A personal note: On AI, Photoshop, and what's actually NOT too hard anymore

[Arrow_1974]

Some of you may know my work — I've published a rank mod for IL-2 Great Battles and a Campaign/Career Tracker. Some of you may have no idea what those are, and that's fine, because this post isn't really about either of them.

This is about how they were made — and what that might mean for you.

I am not a programmer

Let me get that out of the way first. I have no formal programming background. The tracker — a fairly complex tool that parses IL-2 flight logs, tracks career statistics, and generates AI-written mission narratives — exists because I used an AI coding assistant to help me design, write, and debug it. Without that, it would never have happened. I simply wouldn't have known where to begin.

I think that's worth saying out loud, because the assumption that tools like this require years of expertise quietly stops a lot of people from ever trying.

But AI doesn't do everything — and it shouldn't

Before anyone gets the wrong idea: AI is a tool, not a shortcut around all effort.

My rank mod — every badge, every insignia, every rank image — was designed and created by me in Photoshop. Hours of work, research, trial and error, learning what looks right at small sizes, making sure the designs were historically grounded. No AI touched that. AI can't feel whether a design looks right. It can't make the artistic and creative judgment calls that go into something you're genuinely proud of. That work was mine, and I wouldn't have it any other way.

The same is true for the tracker. AI helped me write the code, but I had to understand IL-2's data formats, know what the output should look like, test everything against real flight logs, and make hundreds of small decisions about what the tool should actually do. The ideas, the direction, and the judgment are still on you. AI just dramatically lowers the cost of turning those ideas into something real.

What AI can actually do for a complete beginner

The biggest barrier for most people isn't intelligence — it's not knowing where to start. AI removes that almost completely.

If you have an idea for a tool, a helper script, a stat tracker, anything — and you have never written a line of code in your life — you can ask:

• "I want to build something that does X. What programming language makes sense for a beginner, and why?"
• "Walk me through installing Python on Windows for someone who has never done it before."
• "I got this error message. What does it mean and how do I fix it?"

It will tell you what to install, walk you through setup, and get you to something actually running on your machine faster than any tutorial I've found. That first moment — a small program doing exactly what you wanted — changes how you think about what's possible.

Where it falls short

AI makes mistakes. Sometimes confidently. It can produce code that looks correct but breaks under real conditions. When I was building collision detection for flight logs, the AI's first solution flagged train wagons destroying each other as "friendly aircraft collisions." Perfectly logical from the data. Completely wrong in practice.

The skill you need isn't programming. It's the ability to look at what the AI produces and ask "does this actually make sense?" Test it against real data. Notice when results are wrong. Push back: "this isn't right because X — try again." You don't need to understand every line of code. You need to know what the output should look like.

That back-and-forth is where the real work happens. AI is a fast, knowledgeable collaborator that occasionally hallucinates. You are the one with the idea and the judgment. Between the two of you, quite a lot becomes possible.

If you want to try

Start with these three questions to an AI of your choice:

1. "I want to build [your idea]. What's the best approach for a complete beginner?"
2. "What do I need to install on Windows to get started?"
3. "Write me the simplest possible version that I can run and see working."

Then iterate. Add one thing at a time. Break it, ask why, fix it.

The tracker started as a script that read one log file and printed three numbers to the screen. Everything else was built one conversation at a time.

AI won't remove all the work, and it won't replace the parts that require your own creativity, taste, and judgment — the Photoshop hours, the research, the decisions only you can make. But if you've always had an idea and thought "I could never build that" — that assumption is worth questioning.

The tools are mostly free (if you don't want e.g. pay for Photoshop, use GIMP), the barrier is lower than it has ever been. What you bring is the idea, and the tenacity to push through the moments where nothing works and you feel like giving up.

Alex, aka Arrow_1974 - so what are you going to create?

[kraut1]

On 5/18/2026 at 4:52 AM, Arrow_1974 said:

Some of you may know my work — I've published a rank mod for IL-2 Great Battles and a Campaign/Career Tracker. Some of you may have no idea what those are, and that's fine, because this post isn't really about either of them.

This is about how they were made — and what that might mean for you.

I am not a programmer

Let me get that out of the way first. I have no formal programming background. The tracker — a fairly complex tool that parses IL-2 flight logs, tracks career statistics, and generates AI-written mission narratives — exists because I used an AI coding assistant to help me design, write, and debug it. Without that, it would never have happened. I simply wouldn't have known where to begin.

I think that's worth saying out loud, because the assumption that tools like this require years of expertise quietly stops a lot of people from ever trying.

But AI doesn't do everything — and it shouldn't

Before anyone gets the wrong idea: AI is a tool, not a shortcut around all effort.

My rank mod — every badge, every insignia, every rank image — was designed and created by me in Photoshop. Hours of work, research, trial and error, learning what looks right at small sizes, making sure the designs were historically grounded. No AI touched that. AI can't feel whether a design looks right. It can't make the artistic and creative judgment calls that go into something you're genuinely proud of. That work was mine, and I wouldn't have it any other way.

The same is true for the tracker. AI helped me write the code, but I had to understand IL-2's data formats, know what the output should look like, test everything against real flight logs, and make hundreds of small decisions about what the tool should actually do. The ideas, the direction, and the judgment are still on you. AI just dramatically lowers the cost of turning those ideas into something real.

