In the spring of 2023, Ubisoft's La Forge AI research team publicly demonstrated Ghostwriter — a tool that generates first-draft barks (brief ambient dialogue lines) for NPCs. The announcement was careful to frame Ghostwriter as a time-saver for narrative designers, not a replacement. But the subtext was unmistakable: a task that once consumed weeks of a writer-designer's schedule could now be seeded in hours. The designer's job didn't disappear. It transformed. Instead of authoring every line, the designer became a curator — evaluating, refining, and injecting personality into outputs the machine had already roughed in.
That shift — from sole author to creative director of an AI collaborator — is the defining professional transition of this era in game design.
For most of the industry's history, the game designer occupied a role that blended craft, intuition, and exhaustive manual iteration. Level designers hand-placed every enemy, every pickup, every sight line. Narrative designers wrote tens of thousands of lines of branching dialogue. Systems designers balanced spreadsheets by hand, running play sessions to test each tweak. The work was granular, painstaking, and deeply personal — the designer's fingerprints were on every decision.
This model produced masterpieces. It also produced chronic crunch, ballooning budgets, and teams of hundreds required just to ship a mid-size open-world title. The human cost of total authorship at scale became increasingly unsustainable as player expectations for content volume kept rising.
The emergence of procedural generation tools in the 2000s (Spelunky's cave generator, Minecraft's terrain system, No Man's Sky's planetary engine) hinted at a different model — one where the designer authored systems rather than content. AI-assisted design is the next evolution of that same principle, extended into nearly every domain of the discipline.
The orchestrator model reframes the designer's core responsibility. Rather than producing final artifacts directly, the designer defines intent, configures tools, evaluates outputs, and makes judgment calls about what the AI produced. The craft moves upstream — from execution to direction.
In practice this manifests across multiple disciplines simultaneously. A level designer using NVIDIA's AI-assisted layout tools sets traversal goals and constraint parameters, then reviews generated layouts against feel and pacing criteria. A narrative designer working with a large language model provides character voice, thematic guardrails, and emotional beats — then edits the generated text rather than writing from scratch. A systems designer uses ML-based simulation to run thousands of balance iterations overnight, then interprets the data in the morning to make design decisions.
None of these workflows eliminate judgment. They compress the distance between intent and testable artifact, which means the designer spends more time evaluating and less time producing raw material. The skill set required shifts accordingly: critical evaluation, prompt engineering, AI output literacy, and the ability to articulate design intent precisely become as important as traditional craft skills.
Real Case · Ubisoft Ghostwriter (2023)
Ubisoft's La Forge team built Ghostwriter specifically so narrative designers would never have to write a first draft of NPC barks again. The tool generates contextually appropriate lines based on character parameters, situation tags, and emotional tone inputs. Designers then curate and edit. Ubisoft reported in their GDC 2023 presentation that bark authoring time was reduced significantly — freeing writers for higher-order narrative work: quest design, character arc development, and systemic dialogue logic. The tool is an internal product, not commercially available, precisely because it encodes Ubisoft's proprietary character voice data.
The orchestrator model does not erase traditional design skills — it recontextualizes them. Understanding why a level feels good is still essential; you simply need that understanding to evaluate AI-generated levels rather than to build them tile by tile. Story structure knowledge is still vital; you need it to recognize when an LLM-generated scene violates character or undercuts thematic resonance. Game feel and balancing instincts remain indispensable as interpretive tools for reading AI output intelligently.
What changes is where those skills are applied. The designer as orchestrator is always operating one level of abstraction higher than the designer as direct author. This demands clearer thinking about design intent, more explicit documentation of goals and constraints, and a new literacy around what AI systems can and cannot reliably do. Designers who develop this literacy early gain a meaningful professional advantage in an industry undergoing rapid structural change.
Core Idea
AI tools don't remove the designer from the creative loop — they relocate the designer to an earlier, more strategic position in that loop. The question shifts from "How do I make this?" to "What do I want this to be, and how do I direct the system to approximate it?"
