Building with Claude

The biggest pain point in regular chat is the memory problem. Every conversation starts fresh. If you're working on something over weeks, you keep pasting the same background context into every chat. That gets old fast.

Projects solve this. A Project is a labelled space inside Claude where you can attach documents, instructions, and reference materials that every conversation in that project automatically has access to. You set up the context once. Every chat inside the project picks it up.

Practical advice: don't make Projects until you feel the pain of pasting context twice. The moment you do, make one. The setup takes ten minutes and pays back for weeks.

The free plan gives you up to five Projects, which is plenty to start. If you run out, that's a good sign Pro has become worth it.

Most of what Claude returns is text in a chat. But sometimes you want a thing: a usable document, a working webpage, an interactive tool, a chart. Artifacts are Claude's way of producing those.

When you ask Claude to make something that's better viewed as a thing than as paragraphs, it produces an Artifact alongside the conversation. You can read it, copy it, share it, or download it. The conversation continues; the Artifact updates as you give feedback.

The trick with Artifacts is asking for them on purpose. If you say 'build me a small thing that does X' or 'put this in a document I can copy,' Claude will reach for the Artifact format. If you say 'tell me about X,' you get paragraphs in the chat.

Artifacts are available on both free and Pro. The interactive ones are limited (no external data, no saving), so think of them as sketches. For real apps, there's one more step.

Claude Code is what Claude becomes when you give it hands. Instead of talking to a chat box, you talk to Claude inside your desktop app's Code tab, a terminal window, or your editor, on your own computer. Claude can then read your files, write to them, run commands, and build entire things while you watch.

What changes is who does the work. In chat, you ask Claude something and it tells you. In Claude Code, you ask Claude to do something and it does it. Modifies the file. Runs the program. Fixes the bug. Reports back.

This is how people build websites, automate workflows, write programs, and create tools from a single conversation. Without ever having learned to code in the traditional sense. The tool does the typing. You stay the director.

My take on order: chat → projects → artifacts → Claude Code is the path that works for most people, because each step makes the next feel reasonable. But if the desktop app's Code tab already clicked for you, you've effectively started here already, and that's fine. What makes people bounce is starting with no idea what they want to build.

Claude isn't the only option. Once you know what AI can do, there are other tools that do specific things better than Claude does, or do similar things differently enough to be worth knowing.

There's a natural progression for most people. Chat for a few weeks. Projects when context starts to feel heavy. Artifacts when you start wanting to make things. Claude Code (or Cursor) when you want to build. Other tools when you need something Claude can't do.

Nobody does this on a timeline. Some people stop at chat forever and get tremendous value. Some hit Claude Code in a week. There's no right pace. The only mistake is learning a tool you have no use for, then forgetting it, then feeling bad. Go further when you have something specific to build.