# Game Development Roles: What AI Changes for Each (2026) > Source: > Published: 2026-09-05 00:00:00+00:00 # Game Development Roles: What AI Changes for Each (2026) Game development roles fall into four groups: designers and producers who decide what to build, artists who make it visible, programmers who make it run, and marketers who get it to players. [The 2026 GDC State of the Game Industry survey](https://gdconf.com/article/gdc-2026-state-of-the-game-industry-reveals-impact-of-layoffs-generative-ai-and-more/)  of more than 2,300 developers found that 36% now use generative AI at work, but only 30% of studio staff do, against 58% at publishing and marketing firms. This post covers what each role owns, where its work gets handed to someone else, and what changes when an AI agent runs inside the [game engine](/blogs/what-is-a-game-engine) instead of in a chat window. ## The four roles and their handoffs Most guides list roles as job titles. It’s more useful to look at what each role hands to the next one, because the handoffs are where the time goes. | Role | Owns, and what it hands off | GDC 2026 view of generative AI | |---|---|---| | Designers and producers | Concept, scope, what’s fun, what makes money. Hands a design doc and a task list to everyone else. | 63% of design and narrative staff say it harms the industry | | Artists | Concept art, models, textures, animation. Hands finished assets to whoever puts them in the engine. | 64% say it harms the industry; 19% of AI users generate assets with it | | Programmers | Scenes, scripts, physics, performance, builds. Hands a build to QA and playtesters. | 59% say it harms the industry; code assistance is still the second most common use, at 47% | | Marketers and publishers | Store pages, wishlists, press, localization, ports. Hands sales and player feedback back to the team. | 58% of staff at publishers and marketing firms use it | Sources: [GDC 2026](https://gdconf.com/article/gdc-2026-state-of-the-game-industry-reveals-impact-of-layoffs-generative-ai-and-more/)  and [our breakdown of three years of GDC data](/blogs/gdc-2026-generative-ai-game-development-data). Unity’s [2025 Gaming Report](https://unity.com/blog/2025-unity-gaming-report-launch) , drawn from its own developer base, found 96% of studios use AI tools in at least one workflow, so the disagreement is about where AI belongs, not whether it’s in the building. ## Artists: put the art in the engine yourself In a traditional pipeline the artist finishes a character or an animation and hands it to a programmer. Studios employ technical artists because that handoff breaks: Unity’s own guide for the role calls them [“the bridge between programmers and artists”](https://unity.com/blog/engine-platform/complete-overview-of-unity-toolsets-workflows-for-technical-artists) . Godot’s [3D import guide](https://docs.godotengine.org/en/stable/tutorials/assets_pipeline/importing_3d_scenes/index.html)  shows what sits between the file and the game: import settings, materials, animation libraries, and the scene that ties them together. The jump that looked right in Blender lands in the game with different timing, and the artist finds out a sprint later. An agent inside the editor changes who does that step. With [Ziva](/blogs/what-is-ziva), the artist drops the animation into the Godot scene, asks for it to play on jump, and adjusts the timing while the game runs. The programmer never enters the loop, and what ships is what the artist saw. Generation is the second change, and studios are careful about it. Only [19% of AI-using developers](/blogs/gdc-2026-generative-ai-game-development-data) generate assets with it. Capcom’s technical director Kazuki Abe [built a prototype on Gemini and Imagen](https://automaton-media.com/en/game-development/capcom-is-experimenting-with-generative-ai-to-help-generate-the-hundreds-of-thousands-of-ideas-needed-for-game-development/)  to draft the “thousands to tens of thousands” of background objects a game needs, with the system scoring its own proposals against set criteria before an art director sees them. Ziva’s sprite and 3D model generation is built for the same job: it takes the existing art as the reference and fills in the missing prop or finishing detail, so art direction stays with the artist. ## Designers and producers: plan with the codebase open Research and brainstorming is the most common use of AI in game development, at [81% of the people who use