The orchestrator model requires you to articulate design intent with precision before interacting with any AI tool. In this lab you'll practice that skill: describe a game design problem or creative goal in clear terms, and the AI will help you refine your intent statement, identify what constraints you'd need to specify, and discuss how a real orchestrator workflow would handle it.
Think about a design task — a level feeling, a character voice, a balance problem — you'd want AI help with. Then describe it to the assistant below and work through how you'd frame it as orchestrator direction rather than direct authorship.
When Hello Games shipped No Man's Sky in 2016, its 18-person team had authored a universe of 18 quintillion planets without placing a single rock by hand. The designers' job wasn't terrain art — it was rule authorship: what parameters govern atmospheric color? What biological logic determines creature morphology? What economic rules make star systems feel distinct? The artifact of their labor was an algorithm, not a landscape. By the time players arrived, Hello Games designers had become expert systems thinkers — people who could reason about emergent complexity and hold entire rule structures in their heads simultaneously. That competency profile is increasingly what the industry needs from everyone.
Prompt engineering — the craft of communicating with a generative AI system to elicit useful output — has rapidly become a recognized professional skill in game development contexts. It is not simply typing instructions; it is a structured communication discipline that requires understanding how a model represents intent, where it tends to over-generalize or hallucinate, and how to use constraints, examples, and iteration to guide outputs toward a specific creative goal.
For narrative designers, effective prompting means providing character voice anchors, genre context, tone boundaries, and negative examples (what the output should not sound like). For level designers using AI layout tools, it means specifying spatial constraints, player ability assumptions, pacing goals, and reference touchstones. The designer who can prompt precisely generates useful drafts; the designer who cannot gets generic output that requires as much rework as starting from scratch.
Practical prompt skills include: writing role-context primes ("You are assisting a designer working on a post-apocalyptic survival game with a grim tone and low-magic world"), using few-shot examples to anchor style, decomposing complex requests into sequential prompts, and knowing when to reject a model's output entirely versus when to iterate on it.
AI output literacy is the ability to evaluate what a generative system has produced — to read it critically, identify where it succeeds and fails against design intent, and make informed decisions about what to keep, what to modify, and what to discard. This is a skill distinct from prompt engineering; it operates on the other side of the interaction, after the output has been generated.
In narrative contexts, AI output literacy means recognizing when generated dialogue has the right surface structure but wrong character voice — when it sounds plausible but lacks the specific idiosyncratic quality that makes a character feel real. In systems contexts, it means reading ML-generated balance data critically rather than deferring to whatever the model converged on — asking whether the "optimal" configuration a simulation found is actually fun, fair, or aligned with the game's intended difficulty curve.
This critical literacy is what separates a designer who uses AI well from one who simply accepts its outputs. Garbage in, garbage out is the old data maxim — but in AI-assisted design, the more insidious failure is plausible output uncritically accepted. AI systems are very good at producing outputs that feel acceptable. The designer's job is to hold them to a higher standard: not just "does this work?" but "does this serve the specific design goal?"
Real Case · Insomniac Games · Systemic Design
Insomniac Games' design team on Marvel's Spider-Man (2018) and its sequels used simulation tools to run combat encounter balance tests at scale — testing enemy compositions, spawn timing, and arena geometry across thousands of simulated play-throughs before human testers saw the content. The designers reading that simulation data needed to know not just what the numbers said, but what they meant for moment-to-moment feel. High kill rates in a combat arena didn't simply mean "make enemies weaker" — they required the designer to reason about whether the arena layout was the real problem, whether the encounter pacing needed restructuring, or whether the player's available toolkit in that section was incomplete. Simulation fluency is now a core competency, not a specialist one.
Not every traditional design skill is disrupted equally. Some become more important in an AI-assisted workflow, not less. Game feel — the embodied, intuitive sense of whether an interaction is satisfying — cannot currently be generated or evaluated by AI. A designer must still possess the tacit knowledge to play a prototype and know, in their body, that the jump arc is slightly too floaty. That judgment cannot be offloaded.
