--- title: "Claude Code Setup for Product Managers" description: "Install Claude Code, log in, write a CLAUDE.md file, choose a permission mode and get one useful product artefact out of it in a single sitting as a PM." canonical_url: "https://builderscamp.com/guides/tools/claude-code-setup-for-product-managers" date_published: "2026-09-18" date_modified: "2026-09-18" author: "Andre Albuquerque, Guilherme Salgueiro" publisher: "Builders Camp" guide_class: "tools" --- # Claude Code setup for product managers: install to first useful output **TL;DR:** A working Claude Code setup takes one install command, one login and one project folder, and Anthropic's quickstart confirms the install by printing a version number. The part that decides whether the tool sticks is the CLAUDE.md file you write afterwards, kept under 200 lines, and starting in a permission mode that asks before it edits. ## What do you need in place before the install command? Three things, and none of them is a coding background. You need a terminal you can type into, a folder to point the tool at, and an account. Anthropic's quickstart lists the valid account types as a Claude subscription on the Pro, Max, Team or Enterprise plans, a Claude Console account with prepaid credits, or access through a supported cloud provider such as Amazon Bedrock or Microsoft Foundry. Check which one your company already pays for before you put a personal subscription on your own card, because the Console route creates a workspace for centralised cost tracking that your finance team may already be using. The folder matters more than most setup guides admit. Claude Code is a file-based tool: it reads and edits what is on disk, and in Manual mode it writes only inside the directory it was started in and that directory's subfolders. If you have repository access, a real project is the best starting point. If you do not, a folder holding your PRDs, research notes and a CSV export is a perfectly valid working directory, and it is where most product managers get their first genuinely useful output. ## Installing Claude Code takes one command, and confirming it takes one more On macOS, Linux and WSL, Anthropic's quickstart gives a single curl command that pipes an install script to bash. Windows users get a PowerShell equivalent, and there are Homebrew and WinGet packages as well, with the note that Homebrew and WinGet installs do not auto-update while the native install does. Pick the one your machine is already set up for rather than the one that sounds most official. Then confirm it. Run `claude --version` and check that it prints a version number followed by the words Claude Code. This step gets skipped constantly and it is the cheapest debugging you will ever do: a half-finished install produces confusing failures ten minutes later, when you have already started blaming your prompt. Log in by running `claude` in a project directory and following the browser prompt. Credentials are stored after the first time, so this is a once-per-machine step rather than a daily one. ## Your first session should be a read, not an edit Open the folder you picked, start a session, and ask a question that changes nothing: what does this project do, or what is in this folder and how is it organised. Anthropic's quickstart opens with exactly this move, and the reason is practical rather than ceremonial. A read-only first question tells you whether the tool can actually see what you think it can see, before any change depends on that assumption. For a product manager, the first read is also a cheap test of context quality. If you ask what this folder is about and the answer is vague, that is not the model being weak. It is a folder with no README, no naming convention and three versions of the same spec, which is the same problem your new joiners have. ## CLAUDE.md is the setup step that decides how the next 50 sessions go Each session starts with a fresh context window. CLAUDE.md is the file that carries your instructions across that reset, and it is the difference between a tool you re-explain yourself to every morning and one that already knows how your team works. A project file lives at `./CLAUDE.md` or `./.claude/CLAUDE.md` and is shared with your team through version control; a personal one at `~/.claude/CLAUDE.md` applies across all your projects. Anthropic's guidance on writing these files is specific, and it cuts against the instinct to write everything down. Target under 200 lines per file, because longer files consume context and reduce how consistently the instructions get followed. Write rules concrete enough to verify: "run npm test before committing" rather than "test your changes". Review them periodically for contradictions, since two conflicting rules mean the agent picks one arbitrarily. Running `/init` generates a starting file from the codebase, which is a faster first draft than a blank page. The rule of thumb that works for product work is narrower than the rule for engineers. Put in the things you would tell a contractor on day one: what the product is, who the users are, which documents are canonical and which are stale, and the naming convention for specs. Leave out anything that only matters for one task. ## Manual mode is the right starting permission mode for a PM Anthropic's security documentation describes a permission-based architecture: in Manual mode, Claude Code starts with read-only permissions and asks before it edits files or runs commands that change your system, running a small built-in set of read-only commands like `ls` and `git status` without asking. There is also an auto mode, where a separate classifier model reviews actions instead of you, and a plan mode you reach with Shift+Tab that lets the agent read and propose without changing anything. Start in the mode that asks. The approval prompts are annoying by design, and the annoyance is doing real work in week one: it is the only thing forcing you to read what the agent is about to do while you still have no instinct for which requests are risky. Turn the friction down once you can predict the answer before the prompt appears, not before. ## Three