I love card games, and I can usually get decent at them with some time and effort. I’ve climbed my way up to the top rank—Legend—in Hearthstone, and competed in some local (and one regional) tournaments in Weiss Schwarz. However, Magic: The Gathering has long been my white whale—a game that always intrigued me but just seemed too tough to get into, due in part to its 30-year history and tens of thousands of cards. But that all changed after I played a few hundred games, watched a lot of videos…and vibe coded an app (called Replay Coach) with GPT-5.6 to help me understand why I lost when I did.
Here’s the full story of how AI helped me finally break into the Magic: The Gathering scene and reach the top rank in its video game adaptation.
Meet Replay Coach: An AI Reviewer, Not a Cheat Engine
AI is a touchy subject in the video games industry, and even a wrong comment can create massive controversy. So, putting my cards on the table (heh), here’s what my app does and why it’s not remotely unsportsmanlike.
Replay Coach parses the local log data of Magic: The Gathering Arena (Magic’s most popular video game adaptation) to create a summary of what happened in games after they are over. It then packages that data, along with some supplementary information I’ll explain later, into a coaching packet of sorts that you can send to an AI chatbot.
The packet contains instructions for an AI chatbot to explain what went wrong, how you could have prevented those mistakes, and what to look out for next time. So, for example, one analysis I received highlighted my major mistake of letting Lord Skitter resolve when I had a counter in hand, Get Out. Other smaller issues it identified included a sub-optimal discard of Winternight Stories over Sunderflock when I didn’t have another creature in play and the waste of Spider-Sense on a low-priority triggered ability.
Replay Coach also tracks removal, which refers to cards that can destroy creatures or counter other cards, especially board wipes that destroy all the creatures currently in play. Put simply, the app lists instances of removal in my games and allows me to filter them by turn or type. This was incredibly helpful when I was trying to learn which cards to play against the opponents I faced. For example, Replay Coach taught me to watch out for Avengers Disassembled in red decks, Day of Judgment in white decks, and Deadly Cover-Up in black decks on turns three, four, and five, respectively.
Why not record games and watch them back to figure out what happened on my own? Well, I just didn’t know enough as someone new to Magic to do that effectively. Hiring an actual coach would have definitely helped, but it wouldn’t have done much for me when I ran into unfamiliar cards and new situations that those coaching sessions didn’t cover. Endless coaching sessions wouldn’t be very cost-effective either. Most importantly, I wouldn’t have been able to write this article.
Vibe Coding the Tool: How GPT-5.6 Built the Engine
To build Replay Coach, I started with GPT-5.5, along with a dash of Opus 4.8 and Fable 5, for the first iteration of its interface. However, I developed the bulk of it with GPT-5.6 Sol, primarily at the Extra High intelligence setting. I have a ChatGPT Pro 5x subscription ($100 per month), but OpenAI gave out a number of free usage resets during development. I used OpenAI’s Codex app for all the coding, which later became the ChatGPT app.
The concept of Replay Coach is simple, but the hardest part of its development was making it generate useful game recaps. The log data Magic: The Gathering Arena generates isn’t human-readable or even something an AI can understand by itself, so turning it into a summary isn’t easy. For example, my current log file is a 5.81MB block of text filled with numerical IDs and technical jargon, which Replay Coach needed to translate into something AI could analyze effectively and efficiently.
Replay Coach’s parser turns log data into readable summaries of games, giving you the full picture of what happened during them (Credit: Ruben Circelli)
However, mere translation wouldn’t cut it, either, because if a summary is too long or not formatted well enough, it can overload an AI’s context window or send it down a rabbit hole of endlessly chasing down details. Accordingly, my summaries needed to balance accuracy, brevity, and formatting. To start, I had GPT analyze as much of my log data as possible to create a parser that could accommodate edge cases and unique interactions, and I insisted on a format that featured legally available actions, mana availability, a turn history, and more.
As development progressed, I often tasked GPT-5.6 with refining the parser by comparing different versions. I needed to ensure that the parser didn’t drop important information as I slimmed down game summaries. In response, GPT-5.6 spun up agents to analyze game summaries produced by different versions of the parses and provide coaching based on them. Then, it created fresh agents to blindly judge each coaching result. Based on the results, GPT-5.6 improved the parser to produce summaries that were both accurate and not overly long. In general, GPT-5.6 made this process a breeze and required minimal oversight.
However, summaries are only one component of the larger coaching packet Replay Coach produces. As mentioned, I also built Replay Coach to track removal, and I include a summary of that data within the coaching packet. Furthermore, I intended for Replay Coach to integrate with Untapped, a popular third-party Magic: The Gathering add-on that lets you watch game replays, track various stats, and more. As extra context, the packet includes the information Untapped collects on games, too, and Replay Coach provides an easy-to-access link to Untapped replays. This lets me quickly visualize what the coaching feedback references if I don’t remember it.
