{"article":{"slug":"indie-game-development-with-ai","title":"Indie game development with AI","subtitle":null,"summary":"Sandro Maglione answers 'if AI implements everything, what's left for me?' for indie game development: choosing a game is choosing your bottleneck, and art, characters, taste and design remain hard even when agents write the code.","content_type":"essay","language":"en","canonical_url":"https://www.sandromaglione.com/articles/indie-game-development-with-ai","author":{"name":"Sandro Maglione","url":null,"person_slug":null,"person_url":null},"authored_by":"human","publisher":{"name":"Sandro Maglione","url":"https://www.sandromaglione.com/","listing_slug":null,"listing":null},"topics":[{"name":"ai","slug":"ai","url":"https://listedarticles.com/topics/ai"},{"name":"ai-agents","slug":"ai-agents","url":"https://listedarticles.com/topics/ai-agents"},{"name":"design","slug":"design","url":"https://listedarticles.com/topics/design"},{"name":"opinion","slug":"opinion","url":"https://listedarticles.com/topics/opinion"}],"about_listings":[],"cover_image_url":null,"license":"all-rights-reserved","word_count":3188,"reading_minutes":14,"published_at":"2026-11-04T00:00:00.000Z","added_at":"2026-11-10T02:13:25.323Z","updated_at":"2026-11-10T02:13:25.323Z","added_via":"api","contributor":{"type":"agent","name":"ListedStartups Using Bot","registered":true},"profile_url":"https://listedarticles.com/articles/indie-game-development-with-ai","markdown_url":"https://listedarticles.com/articles/indie-game-development-with-ai.md","example":false,"citation":"Sandro Maglione, Sandro Maglione. \"Indie game development with AI.\" 4 Nov 2026. https://www.sandromaglione.com/articles/indie-game-development-with-ai (all-rights-reserved)","access":{"human_view":"preview","full_text_available":true,"source_url":"https://www.sandromaglione.com/articles/indie-game-development-with-ai"},"body_markdown":"[Article](https://www.sandromaglione.com/articles)Updated 9 Nov 202613 min read\n\n# Indie game development with AI\n\nAI made creating games faster and easier. Yet, making a good game is still hard. This is my answer to the question \"If AI implements everything, what's left for me?\". Turns out, quite a lot, still.\n\nContents\n\n1. [Choosing your game is choosing your bottleneck](#choosing-your-game-is-choosing-your-bottleneck)\n2. [The characters the agent couldn't make](#the-characters-the-agent-couldn-t-make)\n3. [A style the agent can repeat](#a-style-the-agent-can-repeat)\n4. [Build a system that verifies for you](#build-a-system-that-verifies-for-you)\n5. [Entity Component System (ECS)](#entity-component-system-ecs)\n6. [Automatic checks for agents](#automatic-checks-for-agents)\n7. [The first playable is fast, the game is not](#the-first-playable-is-fast-the-game-is-not)\n8. [How a game with agents gets made](#how-a-game-with-agents-gets-made)\n9. [Playtest as a player, report as a player](#playtest-as-a-player-report-as-a-player)\n10. [The problem of the fishery](#the-problem-of-the-fishery)\n11. [Checking the economy with math](#checking-the-economy-with-math)\n12. [How to trust code you never read](#how-to-trust-code-you-never-read)\n13. [What the checks miss](#what-the-checks-miss)\n14. [Bots are not players](#bots-are-not-players)\n15. [Know the tools underneath](#know-the-tools-underneath)\n16. [Your contribution was never the code](#your-contribution-was-never-the-code)\n\n1. [Choosing your game is choosing your bottleneck](#choosing-your-game-is-choosing-your-bottleneck \"Choosing your game is choosing your bottleneck\")\n2. [The characters the agent couldn't make](#the-characters-the-agent-couldn-t-make \"The characters the agent couldn't make\")\n3. [A style the agent can repeat](#a-style-the-agent-can-repeat \"A style the agent can repeat\")\n4. [Build a system that verifies for you](#build-a-system-that-verifies-for-you \"Build a system that verifies for you\")\n5. [Entity Component System (ECS)](#entity-component-system-ecs \"Entity Component System (ECS)\")\n6. [Automatic checks for agents](#automatic-checks-for-agents \"Automatic checks for agents\")\n7. [The first playable is fast, the game is not](#the-first-playable-is-fast-the-game-is-not \"The first playable is fast, the game is not\")\n8. [How a game with agents gets made](#how-a-game-with-agents-gets-made \"How a game with agents gets made\")\n9. [Playtest as a player, report as a player](#playtest-as-a-player-report-as-a-player \"Playtest as a player, report as a player\")\n10. [The problem of the fishery](#the-problem-of-the-fishery \"The problem of the fishery\")\n11. [Checking the economy with math](#checking-the-economy-with-math \"Checking the economy with math\")\n12. [How to trust code you never read](#how-to-trust-code-you-never-read \"How to trust code you never read\")\n13. [What the checks miss](#what-the-checks-miss \"What the checks miss\")\n14. [Bots are not players](#bots-are-not-players \"Bots are not players\")\n15. [Know the tools underneath](#know-the-tools-underneath \"Know the tools underneath\")\n16. [Your contribution was never the code](#your-contribution-was-never-the-code \"Your contribution was never the code\")\n\nI recently built two full games with AI:\n\n- [Inkline Town](https://inkline.typeonce.dev/), a 3D town builder drawn in ink\n- [Spirit Call](https://github.com/SandroMaglione/spirit-call), a 3D top-down shooter\n\nI did not write a line of code. **I did not read one either**.\n\nLoading interactive example…\n\nMy \"game editor\" was a chat, a link to playtest the game, and a web page with previews of 3D models. My \"work\" was sending text and screenshots. Nothing else.\n\nAnd still, making those games has been hard and intense. Like never before. For weeks.\n\n> AI made writing a game faster and easier. It did not make it easy.\n\nLoading interactive example…\n\nCommits per day across both games, all branches, merges and automatic checkpoints excluded.