Narrative craft knowledge — understanding story structure, character motivation, thematic coherence, and emotional pacing — is essential precisely because AI language models can violate all of these while producing text that reads smoothly. The designer without strong narrative foundations cannot catch those failures. Similarly, systems design intuition — the sense of how emergent complexity arises from simple rules, where feedback loops destabilize, and what makes a decision space feel meaningful — remains entirely a human responsibility. AI can simulate, but it cannot yet intuit.
The designer of the AI era is someone who has deepened traditional craft skills and layered new AI-literacy competencies on top of them. Depth in both dimensions is what the evolving role demands.
Competency Map
New competencies to develop: prompt engineering, AI output literacy, intent documentation, simulation data interpretation, workflow integration. Traditional competencies to deepen: game feel judgment, narrative craft, systems design intuition, player psychology, creative vision-holding under constraint.
Prompt engineering for game design is about specifying context, constraints, tone, and intent with enough precision that an AI collaborator can produce useful drafts. In this lab, the assistant will help you build and critique prompts for real design tasks — level design briefs, character voice profiles, balance problem descriptions, and more.
Try writing a prompt for a specific game design task, then ask the assistant to evaluate whether it's precise enough, what's missing, and how you might tighten it. Compare different versions of the same prompt and discuss what changes.
In the summer of 2023, the Writers Guild of America's strike against major studios included, for the first time in labor history, explicit demands around AI usage in creative work. Writers sought guarantees that AI could not be used to generate scripts that human writers would then be paid to "polish" — which would effectively set a lower wage floor for the same work. The studios sought flexibility to use AI tools however they saw fit. The eventual contract included provisions requiring studios to disclose AI use, prohibiting AI-generated material from being treated as source material writers must adapt, and protecting writers' credit. The video game industry — watching this closely — faces analogous questions. When a tool generates 40% of a game's dialogue, who gets the writing credit? What does authorship mean?
Authorship in creative fields has always been messier than the lone genius myth suggests. Game development is radically collaborative by nature — credits pages list hundreds of contributors across disciplines, and even a "lead designer's" vision is inevitably shaped, constrained, and transformed by every other person on the team. AI tools add a new non-human collaborator to this already complex attribution landscape.
The key legal and ethical question is not simply "did AI help?" — tools always help, from word processors to level editors to physics engines. The question is whether AI contribution is material to the creative output in a way that displaces human authorship. In the U.S., the Copyright Office's 2023 guidance clarified that AI-generated content without sufficient human creative input is not copyrightable. The designer who reviews and selects from AI-generated options has exercised editorial judgment — which constitutes human creative authorship. The designer who accepts AI output wholesale with no meaningful curation may not have produced a copyrightable work.
For working designers, this creates a practical imperative: document your creative decisions. Version history, prompt logs, selection rationale notes, and editorial records all constitute evidence of human authorship in an AI-assisted pipeline.
How studios handle AI use internally varies dramatically. Some have implemented explicit AI ethics policies — Paradox Interactive, EA, and others have issued public statements about how they will or will not use generative AI in their pipelines. Others have adopted AI tools quietly, without communicating clearly with their teams. The absence of clear policy tends to produce anxiety among designers who don't know whether their role is being actively automated, what the studio's legal exposure is, or how AI use will be attributed.
Best-practice studios treat AI tools like any other significant new technology: with transparent policies, clear guidelines for use, explicit protocols for attribution and documentation, and regular team-level conversations about how workflows are evolving. Designers who have used AI tools in a project should be credited for their curatorial and editorial contribution — the same way that a photo editor who selects and processes images from a shoot is credited, even if they didn't operate the camera.
The emerging professional norm is disclose use, document contribution, credit judgment. This protects studios legally, respects designers' professional identities, and builds the team trust that effective human-AI collaboration requires.
Real Case · U.S. Copyright Office Guidance · 2023
In March 2023, the U.S. Copyright Office issued clarifying guidance on AI-generated content in response to a registration application for the graphic novel "Zarya of the Dawn," whose creator, Kristina Kashtanova, had used Midjourney to generate its images. The Office allowed copyright protection for the text and for the selection and arrangement of images, but denied protection for the AI-generated images themselves. This established a key principle: human creative selection, arrangement, and editorial judgment can constitute protectable authorship even when underlying material is AI-generated. Game designers curating AI outputs are in a legally similar position to Kashtanova — their judgment is the copyrightable contribution.