it](https://gdconf.com/article/gdc-2026-state-of-the-game-industry-reveals-impact-of-layoffs-generative-ai-and-more/) . The gap is that a chatbot brainstorms without knowing the game. Most design docs are [abandoned by week two](/blogs/game-design-document), and the planner ends up asking a programmer what the code can support. Ziva’s plan mode reads the project first. It [writes a step-by-step plan and waits for approval before touching a file](/docs/slash-commands), so a designer can ask what it would take to add a daily reward loop or teach a new skill and get an answer scoped against the actual codebase. The suggestions are suggestions. The game idea stays with the studio, and the team picks which ones to build. ## Programmers: the engine is the part a chatbot can’t see Ordinary code is text, and language models are good at text. A game is not only code. Godot stores scenes as [TSCN files](https://docs.godotengine.org/en/stable/engine_details/file_formats/tscn.html)  of node trees and resource references, and Unity’s scenes are [YAML documents](https://docs.unity3d.com/Manual/FormatDescription.html)  full of fileID pointers. Both are checked by the editor, not by the model. That’s why ChatGPT still [mixes Godot 3 and 4 syntax](/blogs/claude-godot), and why a code editor like Cursor [can’t see the scene tree or the running game](/blogs/ziva-vs-cursor-for-godot). Ziva connects models like Claude and GPT to the editor itself; the reasoning is in [why Ziva is built inside a game engine](/blogs/why-ziva-is-built-inside-a-game-engine). The model edits scenes and scripts through the engine, sees the editor’s errors, and runs the game. For a studio, the verification is the part that matters: 1. Properties that don’t exist in the engine are flagged instead of shipped. 2. Code tests run after each change. 3. An [AI playtester](/full-game-tester) launches the game, sends real inputs, and records a pass or fail video. Testing is where the largest studios are heading. Square Enix’s November 2025 plan sets a goal to [automate 70% of QA and debugging tasks by the end of 2027](https://www.hd.square-enix.com/eng/news/pdf/20251106_01_en.pdf#page=5)  with a joint team from the University of Tokyo’s Matsuo-Iwasawa Laboratory, while only [22% of AI-using developers](/blogs/gdc-2026-generative-ai-game-development-data) use it for testing today. Performance work fits the same loop. Jody Mitoma, building [Beat Boxy](https://store.steampowered.com/app/4536120/Beat_Boxy/)  with Ziva at JoMito Games, asked it to find the largest memory, CPU, and GPU costs in his levels. The game ran “even smoother than it already was,” and the [total cost was 58 cents](/blogs/made-with-ziva). ## Marketers and publishers: the work after the build A build is not a launch. About 62% of Steam users run the client in a language other than English, per the [August 2026 hardware and software survey](https://store.steampowered.com/hwsurvey/) , and Valve’s 2024 figures put [Simplified Chinese at 33.7% of users, above English](https://gamedevreports.substack.com/p/valve-chinese-language-surpassed) . Localization, ports, and press are where small teams sign publishing deals, and those deals cost. In [Voyer Law’s review of more than 100 agreements](https://gamedevreports.substack.com/p/voyer-law-publishing-agreements-in) , the median developer share is 50% with an advance and 70% without. Ziva’s [publishing program](/publishing) covers that stretch for its users: builds for other platforms, playtesting, localization, social advertising, and press outreach, through partner publishers at preferential rates. The store-page side is in [how to publish a game on Steam](/blogs/publish-game-steam) and our [Next Fest wishlist data](/blogs/steam-next-fest-2026). ## Will AI replace game development roles? Not on the current data. Only [5% of AI-using developers](/blogs/gdc-2026-generative-ai-game-development-data) put AI in front of players, and 7% of the industry thinks it has a positive effect. What moves is the handoff: the artist implements, the designer plans against the codebase, the programmer verifies instead of reviewing hallucinated properties, and the marketer gets a build that’s ready for more markets. The short version is in [Ziva for game studios](/blogs/ziva-for-game-studios). [Talk to Ziva about your studio](/enterprise). Godot is available today, [Unity](/unity) is in beta for studios, and [Unreal Engine](/unreal) support is arranged with each team.