API credits
Max 20x	$200/month	Yes	20x the usage of Pro, priority access and $200 a month in API credits
Team, standard seat	$25/user/month, or $20 billed annually	Yes	More usage than Pro; for teams of 2 to 150 people
Team, premium seat	$125/user/month, or $100 billed annually	Yes	5x the usage of a standard seat
Enterprise	$20/seat/month (annual, 20-seat minimum) plus usage at API rates	Yes	Based on usage, with security and admin controls
API (Claude Console)	Pay per token	Yes	No subscription: you pay for what you use

Prices in US dollars and before tax, according to claude.com/pricing. Anthropic charges in local currency where it can, so you'll see the exact amount at checkout.

Is Claude Code free?

Not on Claude's free plan: Claude Code is a paid feature, starting with Pro. There are two exceptions. The Claude for Open Source program gives six months of Max 20x to open source maintainers who meet its requirements. And Claude Code can be connected to other models, for example through Ollama, although at that point you're no longer working with Claude's models.

Is Claude Code unlimited? Pro and Max usage limits

No. Every plan has limits that reset in five-hour windows, and the paid plans add weekly limits on top. What's more, the Claude chat, the app and Claude Code all draw from the same allowance: if you spend the morning chatting, you'll have less left for coding.

When you hit the limit, you can wait, upgrade your plan or turn on usage credits to keep going at API rates. Type /usage to see where you stand.

Which models does Claude Code use, and what do they cost on the API?

If you use Claude Code through the API, you pay per million tokens (a token is a chunk of a word). These are the rates for the current models:

Model	Best for	Input (per million tokens)	Output (per million tokens)
Fable 5.1	Long, autonomous jobs	$10	$50
Opus 5.5	The everyday choice for agentic coding	$4	$20
Sonnet 5.5	High performance at a lower cost	$2	$10
Haiku 5.5	The fastest and cheapest	$0.10	$0.50

Haiku 5.5: price for requests of up to 100,000 tokens. Batch processing comes with a 50% discount. Source: claude.com/pricing.

So how much do people actually spend? The only official benchmark comes from companies: according to the Claude Code documentation, enterprise deployments average around $13 per developer per active day ($150 to $250 a month), and 90% of users stay below $30 a day. Those are technical teams working with it every day; your own usage will depend on the model, the size of the project and how much you use it.

Which Claude Code plan should you choose?

I'd choose based on actual usage, not on the fear of running short. This is the table I'd use to decide:

If…	I'd choose	Why
You want to give it a serious try for a few weeks	Pro, billed monthly	It's the cheapest way in and you can cancel anytime
You use it daily and run out of allowance by mid-day	Max 5x	Five times more usage per session, without switching tools
You run several sessions in parallel or very long tasks	Max 20x	Built for people who work with Claude all day long
You're a team and want centralized billing and control	Team	Admin tools, single sign-on and standard or premium seats
You'll use it in automations or only now and then	API	You only pay for what you use and can set spending limits
You care that your code isn't used for training by default	Team, Enterprise or API	Under commercial terms, Anthropic doesn't train on your code unless you allow it

My advice: start with Pro billed monthly and move up to Max only if you're running short every day for a couple of weeks.

Installation

How to install Claude Code step by step (Windows, Mac and Linux)

There are two ways to install Claude Code: downloading the Claude desktop app, which doesn't need the terminal, or running a single command in the terminal. Either way, you need a paid account (Pro, Max, Team or Enterprise) or an API account in Claude Console.

Before you start, check the official requirements:

• Operating system: macOS 13 or later, Windows 10 (version 1809) or later, Ubuntu 20.04 or later, Debian 10 or later, or Alpine Linux 3.19 or later.
• Hardware: 4 GB of RAM or more, with an x64 or ARM64 processor.
• Connection: internet, always. Claude Code doesn't work offline.
• Location: you need to be in a country where Anthropic offers its services. The US, Spain and Andorra are all on the list.

• Choose your way in
• Install
• Sign in
• Open a folder
• Ask and review

Option A: Claude Code Desktop, the desktop app with no terminal

1. Download the Claude app for macOS or Windows from the official website (on Linux it's in beta and installs with apt).
2. Open it and sign in with your Anthropic account.
3. Click the Code tab. If it asks you to upgrade, you're on the free plan.
4. In the row below the text box, choose the Local environment and, with “Open folder…”, select the project folder.
5. Pick the model, type what you want and review the changes before you accept them.

The app already includes Claude Code: there's nothing else to install.