setup mistakes that cost product managers their first sitting - Installing, then immediately asking for a large change, so the first thing you have to evaluate is a multi-file diff in a tool you have used for four minutes. - Writing a 600-line CLAUDE.md on day one, which is both more than you know about your own conventions and more than the guidance recommends carrying in every session. - Letting one session run all afternoon. Anthropic's best-practice guidance is blunt that the context window is the most important resource to manage and that performance degrades as it fills, so `/clear` between unrelated tasks is a habit, not a reset button for when things go wrong. ## What does "first useful output" mean if you do not ship code? Something you would have written by hand this week. A release note drafted from the actual commit history rather than from your memory of it. A spec rewritten against the code that exists, with the three places where the two disagree listed explicitly. A support export read end to end and grouped, with the row numbers kept so you can check any claim. None of these needs you to approve a code change, and each one produces an artefact a colleague will read. That is the honest scope of a first sitting. You will not ship a feature, and a setup that promises you will is selling you something. What you should have at the end of 90 minutes is an installed tool, a short context file, one artefact, and a clear sense of which of your weekly tasks are file-shaped enough to move here next. ## Where the setup ends and the workflow begins Setup is the cheap part. The part that takes practice is knowing which product tasks survive being handed to an agent, how to constrain a request so a small change stays small, and how to read a diff for the thing you did not ask for. Builders Camp's Claude Code for Product Managers bootcamp runs that in 1 week across 2 live sessions and 8 self-paced microlessons, and its practical challenge is a regression review: a PM accepts a clean-looking change, and every form submission silently stops saving the next day. [See the Claude Code for Product Managers bootcamp](https://builderscamp.com/bootcamps/claude-code-for-product-managers?utm_source=guide&utm_medium=organic&utm_campaign=claude-code-setup-for-product-managers) If your goal is building real systems rather than PM workflows, Building with Claude Code goes further into context architecture, skills, agents and automation pipelines, and sits inside the AI Agentic Builders Expert Track alongside the agent design bootcamps. For the wider toolkit question, see [the best AI tools for product managers](https://builderscamp.com/guides/tools/best-ai-tools-for-product-managers), and for the file-based habits that carry over from chat tools, [ChatGPT for product managers](https://builderscamp.com/guides/tools/chatgpt-for-product-managers). The idea underneath the CLAUDE.md file has its own entry in [context architecture](https://builderscamp.com/guides/glossary/context-architecture). ## Frequently asked questions ### Do I need to be able to code to set up Claude Code? No, but you do need to be comfortable opening a terminal and running a command you were given. Installation is a single command, login happens in the browser, and the first session is typed in plain English. What you cannot skip is reading the output: Claude Code shows you the files it touched, and a product manager who accepts every change without looking at it has bought speed with risk. ### What do I need before I run the install command? A terminal, a project folder, and an account. Anthropic's quickstart lists a Claude subscription on the Pro, Max, Team or Enterprise plans, a Claude Console account with prepaid credits, or access through a supported cloud provider. If your company already pays for one of these, ask before buying a personal plan. ### How do I know the install actually worked? Run claude --version. Anthropic's quickstart says the command prints a version number followed by the words Claude Code. If it prints nothing or reports a command not found, the install did not finish and no amount of prompting will fix it. ### What is CLAUDE.md and do I need one on day one? CLAUDE.md is a plain markdown file Claude Code reads at the start of every session, holding the instructions you would otherwise retype. You do not need one to run your first session, and you will want one by your third. Anthropic recommends keeping each file under 200 lines, since longer files consume context and reduce how consistently the instructions are followed. ### Which permission mode should I start in? Manual mode, where Claude Code begins with read-only permissions and asks before it edits files or runs commands that change your system. Anthropic's security documentation describes this as the permission-based architecture the tool is built on. Moving faster later is a deliberate choice, not a default you should inherit on day one. ### Can I use Claude Code if I do not have access to my company's repository? Yes. Point it at a folder of product documents instead of source code. Claude Code reads and writes files, so a directory of PRDs, research notes and exports is a valid working directory, and it is the setup most product managers actually start with. ### How long should the first sitting take? Budget 90 minutes. Install and login take a few minutes, the first read-only question takes one, and the rest goes on the part that matters: writing a short CLAUDE.md and getting one artefact you would have written by hand anyway. ## Sources - [Anthropic: Claude Code quickstart](https://docs.claude.com/en/docs/claude-code/quickstart) - [Anthropic: How Claude remembers your project](https://docs.claude.com/en/docs/claude-code/memory) - [Anthropic: Claude Code security](https://docs.claude.com/en/docs/claude-code/security) - [Anthropic: Best practices for Claude Code](https://docs.claude.com/en/docs/claude-code/best-practices) ## How this guide was made Researched from Builders Camp's bootcamp, track and masterclass material and the sources listed on this page, drafted with AI, and fact-checked against every source cited.