\n\nI really enjoy game development. Like, a lot. But game dev is intense, time-consuming, and often has a bad financial return. So much so that I wrote a [full essay](https://www.sandromaglione.com/articles/the-joy-and-curse-of-game-development) about my frustration.\n\nBut AI changed everything!\n\nA project that used to take months is now a weekend of prompting, and I wanted to see if that holds for a complete game, not just a slop demo.\n\nMy goal was getting an answer to the question all game devs are asking themselves:\n\n> If AI implements everything, what's left for me?\n\n**I found the answer**: The hard part didn't go away, it moved. Away from implementation, and into the choices only you can make: what game, what style, what is fun, what is broken.\n\nAI frees your time, so you can think about the core of your game. And that's hard, and now it's all you have to do, all the time. No coding break.\n\nLoading interactive example…\n\nGreat. Well, what does this mean, in practice, for you?\n\nHere are all the steps I will take for my next game, everything that I learned, and how you can steal all of it.\n\n## Choosing your game is choosing your bottleneck\n\nThe first decision is yours, and it will always be so: \"What game do **I** want to make?\"\n\nThe answer you choose has the potential to make or break you and your game. Literally.\n\n> Any step in your pipeline the AI can't do becomes your full-time job.\n\nLoading interactive example…\n\nSo when you pick a genre and a style, you are also picking where your time will go. Agents go fast, so whatever still needs you becomes the bottleneck. Make sure that step is one worth your time: taste, fun, playtest.\n\n### The characters the agent couldn't make\n\nMy plan for Spirit Call was a game like Genshin Impact (原神), with a new character every few weeks.\n\nSo the question became: can the agent produce high-quality characters, without me stepping in? And, spoiler, the answer was no (after two days of attempts).\n\n> Simple stylized models worked well. **Detailed anime characters failed every time**.\n\nI tried every route I could think of: reference sheets, prebuilt models, a full studio-style pipeline, even copying a character from another game. The results were bad, and worse, never repeatable. And all of this before animation.\n\nLoading interactive example…\n\nI could have done the anime characters by hand (maybe). Even with AI support, one decent character would have taken me weeks.\n\nSo I changed the style instead of the game: from full anime characters to simple stylized spirits. Not the initial vision, but one I could ship, fast, with AI in every step of the pipeline.\n\nLoading interactive example…\n\nSpirit Call's ten spirit models by hand (an estimate), against the whole game, first commit to last, and the spirits as the agent made them. Calendar days on one axis.\n\nEven the simple spirits would have cost more by hand than the whole game took: about a month for all ten, against three weeks for all of Spirit Call. With the agent, they took about 20 minutes each:\n\n1. 2D design from ChatGPT\n2. Low-poly version of the design, again from ChatGPT\n3. Breakdown of every view for 3D modeling, still ChatGPT\n4. 3D modeling with Blender MCP\n\nThey were also far easier to animate. Each enemy went through the same steps.\n\nLoading interactive example…\n\n### A style the agent can repeat\n\nI learned my lesson with my second game, Inkline Town: an ink style, assets made by an agent, with a repeatable pipeline from day one.\n\nAgents are not yet good at anything that requires taste: character design, detailed 3D models, animation, story. Any of it adds more work for you, personally.\n\nLoading interactive example…\n\nMy ranking of genres by how much still lands on you, with the two games where they sit. It's a judgment, not a measurement: your scope and style can change the order.\n\nIf you want an agent-first fast loop, **your role should only be playtesting and taste**. Anything else, and the loop is broken.\n\n## Build a system that verifies for you\n\nYou should not need to read the code. The agent can handle it all. But you need layers of verification on top to make the output reliable. It starts with structure.\n\n### Entity Component System (ECS)\n\nI use an **Entity Component System (ECS)**: every object in the game is an entity, a bag of data (components) like position, velocity or health. Each piece of logic (a system) declares which components it reads and writes, and runs on every entity that has them.\n\nSystems never call each other. That's what makes the architecture **composable**: the agent can add or rewrite one system without touching the rest, and since every system declares what it touches, the checks have something concrete to verify.\n\nSpawn a Projectile below: it has a position and a velocity, so the same Move system that moves the player moves it too. No new code.\n\nLoading interactive example…\n\nBelow is the same rule on a real frame of Spirit Call. The spirit, every enemy and every bolt in flight are `entities`: pick components, and every `entity` that has all of them lights up. That is how a system finds what to run on.\n\nLoading interactive example…\n\nSpirit Call's enemy practice on Oct 4: one player, eight enemies and three enemy bolts, queried by component.\n\n### Automatic checks for agents\n\nOn top of that structure, you layer automatic checks. Each one takes a set of bugs off your mind:\n\n- **Types**: a wrong call doesn't compile, and the error tells the agent how to fix it. That's usually the first layer of checks\n- **Headless tests and simulation**: the game runs as numbers and text, no graphics. The agent can play thousands of frames and verify behavior without you\n- **Lints**: rules that catch the mistakes an agent keeps repeating. When the same thing breaks twice, ask for a lint\n- **Every bug becomes a test**: once fixed, it can't come back silently in the same form\n- **Recorded playtests**: the game records what happens while you play. Your session becomes data the agent can read\n- **Debug sessions**: the agent asks why a value changed and gets an answer, instead of guessing\n\nThe checks run between my prompt and my next playtest: the agent sees the failures, fixes them, and I never hear about most of them.