Option B: Claude Code CLI in the terminal

Open the terminal (on a Mac, the Terminal app; on Windows, PowerShell) and paste the command for your system. It's the method Anthropic recommends, and this installation keeps itself updated in the background.

macOS, Linux or WSL

curl -fsSL https://claude.ai/install.sh | bash

Windows (PowerShell)

irm https://claude.ai/install.ps1 | iex

Windows (Command Prompt, CMD)

curl -fsSL https://claude.ai/install.cmd -o install.cmd && install.cmd && del install.cmd

You can also install it with Homebrew (brew install --cask claude-code) or with WinGet (winget install Anthropic.ClaudeCode), although those versions don't update themselves. To check that everything worked, open a new terminal and type claude --version: you should see a version number.

To get started, go to your project folder and type claude. The first time, your browser opens so you can sign in; after that, it won't ask you again.

Copy and paste

cd path/to/your/project
claude

How do you install Claude Code on Windows?

With the PowerShell command above or with WinGet. Anthropic also recommends installing Git for Windows so Claude Code can use Bash; without it, Claude Code uses PowerShell to run commands. If you'd rather not touch the terminal at all, the desktop app works on Windows with both x64 and ARM64 processors.

A practical tip: if the command fails with “'irm' is not recognized,” you're in CMD, not PowerShell. And if your computer can't find claude after installing, open a new terminal.

How do you use Claude Code in VS Code, Cursor or JetBrains?

With its extension: search for “Claude Code” in the VS Code extensions panel, install it and open it from the command palette. The same extension works in Cursor. In JetBrains IDEs, install the plugin from their marketplace; that plugin needs the CLI installed separately.

Getting started

How to use Claude Code: a tutorial for your first session

The best first session with Claude Code is small and risk-free: a test folder, made-up data and a task you can check at a glance. That way you learn how it asks for permission, how it shows you its changes and how to undo them, with nothing at stake.

For your first real task, Anthropic recommends something that works really well for people who don't code: have Claude interview you before it builds anything, and put in writing what the two of you are going to do. This is the prompt I'd start with:

Ready-to-copy prompt

I want to build [what you want to build] for [who it's for and what it's for].
Before you write anything:
1. Ask me whatever questions you need to fully understand what I need.
2. Propose a short plan: which files you'll create and what each one will do.
3. Tell me what you'll need to install and why.
Rules: don't use real customer data, don't install anything without asking me first, and explain each step in plain English, without jargon.
Once I approve the plan, build it in parts and show me how to test each one.

A good first session isn't the one that builds the most. It's the one that teaches you how to undo.

What are the basic Claude Code commands?

Eight commands are more than enough to get started. You type all of them inside the session, starting with a forward slash:

Command	What it's for
/init	Create the project's CLAUDE.md
/plan or Shift+Tab	Enter plan mode or switch the permission mode
/model	Switch models (Opus, Sonnet, Haiku or Fable)
/clear	Start a clean conversation when you switch tasks
/compact	Summarize the conversation to free up working memory
/rewind	Roll back the code, the conversation or both
/usage (or /cost)	See your usage and what's left on your plan
/help	See the help and the available commands

Method

Claude Code best practices to avoid nasty surprises

Using Claude Code well comes down to five habits: giving it context with a CLAUDE.md, asking for a plan before it acts, controlling permissions, reviewing every change and never giving it secrets. In my experience, almost every problem comes from skipping one of them.

What is the CLAUDE.md file and what goes in it?

CLAUDE.md is a text file that Claude Code reads at the start of every session. It holds the project's rules: what the project is, how the work gets done and what's off-limits. You create it with /init and edit it with /memory. If your project already has an AGENTS.md for other agents, Claude Code can read that too.

Anthropic recommends keeping it under 200 lines: in a long file, the important rules get lost and Claude stops following some of them.

CLAUDE.md template for a business project

# Project: weekly store reports
- Always reply in plain English, without jargon.
- The original data is in /original-data: read it, but never modify it.
- Work only with copies in /working-data.
- Don't install programs or libraries without asking me first.
- Before changing more than two files, propose a plan.
- Never write passwords, keys or tokens in the code.
- When you're done, explain how I can check that it works.

What is plan mode and why should you use it?

It's the mode in which Claude Code investigates and proposes a plan without touching anything, and it's worth using because the most expensive mistake is solving the wrong problem well. The Claude Code documentation suggests a four-phase workflow, and it's the one I follow:

• Explore
• Plan
• Implement
• Commit the changes

In plan mode, Claude Code reads and proposes, but doesn't edit anything until you approve. You turn it on with Shift+Tab, with /plan or by starting with claude --permission-mode plan. The documentation itself adds an important nuance: if you can describe the change in one sentence (fixing a typo, changing a color), skip the plan.

