About | Apex Prompt Works

Apex Prompt Works is a practical workshop for experiments at the intersection of technology, artificial intelligence, and tabletop gaming.

The emphasis is on using these tools to do real work: running games, building workflows, testing ideas, solving problems, documenting failures, and seeing what survives contact with actual use.

Much of the work here grows out of tabletop roleplaying games and Commander. That includes AI-assisted campaign preparation, Foundry VTT and Obsidian workflows, voice and media tools, deckbuilding experiments, and the Grand Round Robin—a long-form test of whether a general-purpose conversational AI can actually pilot complete games of Commander against a human opponent trying to win.

The goal is not to prove that AI is magical, autonomous, or universally useful. Quite the opposite. Useful tools tend to become more interesting once their limitations are visible.

Apex Prompt Works documents what worked, what did not, what required human correction, and what was worth keeping.

Why This Exists

Most discussions of AI tools happen at one of two extremes: marketing demonstrations where everything works perfectly, or arguments about what the technology might someday do.

I am more interested in the middle.

Can a model manage a complicated game state for several hours? Can it make decisions that remain coherent after earlier choices constrain later ones? Can it help maintain a campaign without replacing the GM? Can an AI-assisted workflow save enough time to justify the extra machinery? Can a bad experiment teach us something useful?

Those are testable questions.

So the projects here are built around actual use rather than hypothetical capability. When possible, methods, prompts, corrections, results, and failures are preserved so that someone else can understand how a conclusion was reached.

Tabletop Gaming

Tabletop gaming is where many of these experiments become unusually demanding.

A roleplaying campaign may contain years of accumulated continuity, hundreds of NPCs, maps, notes, rules interactions, and player decisions. Commander asks a different set of questions: hidden information, sequencing, resource management, threat assessment, unfamiliar board states, and the need to live with decisions made several turns earlier.

These are useful environments for testing tools because they are messy in exactly the ways polished demonstrations usually are not.

The objective is not to automate the hobby out of existence. The useful question is where technology can remove clerical work, extend memory, improve preparation, or create new capabilities without displacing the human decisions that make the game worth playing.

The Grand Round Robin

The Grand Round Robin is the largest experiment currently documented here.

Twenty-eight Commander decks are being played through every unique pairing: 378 games in total. In each game, one deck is piloted by me and the other by ChatGPT. The games are actually played using Moxfield's playtest environment; future draws remain randomized outside the language model, and both pilots are trying to win.

The primary subject is not which deck is strongest. It is whether a conversational AI can maintain game state, make strategically coherent decisions across an entire match, adapt when its plans fail, and convert those decisions into real wins.

The results are recorded whether they make the AI look good or bad.

The Augmented Gamemaster

The Augmented Gamemaster is a broader project about using modern tools to support long-running tabletop campaigns.

It covers practical workflows involving AI, campaign notes, continuity, VTT preparation, images, maps, audio, voice, automation, and knowledge management.

The premise is simple: the GM remains the GM.

Technology is useful when it increases what one person can reasonably prepare, remember, create, or manage. It becomes considerably less useful when the tool itself becomes the project.

Who Is Behind This?

Apex Prompt Works is run by Gavvin, a longtime tabletop gamer, Commander player, game master, and IT professional.

The perspective here comes less from trying to predict the future of technology than from repeatedly asking a much less glamorous question:

Does this actually work?

If the answer is yes, I want to know how.

If the answer is no, I usually want to know why.

Contact