How games are made, what people use to make them, and the sockets an AI can actually hold.
This is the library behind the studio: the eight-stage process, engines, the 2026 AI stack, MCP servers that give a model hands, and the places a first game can live.
The process
01
Spark — Name the desire
Every game begins as a feeling you want someone else to have. Capture the high concept, the logline, and the constraint that will keep the project honest.
You leave with: A working title, a one-sentence pitch, and a reason this game should exist.
Tracy Fullerton calls this conceptualize: a playable idea small enough to prototype. Jesse Schell would ask: what is the essential experience?
02
Pillars — Decide what never moves
Design pillars are the three truths the team can use to cut ideas. Audience, platform, tone, and non-goals stop a solo project from becoming every game at once.
You leave with: Three pillars, a player fantasy, comparables, and an explicit not-list.
Naughty Dog, Insomniac, and most mid-size studios hang pillars on the wall so every feature can be asked: does this serve the fantasy?
03
Loop — Make the verb addictive
The core loop is the minute-to-minute activity the player will do a thousand times. Verbs first, then failure, then the feeling (MDA aesthetics) those verbs produce.
You leave with: Player verbs, a written core loop, failure, session length, and target aesthetics.
MDA (Hunicke, LeBlanc, Zubek, 2004): designers build Mechanics → Dynamics → Aesthetics; players feel it in reverse. Name the aesthetic first.
04
World — Give the loop a place to live
Setting, characters, and art direction are not decoration — they tell the player why the verbs matter. Keep lore load-bearing. Cut anything the loop does not touch.
You leave with: Setting, premise, cast, story beats, and art/audio direction notes.
Environmental storytelling (Bioshock, Dark Souls) beats cutscene dumps. If a fact never changes a decision, it does not belong in a first slice.
05
Blueprint — Specify what you will actually build
A modern GDD is an index, not a novel: systems, levels, economy, controls, UX, risks, and a cut list. Rosa Carbo-Mascarell's shape — Summary, Core, Features, Content — still holds.
You leave with: A buildable feature list, a scope cut list, and a control scheme.
Stone Librande (GDC 2010) argued one-page designs people will actually hang up beat hundred-page docs nobody reads.
06
Build — Make it playable today
A vertical slice is worth more than a perfect document. Forge a prototype from the design — geometric, juicy, honest about what the loop feels like in the hands.
You leave with: A running game you can play, share, and iterate.
Iterate: conceptualize → prototype → playtest → evaluate (Fullerton). Ship the ugliest thing that tests the fantasy, then keep the parts that are fun.
07
Polish — Make the same game feel expensive
Juice is the gap between a prototype and a toy people want to touch: hit-stop, squash, particles, audio ticks, readable failure. Accessibility and difficulty are polish too.
You leave with: A feel checklist, audio plan, accessibility notes, and a difficulty curve.
Steve Swink's Game Feel and Martin Jonasson's juice talks: the same mechanic with better response reads as a different game.
08
Launch — Put it in someone else's hands
A game that lives on your machine is a sketch. Write the listing, the trailer script, the tags, and a share plan. Export a file a stranger can open.
You leave with: Store copy, tags, a press blurb, and a shareable build.
Itch, Steam Next Fest, and web portals (Poki, CrazyGames) reward a clear one-sentence promise and a GIF of the loop — not a lore dump.
From desire to constraint
A game idea is not a plot. It is a feeling you can defend under a constraint: solo, 2D, a week, a browser. The constraint is a gift. It tells you what the game is allowed to be.
Pillars exist to say no
If a feature does not serve a pillar, it is a different game pretending to live in this one. Write three sentences a tired future-you can use as a knife.
The loop is the product
Players will do the core action a thousand times. If that action is not interesting on the third repeat, no story, art, or soundtrack will save it. Prototype the verb first.
Vertical slice, not horizontal sprawl
One complete moment — start, play, fail, retry, a hint of the fantasy — beats five disconnected systems. Production is copying that moment until it is a game.
Watch hands, not opinions
Sit behind someone. Do not explain. The first place they hesitate is the design. Ask what they thought the game wanted, not whether they liked it.
MCP is a studio in a socket
A language model without tools writes design fiction. Wired to Blender, Unity, Godot, Figma, and GitHub it becomes a junior who can move the actual files. Pair a knowledge server (how to design) with an engine server (how to act).
Tools, engines, MCP
In 2026 the stack split into four jobs: prompt-to-playable (Rosebud, Chatforce), in-engine agents (Summer, Unity AI, Godot MCP), asset generators (Meshy, Scenario), and code copilots (Cursor). MCP is how the last three grow hands.
