The AI Game-Buildability Radar

Score 1.80 — the lowest of any genre. Style-locked illustration, backgrounds, branching dialogue, worldbuilding, voice and music are all in AI's strong zone, and there's no real-time feel, netcode or level design to gate it. A single non-specialist can generate the art, voice and prose of a competent visual novel in a weekend.

Which is exactly why the remaining work is editorial, not technical: which 50 of 500 generated portraits to keep, and whether the writing has a soul. Build it for the taste, not the production.

Score 1.83, with no art pipeline at all: writing, worldbuilding, flavour text and a text shell. The bottleneck is narrative quality, which is editorial, not technical. With 48% of the catalogue released since 2023, the audience is live and arriving.

Score 1.94 — visual-novel-easy on art and writing, held just off the floor by one ingredient: puzzle design, a 4. Let AI build the scenes and the script, and spend your effort on solvable, fair, escalating puzzles.

Score 2.07. The art and scenes are exactly what image generation does best, and the systems are light. An evergreen casual buyer base with 45% of titles recent — easy to build and still bought.

Score 2.59. An environment-and-writing genre: AI handles the prose, the ambience and a good share of the props, while a human directs the environmental storytelling. With 51% of titles recent, it's one of the live ones.

Score 2.62. The card art — the visible bulk of the work — is instant, and the rules systems are well-trodden. The craft that remains is encounter balance, light here but the seed of what makes deckbuilders hard.

Score 2.77 — far easier than its "RPG" reputation. Strip it down and a JRPG is writing, worldbuilding, quests, character art and music over a simple turn-based combat skeleton: all four of those are AI's home turf. The turn-based structure is what spares it the combat-feel wall that action RPGs hit.

Score 2.23. Clickers carry almost no art burden; their one genuine craft is economy and progression tuning (a 3), and that's a spreadsheet-and-iterate problem AI is a strong assistant for. The catalogue keeps growing — 47% recent.

Score 2.38, and the single hottest momentum among easy genres: 59% of farming sims are recent. Pixel art, cozy writing, inventory/crafting and economy loops are all AI's home turf. The Stardew recipe is now largely automatable — except the long balance pass that makes the loop feel good.

Score 2.80. Core loops, economy and UI scaffold quickly with AI; the genuine work is QA and balancing the numbers until the simulation is satisfying. A large, steady audience (655M owner-estimate band).

Score 2.83. NPC AI, economy and procgen are solved patterns AI assembles well, but a colony sim lives or dies on emergent balance and the QA loop — heavily human. Half the genre is recent.

Score 3.12. The art and the tower/enemy systems are inexpensive to generate; the genre *is* encounter balance — wave pacing and the difficulty curve — which is among the AI-hardest content work. Build the content with AI, tune the waves yourself.

Score 3.12. Props and core systems are cheap, but a city builder is an economy-balancing and performance problem at scale — the simulation has to stay satisfying and stay fast as the city grows. Both are sustained human work.

Score 3.18, despite trivially generable card art, because the score is dragged up by encounter balance and economy — a great deckbuilder is a balancing achievement (Slay-the-Spire-class tuning), the AI-hardest narrative-adjacent work.

And it's the loudest demand signal on the board: 64% of deckbuilders are recent, the highest momentum of any genre. Highest demand × hardest-resisting core is the defensible niche — let AI generate every card, and spend your whole human budget on making the numbers sing.

Score 3.25. AI implements the procedural generation and the combat code readily, but a roguelike's quality is encounter balance and economy across endless permutations — the kind of tuning that only emerges from playtest data. Strong, recent demand.

Score 3.38. Environment, crafting and core loops are inexpensive, but survival is a long balance problem, frequently with multiplayer attached — and the moment netcode enters, the difficulty jumps. The largest owner-estimate band of any systems-craft genre.

Score 2.91, and the most deceptive on the board. Art, music and enemy AI all generate easily — so it *looks* AI-friendly — but the genre's entire identity is character-controller feel, level design and game-feel juice, the three things AI is worst at. A jump that doesn't feel good kills the game regardless of how the art looks.

Score 2.93. The board, the art and the loop are easy; what separates a good match-3 from a dead one is game-feel juice and a fair, escalating puzzle curve — both human. The lowest momentum among the assisted genres (28% recent).

Score 3.12. Everything the 2D platformer struggles with, raised: character feel, an interconnected world that must be designed to teach and gate, and readable combat. The art generates; the architecture doesn't.

Score 3.14. AI generates the environment, the ambience and the props beautifully, and the catalogue is huge (9,373 titles). But horror is pacing — the level design and moment-to-moment timing that turn assets into dread — and that's human. The biggest audience of any systems-craft genre.

Score 3.17, and the biggest genre we track (15,485 titles). The art and the engine are trivial; a puzzle game is *puzzle design* — solvable, fair, escalating — which AI drafts but cannot tune. The content looks done and plays hollow without the human pass.

Score 3.61. An action RPG layers 3D characters, skeletal animation, combat, environment art, level design, performance and game feel — AI shaves the edges of each, but none of the cores collapse. The contrast with the JRPG (2.77) is the whole thesis: real-time combat feel is the wall.

Score 3.73. Cars and tracks can be generated, but a racing game is handling feel and a rock-solid frame rate, with networking on top for multiplayer. The lowest momentum here (32% recent) — a mature, expertise-gated niche.

Score 3.87. The genre's identity is deep, interacting systems wrapped in hand-built level design with responsive feel and a punishing QA load. AI assembles the parts; the emergent whole is expert craft. Notably, 61% of titles are recent — a genre in revival.

Score 3.89. Stack 3D characters, animation, readable combat, encounter balance and deliberate level design, and you have a genre defined by exactly the ingredients AI scores 4 and 5 on. The assets help; the fight doesn't tune itself.

Score 3.89. An open world stacks 3D environment and character art, level and quest design, NPC AI, combat, performance and game feel — and then multiplies all of it by sheer scale. The largest owner-estimate band of any genre (3.3B), but the least tractable for a small team.

Score 3.82. A MOBA is authoritative netcode, sustained live-ops infrastructure, and competitive balance, all at once — three of the things AI helps with least. The smallest catalogue on the board (324 titles) and 28% recent: not where small teams break out.

Score 3.93. Single-player shooters are gated by combat feel; competitive ones add authoritative netcode — and that's where the modern genre lives. The biggest owner-estimate band on the board (3.4B), and among the least AI-automatable games that exist.

Score 4.09. Fighting games demand frame-perfect combat feel, rollback netcode and meticulous balance — the three hardest problems for AI, all load-bearing at once. And the data agrees it's not where small teams break out: only 17% of titles are recent, the most frozen niche on the board.

Score 4.17 — the highest on the board. Authoritative netcode for a hundred players, live-ops infrastructure, performance at scale and combat feel, all required simultaneously. This is the genre AI changes least: it can generate the assets, and almost none of the hard parts.