Which permission mode should you use: Manual, plan or auto mode?

Permission modes decide what Claude Code can do without asking you. Pay attention to this one: in recent versions, the terminal and VS Code start in auto mode, where a second model reviews actions on your behalf and blocks the ones it considers risky. It's convenient, but if you're just starting out, I'd begin in Manual (claude --permission-mode manual) until you understand what it does.

Mode	What it does without asking	When I'd use it
Manual	Read only	When starting out, on sensitive projects or with real data
Accept edits	Read, edit files and run basic file commands	When you're already reviewing changes in the diff view
Plan	Read and propose; it doesn't edit until you approve	Before any big change
Auto (auto mode)	Almost everything, with a classifier that blocks risky actions	Long tasks on projects you understand
Don't ask (dontAsk)	Only what's pre-approved; everything else is denied	Automations with a closed list of actions
Bypass permissions (bypassPermissions, the --dangerously-skip-permissions flag)	Everything	Never on your own computer: only in containers or isolated machines

Some actions are never approved automatically in any mode, such as deleting critical system paths, and any deny rules you add are respected in every mode.

How do you undo what Claude Code did?

With checkpoints. Claude Code saves the state of your files before each request, and with /rewind, or by pressing Esc twice, you can roll back the code, the conversation or both.

Checkpoints have two limits: they don't track changes made by terminal commands, like deleting or moving files, and they're no substitute for version control. As soon as the project matters, use git; Claude Code handles it for you if you ask it to “commit these changes with a descriptive message.”

Features

Claude Code features: skills, subagents, hooks, MCP and plugins

You can extend Claude Code with several building blocks, and you don't need to set them all up on day one. Anthropic suggests adding each one when the problem it solves shows up: if Claude gets the same thing wrong twice, it goes in the CLAUDE.md; if you keep typing the same prompt over and over, it becomes a skill.

Feature	What it is	Example for a business
CLAUDE.md	Standing instructions it reads in every session	“Reply in plain English and don't touch the original data folder”
Skills	Saved instructions or processes you launch with /name, or that load on their own when needed	/weekly-report, which repeats the same steps every Monday
Subagents	Helpers with their own context that handle a task separately and return a summary	Reviewing hundreds of files and bringing back only the conclusions
Hooks	Automatic actions that always run at a specific moment	Formatting the code after every change or notifying you when it's done
MCP	Connections to external tools and data	Reading documents from Google Drive or updating Jira tickets
Plugins	Packages that bundle skills, subagents, hooks and connections so you can install them in one go	The same toolkit across every project on the team
Scheduled tasks	Running Claude Code periodically, in the cloud or on your computer	A website check every morning
Worktrees	Separate working copies for several sessions running in parallel	One session fixes a bug while another builds a new feature
GitHub Actions	Claude Code inside GitHub, triggered when you mention @claude	Turning an issue into a pull request you can review
Agent SDK	Claude Code's building blocks as a library for Python or TypeScript	Building your own agent for a company process

What are Claude Code skills and how do you add them?

A skill is a SKILL.md file with instructions, knowledge or a multi-step process. You launch it by typing / followed by its name, or Claude loads it on its own when it needs it. Unlike CLAUDE.md, a skill's content only loads when it's used, so you can have lots of them without filling up the working memory.

To create one, you save its SKILL.md in the project's .claude/skills/ folder. To install other people's skills, the usual route is installing a plugin from a marketplace with /plugin; Anthropic's official marketplace is added by default.

For a business, this is probably the most useful feature of all: you turn the way you work into a procedure the team can repeat exactly the same way every time. At DigiAcción we have skills for processes we repeat often, like the research we do before writing an article.

What is MCP in Claude Code and what is it for?

MCP (Model Context Protocol) is an open standard for connecting AI tools to external data: with it, Claude Code can read your Google Drive documents or update tickets in Jira. Only use it with servers from providers you trust; Anthropic warns that it doesn't audit the security of any MCP server.

Vibe coding

Vibe coding: what it is, what it's good for and where it falls short

Vibe coding is programming by describing what you want in natural language and letting AI write the code, often without even reading it. The term was coined by AI researcher Andrej Karpathy, and Collins Dictionary named it its Word of the Year. It's useful for prototyping, automating tasks and validating ideas without writing code by hand.