Many experienced game designers report a complex emotional relationship with AI tools — recognizing their utility while feeling threatened by the implication that the craft skills they've developed over careers are being devalued. This is a legitimate response to real structural change, not simply irrational anxiety. Industries that have undergone similar tool-driven transitions (photography replacing portrait painting, digital audio workstations transforming music production, CAD replacing hand-drafted architectural drawings) show consistent patterns: the practitioners who adapt become more productive and creatively ambitious; those who resist adaptation struggle to find work.
The key psychological reframe for designers is from craft identity to creative vision identity. Your value is not in your ability to produce artifacts manually at speed — it is in your ability to hold a coherent creative vision, make judgment calls that advance that vision, and direct whatever tools are available toward it. That identity is AI-resistant in a way that manual production skill is not. A designer who identifies primarily with "I write good dialogue" is more vulnerable than one who identifies with "I create characters players genuinely care about."
Professional Imperative
Document your creative decisions when using AI tools. Record what you prompted, what you received, what you accepted, modified, or rejected, and why. This record is both your legal protection and your professional portfolio — evidence that the design work reflects your judgment and vision.
When AI contributes to a game's creative content, designers face practical questions about credit, documentation, and professional identity. In this lab, present authorship dilemmas or credit scenarios you'd face as a designer using AI tools, and work through how the principles from this lesson apply to real decisions.
You might describe a situation — e.g., an AI generated 60% of your game's ambient dialogue, or your studio wants to use AI-generated concept art without disclosure — and ask the assistant to help you reason through the ethical, legal, and professional dimensions.
The most common mistake designers make when encountering AI tools is trying to evaluate the whole landscape at once — reading about diffusion models, transformer architectures, real-time NPC systems, and procedural level generators simultaneously, and concluding that the space is too large to enter. It is not. AI literacy for game designers is a skill that builds sequentially, and the entry point is lower than most people expect. You do not need to understand how a neural network learns. You need to understand what these tools can do, what they cannot do, and how to direct them toward your specific design goals. This lesson gives you a concrete starting sequence.
Image generation is the most accessible AI domain for game designers because the feedback loop is immediate and visual. Tools like Midjourney, Stable Diffusion, and Adobe Firefly let you generate concept art, environment mood boards, character studies, UI mockups, and prop references in minutes. You do not need to draw. You do need to develop taste and judgment about what good output looks like and how to direct the system toward it.
Start by using image generation for a current project or personal practice. Pick a specific design problem — a character you are imagining, an environment that needs a mood reference, a UI style you want to explore — and generate 20 to 30 variations. Notice what the tool does well by default. Notice where it fails. Notice what prompting decisions produce qualitatively different results. This hands-on iteration is how you develop AI output literacy; reading about it produces almost none.
Practical starting projects: generate a style reference sheet for a game you are designing; create 10 variations of the same character in different emotional states; produce environment concept art for a level that needs visual development. The goal is not polished output — it is developing your ability to direct the tool and evaluate results.
Prompt engineering is not a mysterious technical skill — it is precision communication. The core principles apply across all generative AI tools, visual or text-based. Learning them in one domain makes you faster in all others.
Role and context priming. Open every prompt by establishing who the AI is working with and in what context: "You are assisting a narrative designer working on a 2D side-scrolling platformer set in 1980s Tokyo. The tone is melancholy but not depressing — think early Studio Ghibli." This context shapes everything the model produces subsequently.
Specificity over generality. "Generate an enemy character" produces generic output. "Generate an enemy character who is an aging bureaucrat who has been corrupted by power — visually conveys authority and physical decline simultaneously, wears formal clothes that no longer fit well" produces something useful. Every vague word in your prompt is an invitation for the model to default to statistical average.
Negative specification. Tell the model what you do not want: "Do not make this character conventionally attractive. Do not use fantasy armor. Avoid European medieval visual references." Negative constraints are often more effective than positive ones at steering away from default outputs.