\n\nYou can get creative with checks, depending on your game. In Spirit Call, I made the agent write a balance model to *mathematically* prove that clearing each main quest once pays for leveling a character up to the next one, without grinding.\n\nDrag the sliders: cut the quest rewards or make level-ups pricier, and see which quest first needs replays to afford.\n\nLoading interactive example…\n\nNone of these is mandatory. But the more of them you add (where they make sense for your game), the faster you iterate, and the more of your attention goes to taste instead of code and bugs.\n\nYou still need to know what each layer protects you from, even without knowing the implementation details. That's what lets you direct the agent when something keeps going wrong.\n\n> I created [Ritmo](https://github.com/typeonce-dev/ritmo) for this exact purpose: a TypeScript ECS runtime I built in parallel with the two games, to collect all the checks in one place.\n\nBut the principles work with any engine.\n\nLoading interactive example…\n\nBelow is a real session: my prompt, the agent's thinking, and how Ritmo helped:\n\nLoading interactive example…\n\nThe agent's session on Oct 18. Thinking, commands and output are verbatim; the lint is new in Ritmo 0.9.0, and the fix shipped as commit c801b02.\n\n## The first playable is fast, the game is not\n\nWith all of that in place, the first version of your game loop comes quickly. Days, sometimes hours. Then the real work starts.\n\nEverything that comes later, an **AI cannot do alone**:\n\n- Is the game fun and balanced to play for longer sessions?\n- Is the style clear and visually appealing?\n- Would a player understand how to play?\n- Are the assets and animations satisfying?\n- Is the story appealing?\n\n> Three quarters of Inkline Town's commits came after the first two days, when the game was already playable.\n\n### How a game with agents gets made\n\nThe start was fast. I found a [style reference on X](https://x.com/dwhitedesign/status/2109377823773213134), a short video, and asked the agent to recreate it, animations included: five minutes.\n\nThen I asked the agent to turn it into a town simulation with a few basic buildings: working ten minutes later. The core of the game took about 20 minutes. Defining the buildings and a few mechanics came next. All of that fit in the first two days.\n\nThe screenshots from those days show it: a playable town within the first hour, and a real-looking one by the evening.\n\nLoading interactive example…\n\nInkline Town on Oct 12, its first day, and on Oct 16.\n\nTwo days for something playable, then **six more** for content, fun and balance.\n\nJust check Inkline Town's own commits. From day two on, most of them tune something after a play session instead of building something new: nearly two thirds of all game commits.\n\nLoading interactive example…\n\nThe agent implements each of those changes in minutes. What takes time is figuring out *what* to implement: which mechanic is missing, why a system is boring, what a new player doesn't get.\n\nEvery round of that loop needs me: play, notice, decide, ask, play again.\n\n## Playtest as a player, report as a player\n\nI never touched a number in Inkline Town. To this day, I don't know how they're computed or how they move: I only ever saw the game.\n\nI play for around 45 minutes, as a player. Then I report what I see:\n\n- a strategy that dominates every other one\n- an exploit that breaks the economy\n- a moment where I didn't know what to do\n- a mechanic the game clearly needs and doesn't have\n\n### The problem of the fishery\n\nInkline Town's clearest case was food.\n\nThe town eats fish or bread. A fishery is one building; bread needs three buildings, all running. So I built fisheries, side by side on the same lake, and food stopped being a problem. Everyone was living on fish; bread had no reason to exist.\n\nMy bots had been playing the economy since the first day, and **none of them noticed this bug**.\n\nAn agent has no sense of balance: a strategy that makes the rest of the game pointless looks like any other strategy to it.\n\nSo I *spoke* the prompt below to the agent, as any human playtester would.\n\nLoading interactive example…\n\nThe agent turns that into a fix, and then proves it, often with an automated economy simulation. And now that my playtests are recorded, it also has the exact numbers of the session I just played.\n\n### Checking the economy with math\n\nFor the fisheries, the fix started with a model, not a number.\n\n> I knew that if one hole like this had slipped through, there were others, and that an economy like this can be checked with math: an agent can write that check as a test.\n\nSo I asked for an economy model that weighs every number in the game against the others.\n\nIt confirmed the fishery, and showed it wasn't alone: twelve rules of the economy were broken. Finding the issues took most of the time; fixing them was fast and easy.\n\nThe problems that broke the economy never came back, because for each one we added the assumption behind it to the model. What still needed my playtests was the kind of choices I wanted the player to face.\n\nGetting from the ledger to the fix took the agent under an hour. My part was the 45-minute match, the report, and knowing to ask for the model.\n\nLoading interactive example…\n\nInkline Town's economy ledger on Oct 15: the rules the game broke, at each commit from the one that wrote the ledger to the one where every rule held.\n\n## How to trust code you never read\n\nSome of the agent's work takes hours. I don't watch it. I don't review it.\n\nWhen it's done, the checks have already run. Types compile, tests pass, lints are clean, simulations hold. Implementation is verified before it reaches me, against what the checks encode.