Engines & environments
Unity
The default 2D/3D production engine. C#, massive asset store, ships everywhere.
Use when You need 3D, mobile, or a large marketplace of plugins.
Free personal; seats for teams · Player builds for every major store
Engines & environments
Unreal Engine
Photoreal 3D, Nanite, Blueprints plus C++. The AAA look, with a royalty after revenue.
Use when Cinematic 3D, large worlds, or you already think in Blueprints.
Free; 5% royalty after $1M · Console, PC, and high-end 3D
Engines & environments
Godot 4
Open-source, scene-tree mental model, GDScript that reads like Python. No royalties.
Use when You want to own the stack and ship 2D or modest 3D without a license worry.
Free, MIT · Desktop, web, mobile exports
Engines & environments
GameMaker
The 2D specialist behind a generation of indie hits. Visual plus GML.
Use when Pixel 2D you actually want on Steam and consoles.
Free to start; paid for export · Steam, console, desktop
Engines & environments
Construct 3
Event-sheet 2D in the browser. No traditional code required.
Use when You think in events, not classes, and want HTML5 yesterday.
Subscription · HTML5, wrappers
Engines & environments
GDevelop
Open-source event engine with a growing AI assistant and gd.games hosting.
Use when You want free, visual, and a one-click web page for the build.
Free; credits for AI · Web, desktop, mobile
Engines & environments
Phaser
The HTML5 2D framework. If the game should live on a URL, start here.
Use when You already write JavaScript and want arcade, platformers, or puzzles in-browser.
Free · Any static host
Engines & environments
Defold
Lua, tiny runtime, blessed by a lot of mobile and web arcade teams.
Use when Performance on phones and a Lua-sized brain.
Free · Mobile and HTML5
Engines & environments
Bevy
Rust ECS engine. Fast, code-first, still growing its editor story.
Use when You like Rust and want the game to be a real crate, not a hidden binary.
Free, MIT/Apache · Desktop; web via wasm
Engines & environments
Roblox Studio
Luau, a built-in audience, and 2026's prompt-native Build flow expanding beyond NZ.
Use when Distribution is the hard part and you are willing to live inside their platform.
Free; platform split · Roblox, with web play rolling out
Engines & environments
UEFN / Fortnite Creative
Unreal for Fortnite islands. Verse language, a gigantic installed base.
Use when You want Fortnite's players more than your own launcher.
Free; island revenue share · Fortnite
Engines & environments
PlayCanvas
WebGL editor in the browser. Collaborate like a Google Doc for 3D.
Use when 3D that must load on a link with no install.
Free tier; paid seats · Web
AI in the pipeline
Rosebud AI
Prompt-to-playable browser games with a remix culture. Fastest way to see a loop.
Use when You want a playable concept this afternoon, not an engine project.
Free tier; credit plans · Playable web game, JS export
AI in the pipeline
Summer Engine
AI-native engine compatible with Godot 4. The agent creates scenes, writes GDScript, runs the game, reads errors, and fixes itself.
Use when You want to own the project and still let an agent drive the live editor.
Free; Pro around $20/mo · A real Godot project
AI in the pipeline
Chatforce
Multi-agent studio (director, coder, artist, sound) that aims for a complete 2D browser game in minutes.
Use when You want art, audio, and a loop from one conversation.
Free + $20/mo · Polished 2D web game
AI in the pipeline
Meshy
Text/image to game-ready 3D: remesh, PBR, auto-rig. The broadest 3D asset API.
Use when You need a prop or character in the engine before lunch.
Free credits; paid plans · FBX/GLB with textures
AI in the pipeline
Tripo
Fast image-to-3D with cleaner topology and part segmentation.
Use when Hero assets that have to survive a sculpt pass.
Free; Pro ~$20/mo · Quad-friendly meshes
AI in the pipeline
Scenario
Train a style on your own art, then generate consistent 2D sprites and concept sheets.
Use when The game has a look and you cannot afford drift between characters.
From ~$15–20/mo · Consistent 2D sheets
AI in the pipeline
Inworld AI
Runtime characters with memory, voice, and unscripted dialogue.
Use when NPCs should talk back without a dialogue tree the size of a novel.
Usage; indie tiers · In-engine character brain
AI in the pipeline
Convai
Conversational NPCs with vision and gesture, Unity and Unreal SDKs.
Use when You want talking characters inside an existing engine project.
Free limited; Indie ~$29/mo · SDK characters
AI in the pipeline
Ludo.ai
Ideation, market research, and asset exploration — a pre-production toolkit, not a builder.
Use when You are still asking whether the idea has a player.