Claude Code is one of the tools people use most for vibe coding. And it works: you can have a working prototype in a single afternoon. The problem is what you can't see. These three figures sum it up well:

I'm not saying this because I distrust the tool. In that same Veracode report, the Claude models analyzed stand out on functional tasks, but their security didn't improve compared with previous generations. Something that works and something that's secure are two different things.

How do you vibe code with Claude Code?

In a four-step loop: you describe what you want in plain language, Claude Code proposes a plan and writes the code, you test the result, and you explain what's wrong so it can fix it. The difference between doing it well and doing it badly lies in reviewing every round, not just the last one.

Where does vibe coding fall short?

It falls short on everything that doesn't show up in a demo. These are the six limits I'd always check before signing off on something built with AI:

Limit	What usually happens	What I'd do
Security	Unvalidated forms, keys written into the code, dependencies with known flaws	Ask for /security-review and don't publish without a professional review
Maintenance	Nobody understands the code a few months later and every fix breaks something else	Git from day one and decisions written down in the CLAUDE.md
SEO	Websites that look good but that Google can't read properly: content that only appears with JavaScript, generic titles, poor structure, slow on mobile	Have someone who knows SEO review indexing, speed and architecture
Accessibility	Unlabeled buttons, low contrast, forms that can't be used with a keyboard	Audit it with tools and with real-world testing
Personal data	Forms that store customer data without the necessary protection or legal notices	Have it reviewed by someone who knows the data protection rules that apply to you
Costs and performance	Paid services connected with no limits, poorly sized hosting, slow queries	Review the architecture and set spending limits

Can you build a website with Claude Code?

Yes. A real estate agency in Miami can have a clickable website with its property listings in an afternoon, running on its own computer. To validate an idea, pitch a proposal or prepare a visual brief, it's fantastic.

Publishing it and getting it to sell is another matter entirely. A business website needs a properly configured domain and hosting, secure forms, legal pages, speed on mobile, analytics, an architecture designed with SEO in mind and someone to maintain it. And if you're going to sell through it, add payments, inventory, shipping and taxes: that's the work we do as an e-commerce agency in Andorra, and I wouldn't leave it in the hands of a prototype.

When is an AI-built prototype enough, and when do you need a professional?

It depends on who uses it and on what happens if it fails. This is the decision table I'd apply:

Situation	Is Claude Code enough?	Why
An internal tool you use yourself, on your computer, with test data	Yes	If it fails, it doesn't affect anyone else
A script that transforms your Excel or CSV files	Yes, as long as you check the output	Always compare a sample against the original data
A website prototype to validate an idea or create a brief	Yes	It's a draft: its value lies in showing, not in publishing
A tool the whole team uses every day	With a review	It needs version control, documentation and an owner
Anything that touches customer data or payments	No, not without a professional	Security and data protection don't allow for trial and error
Your business website, the one that has to rank and win customers	No, not without a professional	It needs strategy, design, SEO, accessibility and maintenance

Prototype it yourself. Publish it with someone who knows what happens when things break.

Comparison

Claude Code vs Codex, Cursor, GitHub Copilot and Antigravity

Claude Code competes with other AI coding tools that do similar things: OpenAI Codex, GitHub Copilot, Cursor and Google's Antigravity. They all read projects, edit files and carry out tasks; the differences lie in where they live, which models they use and how they charge.

Tool	What it is	Where it runs	Models	Starting price
Claude Code	Anthropic's coding agent	Terminal, desktop app, VS Code, JetBrains, web, mobile and Slack	Claude: Opus, Sonnet, Haiku and Fable	From $20/month (Pro) or pay-as-you-go with the API
OpenAI Codex	OpenAI's coding agent	Desktop app, web, mobile, CLI, editor extension and cloud	OpenAI models	Included in ChatGPT from the free plan for quick tasks; Plus, $20/month
GitHub Copilot	GitHub's assistant and agent	Editors, CLI, GitHub and desktop app	Multiple providers, including Claude, OpenAI and Gemini	Free with limited usage; Pro, $10/month
Cursor	A code editor with its own agent	Editor, CLI, cloud agents and mobile	Grok on the basic plans; third-party models are paid for separately	Limited free tier; Plus, $8/month (Grok only); Premium, $30/month
Google Antigravity	Google's agent platform: app, editor, CLI and SDK	App, editor and CLI	Gemini	Free with basic weekly limits; more usage with Google AI Pro and Ultra

Prices in US dollars and before tax, according to each tool's official website. They change often, so check them before you pay. Gemini CLI, Google's open source terminal agent, has been replaced by Antigravity CLI for free and Google One users; it's still available with an API key, Code Assist or Vertex AI.