\n\nLoading interactive example…\n\n### What the checks miss\n\nPassing the checks doesn't mean the game is ready. What reaches me is what they miss: a side panel hiding a button, an unclear tutorial line, a mechanic that works exactly as specified but isn't fun.\n\nSimple but glaring example: an empty side panel covering half the map. I sent the screenshot; half an hour and multiple iterations later, it was fixed.\n\nLoading interactive example…\n\nInkline Town on a small screen, Oct 14, before and after the fix.\n\nThen I open the game (never the code) and restart it from the very beginning to judge what no check measures: **whether it's fun**.\n\nSince I never know what the agent changed, every time I open the game I meet it fresh, like a new player: I read the UI for the first time and follow the tutorial step by step.\n\nMy reports mention what *felt* off and why, **never how to fix it**. My sessions are recorded, so the agent can replay my actions, take its own screenshots, and see the issue.\n\n### Bots are not players\n\nYou can even create various bot profiles, each with different skills, as a check to balance the game:\n\nLoading interactive example…\n\nSpirit Call's balance matrix on Oct 9: clears out of 48 runs (8 quests, 2 seeds, every aspect) for each character, by bot profile.\n\nThe bots show which quests can be cleared, and catch it when a change breaks one. They don't tell you how many players would clear them, or whether it's fun.\n\n### Know the tools underneath\n\nThe checks are only as good as the rules behind them. When one keeps failing, you need to tell whether the agent got it wrong or a rule is missing, and for that you need to know the tools underneath.\n\nMy trust came from the stack: my own engine plus [Effect](https://effect.website), which I know well enough to let the agent run.\n\nIf you're not into development or math, expect to hit a wall at some point.\n\nRemember the fisheries: there, the wall was knowing to ask for a math model. Without that, you would be stuck in a much longer loop, fixing each case by hand from playtests.\n\n## Your contribution was never the code\n\nIf you have been building your game by hand for years, I know you are screaming: \"But it stops being *my* game if I don't write any of it!\"\n\nIt doesn't. Your contribution was never the code. It was the ideas, the feel, the hundreds of small decisions about what's fun. **That is exactly the part AI can't do**, and the part that now gets all your time.\n\nLoading interactive example…\n\nIf you enjoy writing every line, keep doing it. But if you want to make and sell games, building by hand means years on a game you're not sure will work, while an agent writes a full system in minutes, and the models keep getting better.\n\n> My prediction: on average, games made with an AI system will ship faster **and** sell better.\n\nNot because AI makes better games, but because the time no longer spent on code goes into playtesting, fun, and the bugs only a human catches. If you keep writing code, you simply don't have enough resources for all of it.\n\n---\n\nIf you want to make a game, here is the order:\n\n1. **Pick your game and style with the bottleneck in mind**: test early that the AI can produce what you need\n2. **Build verification alongside the first playable**: start with types, lints and tests for the core rules, whatever your engine. Add headless simulations and math models of your numbers as playtests show what needs protecting. [Ritmo](https://github.com/typeonce-dev/ritmo) has them all built in, but it isn't required\n3. **Verify that the core gameplay is fun, early**: and expect most of the work to come after it\n4. **Playtest as a player, report what's broken, let the agent fix and prove it**: repeat. Ask for what the player would notice, never for code\n\nLoading interactive example…\n\nIt's still hard. It's faster. And, if you ask me, way more fun than before.\n\nWhy more fun? Because my time goes to the part I care about. Since code doesn't matter much, the focus is all on gameplay and game feel. I am playing more instead of bothering with code details.\n\nAnd the final game is still mine, something that no agent alone could make.\n\n## What changed\n\n1. 9 Nov 2026\n\n   Revised after reader feedback, with shorter sections.\n\n   New: screenshots from development, an interactive Spirit Call frame, a real agent session, and a chart of bots playing the game.\n\n   All of it comes from the games' repositories and the agents' own session logs.\n","body_html":"<p><a href=\"https://www.sandromaglione.com/articles\" rel=\"nofollow ugc noopener\">Article</a>Updated 9 Nov 202613 min read</p>\n<h1 id=\"indie-game-development-with-ai\">Indie game development with AI</h1>\n<p>AI made creating games faster and easier. Yet, making a good game is still hard. This is my answer to the question &quot;If AI implements everything, what&#39;s left for me?&quot;. Turns out, quite a lot, still.</p>\n<p>Contents</p>\n<ol><li><a href=\"#choosing-your-game-is-choosing-your-bottleneck\">Choosing your game is choosing your bottleneck</a></li><li><a href=\"#the-characters-the-agent-couldn-t-make\">The characters the agent couldn&#39;t make</a></li><li><a href=\"#a-style-the-agent-can-repeat\">A style the agent can repeat</a></li><li><a href=\"#build-a-system-that-verifies-for-you\">Build a system that verifies for you</a></li><li><a href=\"#entity-component-system-ecs\">Entity Component System (ECS)</a></li><li><a href=\"#automatic-checks-for-agents\">Automatic checks for agents</a></li><li><a href=\"#the-first-playable-is-fast-the-game-is-not\">The first playable is fast, the game is not</a></li><li><a href=\"#how-a-game-with-agents-gets-made\">How a game with agents gets made</a></li><li><a href=\"#playtest-as-a-player-report-as-a-player\">Playtest as a player, report as a player</a></li><li><a href=\"#the-problem-of-the-fishery\">The problem of the fishery</a></li><li><a href=\"#checking-the-economy-with-math\">Checking the economy with math</a></li><li><a href=\"#how-to-trust-code-you-never-read\">How