AI is making extremely niche software viable

AI is filling the Pathfinder character builder gap

Oct 28, 2026 · 4 min read

I’ve always enjoyed perusing projects that make extremely complicated game systems legible. My gold standard is the Hitchhiker’s Guide to the Cataclysm, a companion tool to the open-source roguelike Cataclysm: Dark Days Ahead that automatically extracts information from the game’s GitHub repository and updates every twelve hours to keep up with the game’s daily updates.

The Pathfinder games developed by Owlcat Games, most recently _Wrath of the Righteous_, adapt the Pathfinder 1e tabletop system and would also benefit greatly from a companion tool. This is because the games feature a dizzying array of options for character builds—a dozen playable races, more than 150 classes and subclasses, and over 1,000 spells, feats, and abilities, each with their own prerequisites and restrictions (not to mention countless items to choose from). Players are relatively free to multiclass, which only adds to the complexity. For this reason, planning the journey from level 1 to level 20 is half the fun for many players. There’s a reason Pathfinder 1e is commonly called “Mathfinder”.

The problem is, there’s no convenient way to actually plan builds outside of the game. Your careful spreadsheet planning might be undone by forgetting that a particular ability becomes available at level 5 instead of level 4, or that a certain feat has a prerequisite that is only available at level 10. There’s no way to automatically test potential builds against the actual game rules to determine if they are valid.

I’ve looked for tools like this in the past, but they didn’t exist. There’s a GitHub project that started in 2023, Pathfinder-Wrath-of-the-Righteous-Build-Tool, but it’s more of a virtual character sheet, allowing you to record your choices without validating them against the actual game rules.

The reason building a tool to plan and validate builds for Pathfinder is difficult is not the huge number of feats, spells, abilities, items, and the like. These are relatively easy to extract from the game’s files, since much of this information is stored as JSON in a compressed archive. The hard part is how all of these features interact, i.e., implementing the game logic to validate that a build is actually legal in the game.

Pathfinder: Wrath of the Righteous was released in September 2021 and received its final patch in May 2025, nearly a year and a half ago. So I figured there wasn’t much hope of such a tool suddenly being developed.

I looked again today. I found three build planning tools, all released in the last few months:

The three tools all have different functionality. AutoLevelUp is an in-game mod and uses the game’s own level-up screen to create and validate builds, so there is no need to reimplement the game’s rules. Wrath Build Planner has a browser companion that offers build choices based on the game’s available options, but the in-game mod is needed to actually apply and validate them. So again, no full validation external to the game. WotR Database is the most ambitious project, a completely browser-based build planner that attempts to reimplement the game’s features and rules, including prerequisites and restrictions.

What do these three projects have in common? They all appear to have been heavily assisted by AI agents. AutoLevelUp explicitly states that “the C# implementation was written by AI agents under my direction.” The Wrath Build Planner GitHub repo has Claude as a co-signer on all of its commits. WotR Database is less explicit, but its articles describe decompiling the game rules, including those on attack rolls and damage and feat prerequisites. Not to mention the very AI-generated nature of the writing on these pages, and especially the site’s roadmap.

The point here is not so much the sudden emergence of these particular niche software tools, but rather what they suggest about how AI coding agents are changing the economics of software development. It was always possible to painstakingly recreate a complex game’s rules in an external software tool, but the enormous effort required for a relatively small audience made projects like this non-viable. AI-assisted coding (and decompilation) has lowered that barrier immensely, making projects viable that previously weren’t worth the effort.

There must be countless similarly niche software projects that nobody bothered to build for precisely this reason.