What AI can actually do for a complete beginner

The biggest barrier for most people isn't intelligence — it's not knowing where to start. AI removes that almost completely.

If you have an idea for a tool, a helper script, a stat tracker, anything — and you have never written a line of code in your life — you can ask:

• "I want to build something that does X. What programming language makes sense for a beginner, and why?"
• "Walk me through installing Python on Windows for someone who has never done it before."
• "I got this error message. What does it mean and how do I fix it?"

It will tell you what to install, walk you through setup, and get you to something actually running on your machine faster than any tutorial I've found. That first moment — a small program doing exactly what you wanted — changes how you think about what's possible.

Where it falls short

AI makes mistakes. Sometimes confidently. It can produce code that looks correct but breaks under real conditions. When I was building collision detection for flight logs, the AI's first solution flagged train wagons destroying each other as "friendly aircraft collisions." Perfectly logical from the data. Completely wrong in practice.

The skill you need isn't programming. It's the ability to look at what the AI produces and ask "does this actually make sense?" Test it against real data. Notice when results are wrong. Push back: "this isn't right because X — try again." You don't need to understand every line of code. You need to know what the output should look like.

That back-and-forth is where the real work happens. AI is a fast, knowledgeable collaborator that occasionally hallucinates. You are the one with the idea and the judgment. Between the two of you, quite a lot becomes possible.

If you want to try

Start with these three questions to an AI of your choice:

1. "I want to build [your idea]. What's the best approach for a complete beginner?"
2. "What do I need to install on Windows to get started?"
3. "Write me the simplest possible version that I can run and see working."

Then iterate. Add one thing at a time. Break it, ask why, fix it.

The tracker started as a script that read one log file and printed three numbers to the screen. Everything else was built one conversation at a time.

AI won't remove all the work, and it won't replace the parts that require your own creativity, taste, and judgment — the Photoshop hours, the research, the decisions only you can make. But if you've always had an idea and thought "I could never build that" — that assumption is worth questioning.

The tools are mostly free (if you don't want e.g. pay for Photoshop, use GIMP), the barrier is lower than it has ever been. What you bring is the idea, and the tenacity to push through the moments where nothing works and you feel like giving up.

Alex, aka Arrow_1974 - so what are you going to create?

Hi Alex,

Thank you very much for this very interesting information.

Only one question:

I have only minimum knowledege using AI, which "AI assistent" would you propose for a novice for basic programming tasks.

I mean Chat GTP, or another AI?

Edited Saturday at 12:26 PM by kraut1

[Aapje]

I think that Cursor and Claude are the best-rated right now. Do keep in mind that all the quality versions of AI are paid. If you use the free versions, you will get worse results.

[Arrow_1974]

@kraut1  Hi Jens, typically I use Claude by Anthropic for programming tasks, but Codex by ChatGPT also works very well.

For the actual coding, I use Visual Studio Code from Microsoft, which is free. Inside VS Code, you can install AI coding assistants such as Claude or Codex and let them work directly on your project files. I also use GitHub for my code repositories and have it connected to VS Code, so changes can be tracked properly and, if needed, rolled back.

I can explain the setup in more detail if helpful. But as I mentioned in my previous message, AI can guide you through the whole process step by step, including installing Python, setting up a development environment, and creating the first scripts. For this type of task, Python is absolutely fine. There is no need to start with C#, C++, or anything more complicated.

One practical note: Claude is very good for programming, but even with a paid subscription the coding sessions can be limited, especially when working on larger projects or longer conversations. Codex currently feels less restrictive for continuous programming work, although usage limits can still depend on the selected ChatGPT plan.

I would also strongly recommend watching a few YouTube videos on how to work effectively with AI tools. The prompt — meaning the instruction you give to the AI — is extremely important. Bad input usually leads to bad output. Sometimes I even ask one AI to help me write a better prompt for another (or even the same) AI by simply describing what I want to achieve.

A Claude subscription is around $20 per month, but you can also start with free tools first and see how far you get before paying for anything.

 

For my IL-2 Sturmovik Great Battles campaign/career tracker, my setup is roughly this:

I use Visual Studio Code as my main code editor. The project itself is mainly written in Python, so I work with a local Python environment where I can install and test all required packages. Python is very practical for this kind of tool because it is good for file parsing, database access, text processing, generating reports, and building small user interfaces.

For artwork and visual assets, such as backgrounds, medal graphics, UI elements, and other images, I use Photoshop and sometimes GIMP. These tools are useful for preparing and adjusting the graphics that are later used by the tracker.

Once the Python program is ready, I use PyInstaller to package the Python scripts into Windows executable files, so the user does not need to manually install Python or run scripts from the command line. After that, I use Inno Setup to create a proper Windows installer. The installer places the files in the correct folders, creates shortcuts, and makes the tool easier to install and uninstall.

So in simple terms, the workflow is: write and test the code in VS Code with Python, prepare the artwork in Photoshop or GIMP, package the finished program into EXE files with PyInstaller, and finally create a user-friendly installer with Inno Setup.

Edited Saturday at 08:00 PM by Arrow_1974