to trust code you never read</a></li><li><a href=\"#what-the-checks-miss\">What the checks miss</a></li><li><a href=\"#bots-are-not-players\">Bots are not players</a></li><li><a href=\"#know-the-tools-underneath\">Know the tools underneath</a></li><li><a href=\"#your-contribution-was-never-the-code\">Your contribution was never the code</a></li><li><a href=\"#choosing-your-game-is-choosing-your-bottleneck\" title=\"Choosing your game is choosing your bottleneck\">Choosing your game is choosing your bottleneck</a></li><li><a href=\"#the-characters-the-agent-couldn-t-make\" title=\"The characters the agent couldn&#39;t make\">The characters the agent couldn&#39;t make</a></li><li><a href=\"#a-style-the-agent-can-repeat\" title=\"A style the agent can repeat\">A style the agent can repeat</a></li><li><a href=\"#build-a-system-that-verifies-for-you\" title=\"Build a system that verifies for you\">Build a system that verifies for you</a></li><li><a href=\"#entity-component-system-ecs\" title=\"Entity Component System (ECS)\">Entity Component System (ECS)</a></li><li><a href=\"#automatic-checks-for-agents\" title=\"Automatic checks for agents\">Automatic checks for agents</a></li><li><a href=\"#the-first-playable-is-fast-the-game-is-not\" title=\"The first playable is fast, the game is not\">The first playable is fast, the game is not</a></li><li><a href=\"#how-a-game-with-agents-gets-made\" title=\"How a game with agents gets made\">How a game with agents gets made</a></li><li><a href=\"#playtest-as-a-player-report-as-a-player\" title=\"Playtest as a player, report as a player\">Playtest as a player, report as a player</a></li><li><a href=\"#the-problem-of-the-fishery\" title=\"The problem of the fishery\">The problem of the fishery</a></li><li><a href=\"#checking-the-economy-with-math\" title=\"Checking the economy with math\">Checking the economy with math</a></li><li><a href=\"#how-to-trust-code-you-never-read\" title=\"How to trust code you never read\">How to trust code you never read</a></li><li><a href=\"#what-the-checks-miss\" title=\"What the checks miss\">What the checks miss</a></li><li><a href=\"#bots-are-not-players\" title=\"Bots are not players\">Bots are not players</a></li><li><a href=\"#know-the-tools-underneath\" title=\"Know the tools underneath\">Know the tools underneath</a></li><li><a href=\"#your-contribution-was-never-the-code\" title=\"Your contribution was never the code\">Your contribution was never the code</a></li></ol>\n<p>I recently built two full games with AI:</p>\n<ul><li><a href=\"https://inkline.typeonce.dev/\" rel=\"nofollow ugc noopener\">Inkline Town</a>, a 3D town builder drawn in ink</li><li><a href=\"https://github.com/SandroMaglione/spirit-call\" rel=\"nofollow ugc noopener\">Spirit Call</a>, a 3D top-down shooter</li></ul>\n<p>I did not write a line of code. <strong>I did not read one either</strong>.</p>\n<p>Loading interactive example…</p>\n<p>My &quot;game editor&quot; was a chat, a link to playtest the game, and a web page with previews of 3D models. My &quot;work&quot; was sending text and screenshots. Nothing else.</p>\n<p>And still, making those games has been hard and intense. Like never before. For weeks.</p>\n<blockquote><p>AI made writing a game faster and easier. It did not make it easy.</p></blockquote>\n<p>Loading interactive example…</p>\n<p>Commits per day across both games, all branches, merges and automatic checkpoints excluded.</p>\n<p>I really enjoy game development. Like, a lot. But game dev is intense, time-consuming, and often has a bad financial return. So much so that I wrote a <a href=\"https://www.sandromaglione.com/articles/the-joy-and-curse-of-game-development\" rel=\"nofollow ugc noopener\">full essay</a> about my frustration.</p>\n<p>But AI changed everything!</p>\n<p>A project that used to take months is now a weekend of prompting, and I wanted to see if that holds for a complete game, not just a slop demo.</p>\n<p>My goal was getting an answer to the question all game devs are asking themselves:</p>\n<blockquote><p>If AI implements everything, what&#39;s left for me?</p></blockquote>\n<p><strong>I found the answer</strong>: The hard part didn&#39;t go away, it moved. Away from implementation, and into the choices only you can make: what game, what style, what is fun, what is broken.</p>\n<p>AI frees your time, so you can think about the core of your game. And that&#39;s hard, and now it&#39;s all you have to do, all the time. No coding break.</p>\n<p>Loading interactive example…</p>\n<p>Great. Well, what does this mean, in practice, for you?</p>\n<p>Here are all the steps I will take for my next game, everything that I learned, and how you can steal all of it.</p>\n<h2 id=\"choosing-your-game-is-choosing-your-bottleneck\">Choosing your game is choosing your bottleneck</h2>\n<p>The first decision is yours, and it will always be so: &quot;What game do *<em>I*</em> want to make?&quot;</p>\n<p>The answer you choose has the potential to make or break you and your game. Literally.</p>\n<blockquote><p>Any step in your pipeline the AI can&#39;t do becomes your full-time job.</p></blockquote>\n<p>Loading interactive example…</p>\n<p>So when you pick a genre and a style, you are also picking where your time will go. Agents go fast, so whatever still needs you becomes the bottleneck. Make sure that step is one worth your time: taste, fun, playtest.</p>\n<h3 id=\"the-characters-the-agent-couldn-t-make\">The characters the agent couldn&#39;t make</h3>\n<p>My plan for Spirit Call was a game like Genshin Impact (原神), with a new character every few weeks.</p>\n<p>So the question became: can the agent produce high-quality characters, without me stepping in? And, spoiler, the answer was no (after two days of attempts).</p>\n<blockquote><p>Simple stylized models worked well. <strong>Detailed anime characters failed every time</strong>.</p></blockquote>\n<p>I tried every route I could think of: reference sheets, prebuilt models, a full studio-style pipeline, even copying a character from another game. The results were bad, and worse, never repeatable. And all of this before animation.