I cover AI for PCMag, and I spend most of my free time gaming, so it was only a matter of time before the two collided. I have zero interest in making (or playing) AI slop games. But using AI to make mods that fix what bugs me about the games I already play? I’m hooked, and I don’t even know how to code. Here’s what I’ve built, the simple workflow I use, and why you should give it a try.
Mods, Not Slop: What I’ve Made With AI
Have you ever played a game and wished it could do something very specific that it currently doesn’t? For example, I was playing Valheim and trying to find Haldor, a merchant who sells various goods for coins. Valheim relies heavily on procedural generation, so Haldor doesn’t have a set location but rather potential ones based on proximity to the center of the game world.
Now, you can plug your world into an online tool and look at a map to find these locations, but that felt too much like cheating to me. Still, I wanted a way to make sure I wasn’t wasting my time looking for him in the wrong part of the world. So, I went looking for mods, and I found one called Map Rings that adds helpful rings to the in-game map to show you how far sections of your world are from its center. Problem solved, right?
This map rings mod is almost perfect, minus a missing hotkey, but Claude helped me change it (Credit: Thunderstore/Tchernobill/PCMag)
Not quite. See, I didn’t want to ruin the beauty of my map with permanent rings overlaid on it. Basically, I wanted a hotkey to toggle them, but the mod doesn’t have one. So, I used Claude to add that functionality. With just a few prompts, Claude decompiled the mod to figure out how it worked and built an add-on that lets me toggle it. This is the kind of unique, specific functionality you can add to games with AI, and that’s just the beginning of what I’ve made.
For example, Valheim’s Forge of Potential mechanic, which is a system that lets you gamble on gear upgrades with hard-to-find idol items, is unpopular because of its incredibly punishing tendency to destroy your gear when you don’t win a gamble. So, I put on my game designer hat, thought up a better way to balance the mechanic, and implemented it with Claude.
Valheim’s animals were another pain point for me. You can tame and breed a variety of them, but manipulating them (especially transporting them from faraway places back home) is incredibly awkward. Accordingly, I created StayPut with Claude, which lets me issue a stay command at the press of a button, forcing tamed animals to sit wherever I ask them to and not fight random enemies or stumble off the edge of a boat or wall.
Even if you don’t know how to code, you can, right now, make pretty much whatever mod you want. However, I caution you against generating art and assets during your tenure as a modder. Modern AI models can do these things, make no mistake, but it’s much harder to match an existing art style than to end up with something that has the same generic look as the many AI-generated games cluttering up social media feeds. Practical concerns aside, using AI models in the first place raises ethical questions; feeding someone’s art into them and generating more just like it adds a significant extra layer of moral ambiguity.
Steal My No-Code Workflow for Making Mods
This isn’t the only way to make mods with AI, but it’s a reliable, straightforward method I’ve used to great effect. Most importantly, you can do it with absolutely zero programming experience and limited technical savvy.
First, you need to sign up for a premium subscription to a frontier model, such as ChatGPT or Claude. Right now, I lean toward Claude and its excellent Opus 5.5 model, but ChatGPT’s GPT-6 Astra model is also exceptionally powerful. You don’t have to spring for an expensive $100- or $200-per-month subscription, depending on the scope of the mods you plan to create, but you’re going to, at least, need access to the models and usage exclusive to premium plans.
Before you make anything, instill some guidelines into the memory of whatever AI model you choose. For example, when building Valheim mods, I demand that my mods be as lightweight as possible to preserve my frame rate, avoiding per-frame calculations and anything else that might get computationally expensive. I also require new mods to be compatible with all the mods I currently use to prevent potential conflicts. For peace of mind, I enforce a strict read-only policy for any AI that needs to access game files or saves, requiring it to copy them to its work directory before making changes. For Claude specifically, I require it to be concise, given its predilection for needless repetition and verbosity.
For every project I take on, I also establish a sourcing policy when the work requires routine internet access. When modding Valheim, that means preferring the official wiki over the outdated Fandom wiki.
Top models at the highest effort settings can churn through usage quickly, but they require much less babysitting (Credit: Anthropic/PCMag)
When it comes to model selection, an excellent rule of thumb is to use the highest intelligence setting of the most capable model your usage limits reasonably accommodate. Seasoned vibe coders can get great results out of relatively simple AI models with careful prompt calibration and monitoring, but if you don’t know much (or anything) about coding or just want a set-it-and-forget-it solution, more capable models are your best bet. For my Valheim mods, I use Opus 5.5 on its Max intelligence setting most of the time.
Once you start going back and forth with an AI model about a mod you want to create, I recommend the structure I used for my Valheim mods. After you know what kind of mod you want to create, have the AI make a comprehensive plan for creating it, including an independent review of the plan to catch issues and gaps it should fix before presenting it. When you task the AI with making the mod, require it to follow the plan, and then follow up with an independent review of the mod to surface any problems it ought to fix. My mods haven’t been terribly complicated, and I haven’t had any problems.
The specifics of your prompts depend on what you’re trying to do, but in most cases, you really don’t need a carefully calibrated prompt. For example, I was annoyed that Valheim’s recent 1.0 update made a certain building item, sap extractors, able to trigger raids, which are a real pain when I’m just trying to harvest some sap. So, I told Claude I wanted to undo this change. Claude made a plan for developing this mod, had it reviewed to catch any potential issues, then made the mod, which was also reviewed to squash bugs. Then, I dropped it into my mod manager, and voilà, no more raids. It’s as simple as that, assuming you put some care into setting up your workspace and using the right model, as I described above.