One of the last things I did when developing Replay Coach was test out GPT-5.6’s ability to design user interfaces by giving it a fresh, new look to replace the one I originally made with Claude (slide two). To my surprise, GPT-5.6 created a slick, modern interface (slide one) that looks like a website I might stumble across in 2026, albeit one that relies a little too heavily on large, space-inefficient tiles. When I tried to make an interface with GPT-5.5, the result felt incredibly awkward to navigate and dated, for comparison.
The Secret Sauce: Training a Custom GPT on Rules and Strategy
Ultimately, Replay Coach just compiles information for an AI chatbot to receive and then analyze, and while you can use any AI chatbot with Replay Coach, I got the best results with ChatGPT and GPT-5.6. I tried using Claude, but the resultant coaching just wasn’t as useful. This was, in part, due to one of ChatGPT’s lesser-known features: Custom GPTs.
These are the settings for MTGA Coach, my Custom GPT I use for providing the actual coaching of my games (Credit: OpenAI/PCMag)
Custom GPTs are versions of ChatGPT you can customize with your own instructions and knowledge sources. I created the MTGA (Magic: The Gathering Arena) Coach Custom GPT. As you might expect, I gave it instructions to act as a Magic coach, analyzing submitted coaching packets and providing helpful advice to a new player. However, I also uploaded Magic: The Gathering’s comprehensive rule set and Reid Duke’s Level One course on Magic strategy to its knowledge sources to ground coaching in not just the mechanics of the game but also a well-respected approach to playing it. Finally, I set my Custom GPT to rely on GPT-5.6 Thinking and selected the Heavy thinking effort setting to get the best possible feedback.
Recommended by Our Editors
Overframe Wasn’t Cutting It, So I Vibe Coded a Better Warframe App With Claude
I Just Vibe Coded a Global Mass Surveillance Site in 2 Hours With OpenAI’s Codex. It Was Terrifyingly Easy
I Used Vibe Coding to Build My Own Zillow in Just a Few Hours
How I Turn Every Loss Into a 5-Minute AI Lesson
Using Replay Coach is simple: I click the shortcut I made for it, complete with an icon inspired by the Magic: The Gathering Arena branding that GPT-5.6 created, and it opens in my web browser, Google Chrome. Then, I get to playing Magic with the Untapped overlay open. Whenever I lose a game (or if I win one I want coaching on), I switch to Replay Coach, import my logs, select the game I want coaching on, and download the corresponding coaching packet. I then send the packet to my Custom GPT. On average, I get my lesson about five minutes later.
It’s not a perfect app or overall system by any means, but it helped me many times, highlighting criminally obvious mistakes I shouldn’t have made in the first place, identifying mechanics I simply wasn’t aware of, and pointing out cards I should be familiar with. Since a Custom GPT is still an AI chatbot (and one with access to Magic: The Gathering rules and strategy), I could follow up coaching with any questions I had. And I always made sure to double-check whatever ChatGPT said made sense by watching the corresponding moments in my Untapped replay.
The Payoff: 218 Games Later, I Hit Top 1,500 Mythic
Using Replay Coach in conjunction with watching countless videos of more skilled players piloting similar versions of whichever deck I was using at the time helped me climb my way up from Bronze in Magic: The Gathering Arena’s ranked best-of-one ladder to the highest rank, Mythic, where I peaked at the 1,365 spot midway through the July season before losing my motivation to keep climbing higher.
The whole journey took more than 300 games. However, I started at the end of June, so just by the time I climbed my way to Platinum, Magic: The Gathering Arena’s monthly rank reset system reset me back to Silver. Those 73 games aside, and the next 25 I spent idling in Silver before I changed from the more complicated mono-blue elementals deck I ran up to that point to the mono-white auras deck I spent the rest of my climb with, I went from Silver to 1,365 in Mythic across 218 games with a 60% win rate.
To be clear, I’m far from an expert Magic player, especially considering the easy-to-play and very powerful mono-white auras deck I primarily used. However, I can finally say that I know how to play Magic: The Gathering. Most importantly, I can now take enormous pleasure in the knowledge that if I dedicate enough of my precious little time on Earth to playing a decades-old fantasy card game, I can play it relatively well. Worth it, if you ask me!
Try Replay Coach (Or Vibe Code Your Own App)
If you’re interested in using Replay Coach or playing around with its code, you can download it here, and you can access my MTGA Coach Custom GPT here. As mentioned, I recommend using ChatGPT with GPT-5.6 at the Heavy thinking level, but you can use whatever AI chatbot or ChatGPT model you prefer to analyze coaching packets. As a disclaimer, I’m not a programmer, so I can’t vouch for the quality of its underlying code. However, if you’re looking for a fun way to take advantage of the power of an AI model, I definitely recommend giving Replay Coach a shot or making an app of your own to help you be the best gamer possible.
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.
Areas of Expertise
Latest By Ruben Circelli
- The Best Online Collaboration Software for 2026
- Stop Paying for the Newest AI Models. You Really Don’t Need Them for Most Tasks
- Study Smarter: 10 Ways to Use AI for School That Won’t Get You Expelled
- Zapier Review: Boost Your Business With Easy Automation
- IFTTT Review: Automate Your Life, Simply and Seamlessly
- More from Ruben Circelli