Iterative refinement. Treat the first output as a draft, not a result. Identify specifically what is wrong — "the lighting is too dramatic, the character reads as heroic rather than morally ambiguous" — and add targeted corrections. Three cycles of refinement typically produce dramatically better results than one heavily engineered first prompt.
STEP 3: UNDERSTAND MODEL LIMITATIONS
Every AI tool has a characteristic failure mode. Image generators struggle with hands, text within images, consistent character identity across multiple generations, and precise spatial relationships. Language models hallucinate facts, lose track of character voice across long outputs, and tend toward resolution and positivity when ambiguity and tension are what the story needs. NPC dialogue systems have trouble maintaining long-term narrative consistency and often produce characters who feel helpful and compliant rather than genuinely motivated. Knowing the failure modes of the tools you use is not pessimism — it is the quality control knowledge you need to catch problems before they ship.
One of the highest-value, lowest-barrier AI exercises for game designers is using ChatGPT or Claude to prototype NPC dialogue. The process is straightforward: write a character bible entry (name, background, motivation, speech patterns, relationship to the player), paste it as context, then conduct an interactive conversation with the AI playing the character while you play the player.
This prototype method is valuable not because the output will ship — it usually will not — but because it surfaces design problems quickly. Does the character feel consistent across ten exchanges? Does their motivation come through in dialogue without explicit exposition? Do they respond differently to friendly versus hostile player approaches? Problems that would take weeks to discover in a full implementation surface in an afternoon of AI prototyping.
Practical exercise: write a character brief for an NPC in a game you are designing. Include background, motivation, verbal tics, and what they know and do not know about the player. Ask ChatGPT to play this character while you ask it questions a player might ask. After ten exchanges, evaluate: what worked, what felt off, what did the character consistently get wrong? Revise the brief and repeat. You are iterating the character design, not just the AI output.
Being clear-eyed about where AI will and will not change game design workflows is essential for making good career decisions over the next decade. The picture is not uniformly threatening or uniformly reassuring — it is specific.
What AI will substantially augment: Concept art generation and early visual development (AI can produce 50 variations in the time a human produces 5 — the human still art-directs and selects). First-draft ambient dialogue and NPC barks. Procedural asset variation (texture variations, prop recolors, minor mesh modifications). Balance simulation and playtesting at scale. Localization first-pass drafts. Documentation and design brief writing. These are tasks where volume and iteration speed matter more than singular creative vision.
What AI will not replace in any foreseeable timeframe: Creative direction — deciding what the game should feel like and holding that vision consistently under production pressure. Player empathy — the intuitive judgment that comes from deeply understanding what it feels like to be a player in a specific moment. Narrative judgment — knowing when a story beat is landing and when it is not, and why. Game feel — the embodied sense that an interaction is or is not satisfying. Systemic design intuition — knowing that a particular rule change will produce emergent behavior that breaks the game's intended experience in ways that no simulation predicted. These capacities require human experience, human embodiment, and human taste.
The designer who builds AI literacy now is positioned to spend more time doing what AI cannot do, because AI handles more of what it can. That is not job loss — it is a leverage multiplier for the skills that matter most.
YOUR ACTION PLAN — START THIS WEEK
1. Pick one image generation tool (Midjourney free tier, Adobe Firefly, or Stable Diffusion via DreamStudio) and generate 20 variations of a design problem from a current project. 2. Write one prompt engineering brief: context, specificity, negative constraints. Compare output to a vague prompt. 3. Write a one-page character brief and prototype 10 NPC exchanges in ChatGPT. Identify three things the character consistently got wrong. 4. Subscribe to one AI game design newsletter or follow two practitioners on social media. Stay current; this field moves fast. Designers who build this literacy now will have a meaningful advantage as these tools become standard in every studio pipeline.
Apply and extend the concepts from this lesson through guided conversation with an AI assistant.
Use this lab to explore how the concepts from Lesson 4 apply to your own questions and interests. The AI assistant is here to help you think through complex scenarios.
15 questions covering all lessons — free, untracked, retake anytime.