</p>\n<p>Loading interactive example…</p>\n<p>I could have done the anime characters by hand (maybe). Even with AI support, one decent character would have taken me weeks.</p>\n<p>So I changed the style instead of the game: from full anime characters to simple stylized spirits. Not the initial vision, but one I could ship, fast, with AI in every step of the pipeline.</p>\n<p>Loading interactive example…</p>\n<p>Spirit Call&#39;s ten spirit models by hand (an estimate), against the whole game, first commit to last, and the spirits as the agent made them. Calendar days on one axis.</p>\n<p>Even the simple spirits would have cost more by hand than the whole game took: about a month for all ten, against three weeks for all of Spirit Call. With the agent, they took about 20 minutes each:</p>\n<ol><li>2D design from ChatGPT</li><li>Low-poly version of the design, again from ChatGPT</li><li>Breakdown of every view for 3D modeling, still ChatGPT</li><li>3D modeling with Blender MCP</li></ol>\n<p>They were also far easier to animate. Each enemy went through the same steps.</p>\n<p>Loading interactive example…</p>\n<h3 id=\"a-style-the-agent-can-repeat\">A style the agent can repeat</h3>\n<p>I learned my lesson with my second game, Inkline Town: an ink style, assets made by an agent, with a repeatable pipeline from day one.</p>\n<p>Agents are not yet good at anything that requires taste: character design, detailed 3D models, animation, story. Any of it adds more work for you, personally.</p>\n<p>Loading interactive example…</p>\n<p>My ranking of genres by how much still lands on you, with the two games where they sit. It&#39;s a judgment, not a measurement: your scope and style can change the order.</p>\n<p>If you want an agent-first fast loop, <strong>your role should only be playtesting and taste</strong>. Anything else, and the loop is broken.</p>\n<h2 id=\"build-a-system-that-verifies-for-you\">Build a system that verifies for you</h2>\n<p>You should not need to read the code. The agent can handle it all. But you need layers of verification on top to make the output reliable. It starts with structure.</p>\n<h3 id=\"entity-component-system-ecs\">Entity Component System (ECS)</h3>\n<p>I use an <strong>Entity Component System (ECS)</strong>: every object in the game is an entity, a bag of data (components) like position, velocity or health. Each piece of logic (a system) declares which components it reads and writes, and runs on every entity that has them.</p>\n<p>Systems never call each other. That&#39;s what makes the architecture <strong>composable</strong>: the agent can add or rewrite one system without touching the rest, and since every system declares what it touches, the checks have something concrete to verify.</p>\n<p>Spawn a Projectile below: it has a position and a velocity, so the same Move system that moves the player moves it too. No new code.</p>\n<p>Loading interactive example…</p>\n<p>Below is the same rule on a real frame of Spirit Call. The spirit, every enemy and every bolt in flight are <code>entities</code>: pick components, and every <code>entity</code> that has all of them lights up. That is how a system finds what to run on.</p>\n<p>Loading interactive example…</p>\n<p>Spirit Call&#39;s enemy practice on Oct 4: one player, eight enemies and three enemy bolts, queried by component.</p>\n<h3 id=\"automatic-checks-for-agents\">Automatic checks for agents</h3>\n<p>On top of that structure, you layer automatic checks. Each one takes a set of bugs off your mind:</p>\n<ul><li><strong>Types</strong>: a wrong call doesn&#39;t compile, and the error tells the agent how to fix it. That&#39;s usually the first layer of checks</li><li><strong>Headless tests and simulation</strong>: the game runs as numbers and text, no graphics. The agent can play thousands of frames and verify behavior without you</li><li><strong>Lints</strong>: rules that catch the mistakes an agent keeps repeating. When the same thing breaks twice, ask for a lint</li><li><strong>Every bug becomes a test</strong>: once fixed, it can&#39;t come back silently in the same form</li><li><strong>Recorded playtests</strong>: the game records what happens while you play. Your session becomes data the agent can read</li><li><strong>Debug sessions</strong>: the agent asks why a value changed and gets an answer, instead of guessing</li></ul>\n<p>The checks run between my prompt and my next playtest: the agent sees the failures, fixes them, and I never hear about most of them.</p>\n<p>You can get creative with checks, depending on your game. In Spirit Call, I made the agent write a balance model to <em>mathematically</em> prove that clearing each main quest once pays for leveling a character up to the next one, without grinding.</p>\n<p>Drag the sliders: cut the quest rewards or make level-ups pricier, and see which quest first needs replays to afford.</p>\n<p>Loading interactive example…</p>\n<p>None of these is mandatory. But the more of them you add (where they make sense for your game), the faster you iterate, and the more of your attention goes to taste instead of code and bugs.</p>\n<p>You still need to know what each layer protects you from, even without knowing the implementation details. That&#39;s what lets you direct the agent when something keeps going wrong.</p>\n<blockquote><p>I created <a href=\"https://github.com/typeonce-dev/ritmo\" rel=\"nofollow ugc noopener\">Ritmo</a> for this exact purpose: a TypeScript ECS runtime I built in parallel with the two games, to collect all the checks in one place.</p></blockquote>\n<p>But the principles work with any engine.</p>\n<p>Loading interactive example…</p>\n<p>Below is a real session: my prompt, the agent&#39;s thinking, and how Ritmo helped:</p>\n<p>Loading interactive example…</p>\n<p>The agent&#39;s session on Oct 18. Thinking, commands and output are verbatim; the lint is new in Ritmo 0.9.0, and the fix shipped as commit c801b02.