I’m using AI to try and learn PowerShell, and it didn’t take long to prove that human experts are still important

by

I know some people might not like it, but I do use AI. I use it to research topics I’m unfamiliar with – it reminds me of when Copilot first came out, back when it was called Bing Chat.

I’ve always preferred using AI chatbots to traditional web searches. For quick facts, a simple search works, but when I really want to understand a topic or do research, the chatbot’s conversational style is much better. Being able to ask follow-up questions and easily revisit ideas helps me learn, and it reminds me of having a discussion with a teacher – that’s how I absorb information best.

I’ve been meaning to learn PowerShell for a while now, just out of personal interest. I finally started, and I’m using AI tools to help me along the way. I’m not a big fan of just reading books to learn; I prefer a more hands-on approach – experimenting, making mistakes, and then finding the solutions myself.

It’s become clear very quickly that AI isn’t a replacement for skilled professionals, at least not yet. It’s not a simple solution, and without expert guidance, it can easily cause significant problems.

Just so you know, I’m not saying this is the best approach, or even that it *is* the right one. I’m simply sharing what I did and what happened to me as I went through the process.

Starting small, working through a project

When learning something new, it’s best to start with a small project. I decided to work on a PowerShell profile. Even though I was new to it, I understood the basic idea – it’s similar to a .bashrc file on Linux. However, I wasn’t sure what I could actually *put* in it, other than the code for my Starship prompt, which I use on all my systems.

I initially used Google Gemini to explore PowerShell profiles, hoping for easy-to-implement ideas. While I could have found the same information with a standard web search, Gemini’s conversational style is really helpful when I’m researching. I can always click on the links it provides for more details, but being able to ask specific questions and get clear, understandable answers quickly is a big advantage.

Eventually I had a small list of things to add:

  • PSReadLine history and intellisense
  • Invoking the Starship prompt
  • Adding Fastfetch at startup
  • Adding a cooldown so Fastfetch only appears every so often when launching PowerShell
  • A simple word count function for documents that would trigger using the command “wc” instead of the more verbose PowerShell command being needed every time.

Hardly groundbreaking stuff, but, starting small.

Starship simplifies things by providing the exact commands you need. I used AI for everything else – it helped me create those commands, and also explained how they work and how to use them.

Success and failure

This experience proved to me that AI can’t magically make someone an expert, because simply asking it to create what I needed didn’t work. And honestly, I’m relieved it didn’t.

I solved the problem by carefully working through it, identifying specific errors, figuring out *why* those errors occurred, and then trying to fix them. It took a while, but I learned some valuable things along the way. I realized I hadn’t clearly specified that I was using PowerShell 7, not the older Windows PowerShell, and that made a difference.

I also picked up how to write functions – what they’re for and how to build them. Like before, I learned by asking questions, getting explanations, experimenting, and then troubleshooting when things didn’t immediately work.

It’s not quite the same as a skilled coder debugging, but the approach has been similar: when something doesn’t work, keep at it until it does.

I finally achieved my goal! I picked up some useful PowerShell skills, like how to write functions, and everything is working as expected. It wasn’t a straight path, though – I definitely hit a few roadblocks along the way.

Human experts are still key

If I was already an expert in PowerShell, I’d be approaching this differently. But even with this simple project, I found I couldn’t just tell the AI what I needed and have it build it for me – it just didn’t produce the results I wanted.

Although my experience with AI is limited, it’s become clear that it can’t just automatically create working code. While that would be ideal, I’ve seen that without skilled human oversight, AI makes mistakes and creates potential problems.

AI tools shouldn’t aim to eliminate coding jobs, but rather help coders work faster. If I were using AI to generate code, I’d be able to identify and fix any problems. However, my experience showed it took a significant amount of time to debug the AI’s output and fully understand how it worked, even for a simple task. It wasn’t a quick solution, and required effort to ensure accuracy.

I’m continuing to explore how AI can help me learn and grow. However, it’s important that AI is developed as a tool to assist people, not to take our place – especially if we want things done well.

The completed article is available here if anyone would like to offer helpful feedback. I’m always looking to improve, so any constructive criticism is welcome!

Stay up-to-date with the latest from Windows Central by following us on Google News! You’ll get all our news, helpful insights, and great features right in your Google News feed.

Read More

2025-09-20 16:10