One detail almost nobody mentions: these tools aren't mutually exclusive. The Claude Code extension also installs in Cursor, and GitHub Copilot lets you use Claude models and even, on its Pro+ and Max plans, delegate tasks to third-party agents like Claude and Codex (in preview).

Claude Code vs Cursor: which should you choose?

It depends on where you want to work. Cursor is a code editor with its own agent: it's a good fit if you want to see and touch the code while the AI helps you. Claude Code is an agent you delegate tasks to from the terminal, the app or the editor, and its extension works inside Cursor, so you can use both. Keep an eye on the price: Cursor's basic plans only include Grok models.

Is Claude Code the best AI for coding?

There's no single best AI for coding in every case; there's a best one for the way you work. I'd decide using these four criteria, in this order:

If you're asking what we use: Claude Code, every day.

Claude or ChatGPT for coding? Claude Code vs Codex

For talking about code, explaining an error or generating a one-off snippet, both chatbots do the job. For working inside a project, the fair comparison isn't Claude vs ChatGPT, but Claude Code vs Codex, OpenAI's coding agent.

Both are included in their $20-a-month subscriptions (Claude Pro and ChatGPT Plus), and both run in the terminal, in the editor, on the desktop and in the cloud. The difference lies in the models and in the way they work: try them on the same task before you decide.

Summary

How to get started with Claude Code in 7 steps

To get started with Claude Code without knowing how to code, seven steps are enough: choose a plan, install it, create a test folder, write the CLAUDE.md, ask for a plan, review every change and go through a professional before publishing. If you only remember one list, make it this one.

Example

Complete example: an automatic weekly report for a physical therapy clinic

One of the best uses of Claude Code for people who don't code is automating an internal report from an Excel file. Here's a typical case, in the style of the internal tools we build: a physical therapy clinic in Madrid wants to know every Monday how many appointments each therapist had, how many were canceled and which slots were left open. Right now someone does it by hand, using the Excel file exported from the clinic's booking software.

Step	What I do or ask for	What I check
1 · Prepare	I export a month's worth of the Excel file and replace patient names with codes	That no personal data is left in the folder
2 · Context	“Read this file and explain which columns it has and what each one means”	That it correctly understood what a cancellation is and what an open slot is
3 · Rules	/init, and I add to the CLAUDE.md: “never modify the original file; explain everything in plain English”	That the CLAUDE.md is short and correct
4 · Plan	In plan mode: “I want a weekly report with appointments per therapist, cancellations and open slots; propose how to do it”	That the plan doesn't install anything strange and that the report includes what the clinic needs
5 · Build	I approve the plan: it creates a script and a report template that opens in the browser	The changes, file by file
6 · Verify	“Calculate one week's figures by hand and compare them with the report”	That they match a count done by a person
7 · Repeat	I turn it into a skill, /weekly-report	That it works the same way with the following week's Excel file
8 · Save	“Commit everything to git with a descriptive message”	That there's a version to go back to if something breaks

The result is a small tool used by one person, on their own computer, with data that contains no names. Low risk and high value: the ideal territory for Claude Code.

Now imagine the next step: patients booking from the website and the report pulling data from those bookings. Now you're dealing with health data, public forms, backups and a website that needs to show up on Google when someone searches for a physical therapist. I wouldn't make that leap with a prototype: it's a web design for clinics project, with its own strategy, security and maintenance.

What I wouldn't do

• Install Claude Code from websites that aren't Anthropic's, or buy “shared” accounts.
• Paste passwords, API keys or customer data into the chat.
• Work on the original files without a copy or git.
• Use the mode that bypasses permissions on my own computer.
• Accept changes without looking at them because “it seems to work.”
• Launch a vibe-coded website without reviewing security, SEO and accessibility.
• Pay for the most expensive plan on day one “just in case.”
• Mix ten different tasks in the same conversation instead of using /clear between them.

The bottom line

Claude Code is the first tool that truly puts programming within reach of people who don't code. If I were starting from scratch today, this would be my order:

• Pro plan
• Desktop app
• Test folder
• CLAUDE.md
• Plan before acting
• Review and undo
• A professional before publishing

Use it to build what you couldn't build before: tools, automations, reports and prototypes. And save your judgment for deciding what leaves your computer.

AI has made writing code cheap. It hasn't made mistakes in production any cheaper.

Frequently asked questions