Executive Summary

This is the executive summary. It covers what Claude Code, Anthropic's command-line coding agent, costs, what actually matters once you're using it, and how to decide if it fits your team. I've built this guide on my own workflow, not on Anthropic's own documentation, organized around a single frame: Claude Code as a tool for getting real work done, not only for writing software. That framing, including a claim about running a profitable business on it, is my own experience, not an independently verified fact.

No prior technical background is assumed. What follows walks from installing Claude Code and picking an editor, through the single idea that shapes almost everything else, a file called CLAUDE.md, and on to controlling how much freedom you give the agent, planning complex builds before you start them, and running more than one agent at once.

Here's the cost. At least a Claude Pro subscription, listed at $17 a month billed annually, as of this writing. I think that's small next to the gain, though that's my own opinion, not a measured return. Subscription pricing is also one of the most volatile numbers in any technology guide, so verify the current figure yourself before you commit a team's budget to it, rather than trusting a number fixed at one point in time.

Some material sits outside this first pass. It moves past single web pages into backends, larger applications, and tools built for a team rather than one person. Only a small share of Claude Code's users understand those more advanced features, in my own impression, a personal read, not a market survey, and I'm treating it here as exactly that.

What should you take from this if you read nothing else? Three things. First, CLAUDE.md is not a nice-to-have. It loads before your first message and shapes everything that follows, the way a ship's initial heading determines where a long voyage ends up. Second, the core discipline behind this whole approach is simple: give Claude a task, let it act, then verify the result yourself. Skip that last step and you lose most of the value of the whole approach. Third, permission modes and Plan Mode exist because autonomy is a dial, not a switch. More freedom for the agent means faster work and less oversight, and the right setting depends on how much a mistake would actually cost you in that moment.

This is my own approach. I've generalized it where the underlying mechanism holds and flagged it as personal practice where it doesn't. Every claim in the documents that follow traces back to a specific point in the source transcript, or is disclosed plainly as something added for clarity. See the references appendix for the full trace. I've kept hedges as hedges, and stated numbers as stated, vague ones included.

One question to ask before deciding whether this approach fits your own work: does your team already have a habit of defining a task clearly before delegating it? If yes, you're most of the way there already. If not, the CLAUDE.md discipline described in the pages that follow is where that habit gets built. Start with the overview for exactly what's covered and what isn't.