AI development environment in 2026

In 2026 software development changed for good: an AI assistant is now as fundamental to your setup as your code editor or Git. But to work at full power, the assistant needs a properly configured environment — a terminal, Node.js, version control and connected tools.

In this guide we go all the way from scratch: from a blank laptop (macOS or Windows) to a complete environment with the Claude Code AI assistant connected — one that reads your project, writes code, runs commands and plugs into external services via MCP.

The core idea of 2026: you no longer "type code by hand, occasionally asking the AI". You describe the task and the assistant works inside your project — with access to files, the terminal and tools. Your role shifts from typing characters to framing tasks and verifying results.

What we'll build

A final stack that fits almost any modern web development:

• Terminal — your command center (iTerm2 / Windows Terminal)
• Package manager — Homebrew (Mac) or winget (Windows)
• Node.js + pnpm — JavaScript runtime and a fast package manager
• Git + GitHub — version control
• VS Code — code editor with the right extensions
• Claude Code — the AI assistant in your terminal and editor
• MCP servers — connecting the assistant to GitHub, databases and other services

Placeholder: overview of the finished environment.

Step 1. The terminal — your command center

Almost everything below is installed via terminal commands. Let's prepare the terminal first.

macOS

macOS ships with the built-in Terminal (Applications → Utilities → Terminal). For comfort I recommend iTerm2 — better search, tabs and panes. The built-in one is fine to start.

Windows

On Windows, install Windows Terminal — a modern tabbed terminal. On Windows 11 it's already built in. If not, grab it from the Microsoft Store or via winget later.

Windows tip: for web development many choose WSL2 (Ubuntu inside Windows) — you get a Linux environment where most "Mac-style" instructions work one-to-one. It's optional; in this guide we give native Windows commands.

Placeholder: terminal on both operating systems.

Step 2. Package manager

A package manager lets you install software with a single command, no manual installers.

macOS — Homebrew

Install Homebrew with one command:

After installing, follow the terminal hint to add brew to your PATH (on Apple Silicon it's usually commands referencing /opt/homebrew/bin/brew). Verify:

Windows — winget

Modern Windows 10/11 already include winget, the built-in package manager. Verify:

If the command isn't found, update "App Installer" from the Microsoft Store.

Step 3. Node.js and pnpm

Node.js is the JavaScript runtime needed for almost any frontend and for the classic Claude Code install. pnpm is a fast, disk-efficient package manager (this very project uses it).

I recommend installing Node through the version manager fnm so you can switch versions per project.

macOS

Windows

Verify and enable pnpm

Node ships with Corepack, the easiest way to enable pnpm:

Placeholder: checking Node and pnpm versions.

Step 4. Git and GitHub

Git is your version control system. Without it there's no team work and no safe experimentation. The AI assistant relies on Git too: it creates branches, commits and opens PRs.

Installation

Basic configuration

Introduce yourself once — this name goes into every commit:

SSH key for GitHub

To push without typing a password, create an SSH key and add it to GitHub:

Then GitHub → Settings → SSH and GPG keys → New SSH key → paste the key. Install the official GitHub CLI — the assistant needs it to work with PRs:

Step 5. The code editor — VS Code

A minimal set of extensions worth installing right away:

• ESLint and Prettier — linting and formatting
• Tailwind CSS IntelliSense — class hints (if you use Tailwind)
• GitLens — supercharged Git
• Claude Code — the official extension that integrates the assistant into the editor

After installing, open a project with code . from its folder.

Placeholder: VS Code extensions panel.

Step 6. Installing Claude Code

This is the core of our AI environment. Since 2026 the recommended method is the native installer: it ships a self-contained binary with no dependencies and auto-updates in the background.

macOS / Linux

Windows

In PowerShell:

Or via winget:

Verify

claude doctor diagnoses your environment and flags anything missing (PATH, permissions, etc.).

Requirements: macOS 13 (Ventura) or later; Windows 10 build 1809 or later (including Windows 11).

Placeholder: output of claude --version and claude doctor.

Step 7. First run and sign-in

Go to any project folder and launch the assistant:

On first run it will prompt you to sign in. Use the login command and follow the browser flow:

You can sign in with a Claude (Pro/Max) subscription or a Claude API key billed by usage. For regular work a subscription is usually more cost-effective and predictable.

Placeholder: the /login flow.

Step 8. First steps with the assistant

Now the fun part. Claude Code launches right inside your project folder and sees your files.

Your first prompt

Just describe the task in plain language:

The assistant reads the relevant files and answers. Then ask for changes:

Modes

• Normal mode — the assistant asks permission before changes and commands.
• Plan Mode — it drafts a plan first without changing anything; you approve, then it executes. Ideal for larger tasks.
• Auto-accept mode — runs steps without confirmations; enable it deliberately.

Useful slash commands

CLAUDE.md — project memory

A CLAUDE.md file at the repo root is the assistant's "persistent memory": rules, commands, conventions. Create it with /init or by hand:

The more precise your CLAUDE.md, the fewer mistakes and the less the assistant has to ask.

Placeholder: the assistant answering the first prompt.

Step 9. Connecting MCP — the assistant's superpower

MCP (Model Context Protocol) is the standard for connecting the assistant to external tools: GitHub, databases, file systems, APIs. It turns the assistant from a "code writer" into a full agent that works with your services.

Basic syntax:

Examples:

Managing servers:

Scopes

• local — just you, in this project
• project — shared with the team (stored in .mcp.json, committed to git)
• user — personal, available across all your projects

For team servers use the project scope and commit .mcp.json — then the whole team shares the same toolset.

Placeholder: list of connected MCP servers.

Step 10. Real-world workflows

The environment is ready. Here are the workflows it was all for.

A new feature (plan → implementation)

You approve the plan — the assistant implements it step by step, creating files and running the build.

Refactoring

Debugging from an error

Just paste the error text:

Review before committing

Tests

AI best practices

• Keep CLAUDE.md current. It's the cheapest way to improve answer quality.
• Use plan mode for complex tasks. Plan first, code second. Fewer redos.
• Work in small iterations. One task, one clear result that's easy to verify.
• Commit often. A clean git diff lets the assistant (and you) see exactly what changed.
• Verify, don't blindly trust. AI speeds you up, but the code is your responsibility. Run tests and builds.
• Never paste secrets into chat. Keep keys and passwords in .env, not in prompts.
• Match the model to the task. Complex architecture — Opus 4.8; routine — Sonnet 4.6; quick odds and ends — Haiku 4.5.

Command cheat sheet

Conclusion

In a single pass you've assembled a modern development environment and connected an AI assistant to it. Routine work — project setup, boilerplate, tests, debugging — now happens several times faster, freeing you to focus on what matters: the product and the architecture.

The key is not to stop at installation. The real payoff comes when you build a process: a precise CLAUDE.md, plan mode for big tasks, tools connected via MCP, and the habit of verifying results.

If you need help setting up AI workflows for your team or building a project end-to-end — get in touch and let's talk.