I've worked in IT for about five years and currently manage a department. I'm not a hands-on programmer, but I work closely with developers, understand the main concepts and terminology, and regularly take part in discussions about engineering practices and ethics. I also completed some beginner programming courses and small projects at university more than five years ago, so I'm not starting from zero. I still have a virtual machine, VS Code, Git, and some old school projects available.
I'd like to start a few personal projects with AI assistance, probably using Claude. Possible ideas include a weather report, a personal weight-loss tracker, and eventually a statistics comparison tool for items in a game. I don't want to simply ask an AI to build an entire application and end up learning poor habits or producing code I don't understand. I want to learn the foundations first, analyze the generated code, ask useful questions, and minimize low-quality AI-generated code.
Can anyone recommend a YouTube creator or other guide that explains the fundamentals of setting up AI-assisted software projects before getting deep into prompting? I'm especially interested in project structure, choosing a language and tools, version control, testing, documentation, and good practices for working with Claude.
There may also be a work-related angle. I might eventually use these skills for minor UI fixes in our SaaS product, although those tasks are usually postponed in favor of new features and higher-priority work. I understand that contributing to production software could introduce risks, so I'd want to approach that carefully.
2 Answers
For Claude-based projects, a few practical habits make a big difference:
- Break the work into small tasks rather than requesting a complete project in one prompt.
- If the AI misunderstands something, correct the earlier context instead of repeatedly adding contradictory follow-up messages.
- Use a separate conversation for each substantial feature or task so the context stays manageable.
- Learn how project instruction files such as CLAUDE.md work. You can keep general project conventions in one file and add more specific instructions in subdirectories. These can document coding style, file organization, testing commands, and important assumptions.
- Start complex tasks in a planning or read-only mode. Review the proposed approach and ask the AI to identify assumptions before allowing it to change files.
- Ask it to record important technical decisions and project context in documentation, but keep those files organized and concise.
- Have it explain changes, run tests, and check its own work rather than assuming the first solution is correct.
Token-saving tools and techniques can help later, but they’re less important than maintaining a clean project structure and keeping the AI’s instructions accurate. A small project with clear documentation and tests will teach you much more than trying to generate a large application immediately.
You probably don’t need a long AI-specific tutorial before starting. Since you already have VS Code and some programming background, choose a small project and use an AI extension that can inspect the code as you work. The more important first step is selecting an appropriate language and framework for the project, either by researching it yourself or asking an experienced developer for guidance.
Then work in small, clearly defined pieces instead of asking for the entire application in one prompt. Ask the AI to explain its choices, review the output, and help you write tests. Try to worry less about avoiding AI-generated code entirely and more about making sure you understand and validate everything it produces. AI will make mistakes, so generated code still needs review by someone with real development knowledge, especially before it is used in a production system.
That makes sense. I already have VS Code installed from my university projects, though I may start with a clean setup. In our department, people are responsible for reviewing their AI-generated code, and we only use it for tasks where the team has agreed it is appropriate. There’s no unrestricted use in production.
Thanks, this is exactly the kind of practical checklist I was looking for. I understand the general idea of splitting projects into smaller pieces, but the advice about CLAUDE.md files, planning mode, and keeping technical context organized is especially useful.