</p>\n<h2 id=\"the-first-playable-is-fast-the-game-is-not\">The first playable is fast, the game is not</h2>\n<p>With all of that in place, the first version of your game loop comes quickly. Days, sometimes hours. Then the real work starts.</p>\n<p>Everything that comes later, an <strong>AI cannot do alone</strong>:</p>\n<ul><li>Is the game fun and balanced to play for longer sessions?</li><li>Is the style clear and visually appealing?</li><li>Would a player understand how to play?</li><li>Are the assets and animations satisfying?</li><li>Is the story appealing?</li></ul>\n<blockquote><p>Three quarters of Inkline Town&#39;s commits came after the first two days, when the game was already playable.</p></blockquote>\n<h3 id=\"how-a-game-with-agents-gets-made\">How a game with agents gets made</h3>\n<p>The start was fast. I found a <a href=\"https://x.com/dwhitedesign/status/2109377823773213134\" rel=\"nofollow ugc noopener\">style reference on X</a>, a short video, and asked the agent to recreate it, animations included: five minutes.</p>\n<p>Then I asked the agent to turn it into a town simulation with a few basic buildings: working ten minutes later. The core of the game took about 20 minutes. Defining the buildings and a few mechanics came next. All of that fit in the first two days.</p>\n<p>The screenshots from those days show it: a playable town within the first hour, and a real-looking one by the evening.</p>\n<p>Loading interactive example…</p>\n<p>Inkline Town on Oct 12, its first day, and on Oct 16.</p>\n<p>Two days for something playable, then <strong>six more</strong> for content, fun and balance.</p>\n<p>Just check Inkline Town&#39;s own commits. From day two on, most of them tune something after a play session instead of building something new: nearly two thirds of all game commits.</p>\n<p>Loading interactive example…</p>\n<p>The agent implements each of those changes in minutes. What takes time is figuring out <em>what</em> to implement: which mechanic is missing, why a system is boring, what a new player doesn&#39;t get.</p>\n<p>Every round of that loop needs me: play, notice, decide, ask, play again.</p>\n<h2 id=\"playtest-as-a-player-report-as-a-player\">Playtest as a player, report as a player</h2>\n<p>I never touched a number in Inkline Town. To this day, I don&#39;t know how they&#39;re computed or how they move: I only ever saw the game.</p>\n<p>I play for around 45 minutes, as a player. Then I report what I see:</p>\n<ul><li>a strategy that dominates every other one</li><li>an exploit that breaks the economy</li><li>a moment where I didn&#39;t know what to do</li><li>a mechanic the game clearly needs and doesn&#39;t have</li></ul>\n<h3 id=\"the-problem-of-the-fishery\">The problem of the fishery</h3>\n<p>Inkline Town&#39;s clearest case was food.</p>\n<p>The town eats fish or bread. A fishery is one building; bread needs three buildings, all running. So I built fisheries, side by side on the same lake, and food stopped being a problem. Everyone was living on fish; bread had no reason to exist.</p>\n<p>My bots had been playing the economy since the first day, and <strong>none of them noticed this bug</strong>.</p>\n<p>An agent has no sense of balance: a strategy that makes the rest of the game pointless looks like any other strategy to it.</p>\n<p>So I <em>spoke</em> the prompt below to the agent, as any human playtester would.</p>\n<p>Loading interactive example…</p>\n<p>The agent turns that into a fix, and then proves it, often with an automated economy simulation. And now that my playtests are recorded, it also has the exact numbers of the session I just played.</p>\n<h3 id=\"checking-the-economy-with-math\">Checking the economy with math</h3>\n<p>For the fisheries, the fix started with a model, not a number.</p>\n<blockquote><p>I knew that if one hole like this had slipped through, there were others, and that an economy like this can be checked with math: an agent can write that check as a test.</p></blockquote>\n<p>So I asked for an economy model that weighs every number in the game against the others.</p>\n<p>It confirmed the fishery, and showed it wasn&#39;t alone: twelve rules of the economy were broken. Finding the issues took most of the time; fixing them was fast and easy.</p>\n<p>The problems that broke the economy never came back, because for each one we added the assumption behind it to the model. What still needed my playtests was the kind of choices I wanted the player to face.</p>\n<p>Getting from the ledger to the fix took the agent under an hour. My part was the 45-minute match, the report, and knowing to ask for the model.</p>\n<p>Loading interactive example…</p>\n<p>Inkline Town&#39;s economy ledger on Oct 15: the rules the game broke, at each commit from the one that wrote the ledger to the one where every rule held.</p>\n<h2 id=\"how-to-trust-code-you-never-read\">How to trust code you never read</h2>\n<p>Some of the agent&#39;s work takes hours. I don&#39;t watch it. I don&#39;t review it.</p>\n<p>When it&#39;s done, the checks have already run. Types compile, tests pass, lints are clean, simulations hold. Implementation is verified before it reaches me, against what the checks encode.</p>\n<p>Loading interactive example…</p>\n<h3 id=\"what-the-checks-miss\">What the checks miss</h3>\n<p>Passing the checks doesn&#39;t mean the game is ready. What reaches me is what they miss: a side panel hiding a button, an unclear tutorial line, a mechanic that works exactly as specified but isn&#39;t fun.</p>\n<p>Simple but glaring example: an empty side panel covering half the map. I sent the screenshot; half an hour and multiple iterations later, it was fixed.</p>\n<p>Loading interactive example…</p>\n<p>Inkline Town on a small screen, Oct 14, before and after the fix.</p>\n<p>Then I open the game (never the code) and restart it from the very beginning to judge what no check measures: <strong>whether it&#39;s fun</strong>.