Lastly, and most importantly, test your mods. You’re not just the developer (or, rather, manager) of your mod; you’re its chief playtester. Especially if you plan on sharing your mods with anybody, even casually with friends, you need to carefully observe your mod in action whenever possible and note whenever something goes wrong or if there are any potential areas of improvement. For example, with my StayPut mod, I realized that the first version didn’t capture the direction a tamed animal faces, so I had Claude add it.
Modding the Modder: How I Upgraded Claude Itself
Why limit yourself to making mods when you can mod your modding tools, too? Claude, in particular, makes it easy to change the Claude Code application itself with prompting. However, the following is by no means limited to the Anthropic ecosystem.
The most important Claude mod I created is for managing context and usage. When a session reaches 80% full context or 95% of my five-hour usage limit, my mod automatically stops work and generates a handoff to kick off a new session. I also added a new user interface element at the bottom of the app to show me how close I am to these thresholds. This helps preserve performance, which can degrade when Claude operates with nearly full context, and also reduces usage, as large amounts of context require more tokens for Claude to process.
Anthropic lets you change much of how Claude Code works with mods (Credit: Anthropic)
I also created a mod I called Prompt Boost. This feeds my prompt, alongside conversational context and the official guide to prompting Opus, to Opus on high effort. Then, it spits back out a better version of the prompt in under 10 seconds. Effectively, I can type out what I want as a stream of consciousness, and then Prompt Boost cleans it up. Don’t get me wrong, this isn’t a replacement for shrewd prompt engineering, but it does allow me to be less careful with how I phrase instructions.
Creating mods for Claude works the same as creating mods for games. I go back and forth with Claude on whether what I want is possible and how it should work first. Then, I make Claude create a plan that it thoroughly reviews before presenting it. If all looks good, I have Claude execute the plan, review the results, and give me my mod. As I use my mod, I pay attention to how well it works, so I can return to Claude for any improvements.
Recommended by Our Editors
For example, I noticed the first version of my handoff mod didn’t always trigger when it should, and it didn’t stop any background tasks running in a session. I explained my experience to Claude, which then fixed these issues. Then I realized that the mod was automatically stopping sessions in progress rather than allowing them to properly wrap up, so I once again noted this functionality to Claude so it could fix the issue, which it did without a problem.
What AI Modding Can (and Can’t) Do
I’m the first to admit that I have no idea what my code for the aforementioned mods looks like, so I’m a vibe coder in the truest sense of the term. However, considerable testing, rigorous review policies, tight construction constraints, and the limited scope of my mods give me a sense of security. Most importantly, I haven’t run into any issues, so I’m happy.
The reality is that a model like Opus 5.5, especially in its top-effort settings, is simply overkill for most mods. In my case, it automatically decompiles my game and any other relevant mods to closely examine the underlying nuts and bolts during mod creation, and it stops to ask me for design guidance whenever a question comes up. Even if new restrictions make Opus 5.5 a less capable partner, other models, such as GPT-6 Astra, are also powerful, and new models are constantly emerging.
I’ve made a bunch of Valheim mods with Claude, and they work well (Credit: r2modman/PCMag)
Sure, if, for example, I dared Claude to dream up a whole new expansion for a game I’d otherwise have to wait months or years for the developers to create, I would likely run into months of problems myself and end up with a subpar product. AI isn’t magic (at least not yet), so don’t expect a paragraph and some tokens to revolutionize gaming. However, most mods aren’t ambitious overhauls or huge expansions, and yours don’t need to be, either.
Of course, some mods go a lot further. Take Valheim City Builder, for example, which the creator made with help from Claude. This turns Valheim, the third-person survival sandbox game, into a top-down city builder with tons of new mechanics but without any new assets, keeping the game’s original look and feel intact.
AI ethics aside, if you’re interested in treating AI like a hot glue gun to the popsicle sticks of video game code, you’ve got a capable, fun toy to play with that empowers you to make stuff. And these tools are only going to improve. One day soon, you might actually be able to create something like the aforementioned city builder mod with a single prompt.
Even with what’s possible now, though, I’m a different gamer than I was a year ago. Take, for example, my experience with Crimson Desert. The game’s combat just felt too awkward to me to really stick with. But rather than throw my hands up and admit defeat, I whipped up a traditional Dark Souls-style camera lock-on system, which made the gameplay a lot better. Now, if something bothers me about a game, especially if it looks like a deal-breaker, my instinct is to change it.
What Will You Fix First?
I’m a firm believer that more people getting their hands dirty with game design is a good thing, and it has never been easier to start. Think of the one thing that bugs you most in a game you love, like a missing hotkey, a punishing mechanic, or an awkward camera, and describe it to an AI model. You’ll pay for a subscription, but that costs far less than the time it would take to learn to code. Worst case, you cancel after a month. Best case, you stop accepting a game’s flaws as permanent. That’s what happened to me, and I don’t plan to go back.
About Our Expert
Ruben Circelli
Writer, Software
Experience
I’ve been writing about consumer technology and video games for over a decade at a variety of publications, including Destructoid, GamesRadar+, Lifewire, PCGamesN, Trusted Reviews, and What Hi-Fi?, among many others. At PCMag, I review AI and productivity software—everything from chatbots to to-do list apps. In my free time, I’m likely cooking something, playing a game, or tinkering with my computer.
Read Full Bio