Your first week with Claude Code (for non-developers)

Day 1: install, and look before you touch anything

The instinct on day one is to ask Claude Code to build something straight away. Resist it. Install the extension, open a real project (or a small personal one if you don't have one at work yet), and spend the first session just looking: ask it to describe the project structure, or explain what one file does. You're checking that the install works and getting a feel for how it reads code before you ask it to write any.

Our guide to Claude Code in VS Code walks through the install itself, what the extension actually does, and how the agent loop works, so start there if you haven't set it up yet. If you'd rather watch it done than read it, Waseem recorded the whole setup as a ten-minute video, a blank machine to a first working page, which is the fastest way to see the rhythm before you try it yourself.

Day 2: have it explain a file before you change one

Before you ask Claude Code to change anything, ask it to explain something you don't understand. Every project has one of these: an inherited script, a config file nobody documented, a function with a name that tells you nothing. Paste it in, or point the agent at it directly, and ask what it does and why it might be written that way.

This is the single best exercise for building calibrated trust in the tool, and it costs nothing if the explanation is wrong. You'll start noticing the difference between a confident-sounding answer and a correct one, which is exactly the scepticism you'll need once it starts making changes on your behalf. It's the same instinct we push in Claude for data analysis: ask it to show its working, not just its conclusion.

Day 3: hand it one small, real, low-stakes task

Now do something for real, but keep it small: fix a minor bug, add a field to a form, tidy up a formatting inconsistency, add one small feature you've been putting off. The task matters less than the shape of it, small enough that a wrong answer is cheap to notice and undo, real enough that you actually care whether it works.

Describe the goal in plain English and let it show you a plan before it acts, then watch it work rather than switching tabs. If day 3 goes well, day 4 or 5 next week is a natural moment to think bigger: building an MVP is the usual next step once a small task has gone smoothly a couple of times, and if the small task turns into something the whole team leans on, building business apps with Claude covers the governance that keeps that from turning into shadow IT.

• Pick a task you could describe in one sentence.
• Let it show the plan before it edits anything.
• Watch the first few steps rather than walking away.
• Review the diff yourself before you approve anything that commits or runs a real command.

Day 4: connect your first tool

Everything so far has worked on files. Day 4 is where it starts working with your actual systems, through MCP (the Model Context Protocol), the open standard that connects Claude Code to a database, an API, an issue tracker, or a spreadsheet. Ask for something small that touches real data: a page that lists real records from a connected source, rather than an invented example table.

This is the point where the exercise stops feeling like a toy. Teams that get this far have gone on to build a working CRM, and we've written up a fuller worked example in building Orbit and Cortex with Claude if you want to see what the ceiling looks like past week one. Pick one tool, connect it deliberately, and be as careful about what you keep the agent away from as what you give it access to.

Day 5: the habit that actually matters

By the end of the week, the thing you've actually been practising isn't prompting, it's judgement: how to describe a task clearly enough to get it right first time, when to let the agent run unattended and when to sit and watch, and what to verify before you trust what it produced. That's the part that doesn't come from a cheat sheet, and it's the part every non-developer we've trained underestimates going in. Non-developers genuinely can build software with Claude, but the ones who do it well are the ones who built this judgement deliberately rather than by accident.

Set expectations honestly for yourself here: you won't be shipping production systems by Friday, and you shouldn't expect to. What a realistic first week gets you is a working setup, one small real thing delivered, one tool connected, and the beginnings of knowing when to trust it. That's a solid base to build on, and it's the exact skill split our hands-on AI training is built around: the agent does the mechanics, you do the judgement.