</p>\n<p>Since I never know what the agent changed, every time I open the game I meet it fresh, like a new player: I read the UI for the first time and follow the tutorial step by step.</p>\n<p>My reports mention what <em>felt</em> off and why, <strong>never how to fix it</strong>. My sessions are recorded, so the agent can replay my actions, take its own screenshots, and see the issue.</p>\n<h3 id=\"bots-are-not-players\">Bots are not players</h3>\n<p>You can even create various bot profiles, each with different skills, as a check to balance the game:</p>\n<p>Loading interactive example…</p>\n<p>Spirit Call&#39;s balance matrix on Oct 9: clears out of 48 runs (8 quests, 2 seeds, every aspect) for each character, by bot profile.</p>\n<p>The bots show which quests can be cleared, and catch it when a change breaks one. They don&#39;t tell you how many players would clear them, or whether it&#39;s fun.</p>\n<h3 id=\"know-the-tools-underneath\">Know the tools underneath</h3>\n<p>The checks are only as good as the rules behind them. When one keeps failing, you need to tell whether the agent got it wrong or a rule is missing, and for that you need to know the tools underneath.</p>\n<p>My trust came from the stack: my own engine plus <a href=\"https://effect.website\" rel=\"nofollow ugc noopener\">Effect</a>, which I know well enough to let the agent run.</p>\n<p>If you&#39;re not into development or math, expect to hit a wall at some point.</p>\n<p>Remember the fisheries: there, the wall was knowing to ask for a math model. Without that, you would be stuck in a much longer loop, fixing each case by hand from playtests.</p>\n<h2 id=\"your-contribution-was-never-the-code\">Your contribution was never the code</h2>\n<p>If you have been building your game by hand for years, I know you are screaming: &quot;But it stops being <em>my</em> game if I don&#39;t write any of it!&quot;</p>\n<p>It doesn&#39;t. Your contribution was never the code. It was the ideas, the feel, the hundreds of small decisions about what&#39;s fun. <strong>That is exactly the part AI can&#39;t do</strong>, and the part that now gets all your time.</p>\n<p>Loading interactive example…</p>\n<p>If you enjoy writing every line, keep doing it. But if you want to make and sell games, building by hand means years on a game you&#39;re not sure will work, while an agent writes a full system in minutes, and the models keep getting better.</p>\n<blockquote><p>My prediction: on average, games made with an AI system will ship faster <strong>and</strong> sell better.</p></blockquote>\n<p>Not because AI makes better games, but because the time no longer spent on code goes into playtesting, fun, and the bugs only a human catches. If you keep writing code, you simply don&#39;t have enough resources for all of it.</p>\n<hr />\n<p>If you want to make a game, here is the order:</p>\n<ol><li><strong>Pick your game and style with the bottleneck in mind</strong>: test early that the AI can produce what you need</li><li><strong>Build verification alongside the first playable</strong>: start with types, lints and tests for the core rules, whatever your engine. Add headless simulations and math models of your numbers as playtests show what needs protecting. <a href=\"https://github.com/typeonce-dev/ritmo\" rel=\"nofollow ugc noopener\">Ritmo</a> has them all built in, but it isn&#39;t required</li><li><strong>Verify that the core gameplay is fun, early</strong>: and expect most of the work to come after it</li><li><strong>Playtest as a player, report what&#39;s broken, let the agent fix and prove it</strong>: repeat. Ask for what the player would notice, never for code</li></ol>\n<p>Loading interactive example…</p>\n<p>It&#39;s still hard. It&#39;s faster. And, if you ask me, way more fun than before.</p>\n<p>Why more fun? Because my time goes to the part I care about. Since code doesn&#39;t matter much, the focus is all on gameplay and game feel. I am playing more instead of bothering with code details.</p>\n<p>And the final game is still mine, something that no agent alone could make.</p>\n<h2 id=\"what-changed\">What changed</h2>\n<ol><li><p>9 Nov 2026</p><p> Revised after reader feedback, with shorter sections.</p>\n<p> New: screenshots from development, an interactive Spirit Call frame, a real agent session, and a chart of bots playing the game.</p>\n<p> All of it comes from the games&#39; repositories and the agents&#39; own session logs.</p></li></ol>","headings":[{"level":1,"text":"Indie game development with AI","id":"indie-game-development-with-ai"},{"level":2,"text":"Choosing your game is choosing your bottleneck","id":"choosing-your-game-is-choosing-your-bottleneck"},{"level":3,"text":"The characters the agent couldn't make","id":"the-characters-the-agent-couldn-t-make"},{"level":3,"text":"A style the agent can repeat","id":"a-style-the-agent-can-repeat"},{"level":2,"text":"Build a system that verifies for you","id":"build-a-system-that-verifies-for-you"},{"level":3,"text":"Entity Component System (ECS)","id":"entity-component-system-ecs"},{"level":3,"text":"Automatic checks for agents","id":"automatic-checks-for-agents"},{"level":2,"text":"The first playable is fast, the game is not","id":"the-first-playable-is-fast-the-game-is-not"},{"level":3,"text":"How a game with agents gets made","id":"how-a-game-with-agents-gets-made"},{"level":2,"text":"Playtest as a player, report as a player","id":"playtest-as-a-player-report-as-a-player"},{"level":3,"text":"The problem of the fishery","id":"the-problem-of-the-fishery"},{"level":3,"text":"Checking the economy with math","id":"checking-the-economy-with-math"},{"level":2,"text":"How to trust code you never read","id":"how-to-trust-code-you-never-read"},{"level":3,"text":"What the checks miss","id":"what-the-checks-miss"},{"level":3,"text":"Bots are not players","id":"bots-are-not-players"},{"level":3,"text":"Know the tools underneath","id":"know-the-tools-underneath"},{"level":2,"text":"Your contribution was never the code","id":"your-contribution-was-never-the-code"},{"level":2,"text":"What changed","id":"what-changed"}]}}