Claude Code: install and your first agent run

I avoided Claude Code for months, and the reason is embarrassing: the word 'Code' is in the name, and I run demand generation, not engineering. I assumed it was a tool for people who are not me. That assumption cost me a quarter of doing by hand what it now does while I get coffee.

The reframe that unstuck me: Claude Code is not a coding tool, it is a do-things-on-my-machine tool that happens to be very good at code. Anything that can be described as files and commands is fair game, and a shocking amount of GTM work is exactly that.

So this is the fifteen-minute path from nothing to a first useful run, written for the operator I was, the one who almost skipped it. Do not make my mistake.

What Claude Code actually is

Claude Code is Claude running as an agent. You give it a goal, and it reads what it needs, decides the next action, takes it with a tool (read a file, run a command, call an API), checks the result, and keeps going until the job is done. It is the agentic loop from the concepts, turned into a tool you run.

The shift from the chat box is the whole thing. In the app you ask for text and paste it somewhere. In Claude Code you hand off a task and review the outcome. One is an assistant that talks; the other is one that does.

The shift to internalize is from chatbot to coworker. A chatbot answers, Claude Code acts: on your files, with your tools, in a loop, until the job is done. You are not asking it what you should do about the messy account list, you are handing it the list and watching it do the work, step by step, with you approving as it goes. That difference, talking versus doing, is the whole reason it exists.

Install and sign in

Getting running is quick, and none of it requires you to be technical.

Do not overthink the install. It is one npm command and a sign-in, the same as any command-line tool, and if npm is unfamiliar, that is the one genuinely technical minute in the whole process. After that you are typing plain English, not code. The terminal looks intimidating from the outside and turns out to be the least intimidating part once Claude is the one driving.

• Install with npm: run npm install -g @anthropic-ai/claude-code
• Start it: run claude in whatever folder you want it to work in
• Sign in: your Claude Pro or Max subscription, or an Anthropic API key
• It works in the directory you launch it from, so start in a folder that holds the files for your task

Make your first run a real task

Skip hello-world. Give it something from your actual week. A strong first run for a GTM operator: drop a company URL in and ask for a first-call account brief, or hand it a CSV and ask it to enrich and score the rows. You will learn more from one real task than from ten tutorials, because you will immediately see whether the output is worth your time.

Watch how it works as it goes. It narrates its plan, takes steps, and shows you the result, and that transparency is how you build the trust to hand it bigger jobs later. The first time it does in ninety seconds a thing that used to cost you an hour, the tool sells itself.

Resist the urge to make the first task a safe toy. The demo that converts you is a real one, because a toy proves nothing and a real result is the thing that rewires how you think about your week. Pick a job you were dreading, the list you have to enrich by hand, the follow-ups you owe, and let the first run be the one that saves you the hour you were about to spend.

Five real first tasks to try

If you are staring at the blank prompt unsure what to hand it, here are five that land for GTM operators on day one. None of them needs a line of code, and each produces something you would otherwise have done by hand.

• Enrich a CSV: read accounts.csv and add a one-line relevance angle per company from their website.
• Clean the pipeline: find duplicate accounts and flag records missing an owner or a next step.
• Draft at scale: write a tailored follow-up for each closed call in a folder of transcripts.
• Score inbound: rank a list of leads against your ICP with a one-line reason for each.
• Build a brief: turn a transcript and your positioning doc into a one-page pre-call brief.

It asks before it acts

By default Claude Code asks permission before it does anything that changes your files or runs a command, so a first run cannot quietly wreck something. You watch, you approve, you learn its habits. As you get comfortable, you widen what it can do without prompting, but that is a choice you make on purpose, not a default you inherit.

This is the same guardrail principle from the agent concepts, built into the tool: start cautious, loosen deliberately. It is much easier to grant more freedom to a tool you trust than to recover from one you trusted too soon.

This asking behavior is what makes it safe to start bold. You are never one careless instruction away from disaster, because the disaster pauses for your approval first. Read what it proposes and you will quickly learn the shape of its judgment, where it is reliable and where it needs a tighter instruction. The approval step is not just a safety net, it is how you calibrate trust.

Why a non-engineer should bother

Here is the reframe again, because it is the whole point for operators and I want it to stick: Claude Code is a do-things-on-my-machine tool. I am not a software engineer, I run demand gen, and it is one of the most-used tools in my stack. I use it to build and maintain the scripts and automations that move my GTM data around, and I could not tell you the last time I wrote one of them by hand.

Anything that can be expressed as files and commands is in scope, and once you look, an enormous amount of GTM work is exactly that: pulling a list, cleaning it, scoring it, drafting from it, updating a system with the result. You do not need to code. You need to describe the job and read the result.

The reason a non-engineer should care is that Claude Code is not really a coding tool, it is a do-things-on-my-machine tool that happened to be born in engineering. Your machine is full of GTM work that lives in files and systems, and until now the only way to act on it at scale was to be technical or to wait for someone who was. That gate is gone. The skill it asks for is describing what you want clearly, which is a skill you already have.

Where it fits with the app and the API

One thing that trips up first-timers is when to use Claude Code versus the Claude app they already know. The line is talking versus doing. The app is for thinking, drafting, and one-offs, a conversation. Claude Code is for multi-step work against your real files and tools, a job. If you are pasting the same context into the app every day to do the same task by hand, that is the signal you have outgrown the app for that job, and Claude Code is the next floor up.

Past Claude Code sits the API, for when software, not a person, makes the call, like scoring every inbound lead overnight without you initiating it. You do not need to think about that yet. The path most GTM operators walk is app first, then Claude Code the moment a task becomes repetitive and multi-step, then the API only when something should run entirely without you. Claude Code is the middle floor, and it is the one that changes the most about your actual week.

Where first-timers go wrong

The first trap is using it like the chat app, one question at a time, and never handing it a whole task. That works, but it wastes what makes Claude Code different, which is delegation. Give it the end goal, not a stream of tiny asks.

The second is starting too broad. 'Rebuild my entire lead-scoring system' on run one is how you get overwhelmed and quit. Start with one scoped task, get it working, and grow from there. The confidence to hand off big jobs comes from watching it nail small ones.

The most common early mistake is timidity, treating it like a search box and asking one small thing at a time, when its whole strength is taking on a multi-step job end to end. The second is the opposite, handing it something huge and vague with no way to check its work. The sweet spot is a real task with a clear definition of done: big enough to be worth it, scoped enough that you can tell when it succeeded.

Why this is the tool under everything

Claude Code is the engine beneath the automations on this site: the account briefs, the enrich-and-route flows, the weekly ad analysis. Learning it is learning the thing that runs all of them, which is why it is the first Claude Code piece to read and the one that pays back fastest.

Start with one task you repeat, get it working once, and you have the seed of a system. The rest of this pillar, context, plan mode, skills, hooks, is how you grow that seed into something safe, repeatable, and worth trusting.

So what is the one manual chore you did this week that is really just files and commands? Start there.

How to set it up

Frequently asked questions

Sources & further reading

• Claude Docs
• Anthropic news

Claude ships fast. This page was last reviewed Aug 22, 2026; verify time-sensitive details against the official docs above before relying on them.