How 3D Artists Keep Creative Control While Using AI and MCP

At GDC 2026, a survey made the rounds: 52% of game developers said AI is having a negative impact on games. That number should concern anyone building AI tools for creative professionals. Not because the tools are bad, but because the conversation about AI in game development has been framed wrong from the start.

The dominant narrative has been one of replacement. AI generates art. AI writes code. AI designs levels. The implicit message: creative professionals are costs to be eliminated. That framing is wrong, it's harmful, and it's not how AI tools actually work in production.

We build AI-powered tools at StraySpark — the Blender MCP Server and the Unreal MCP Server. We've watched our users work with these tools daily. And the pattern we see isn't replacement. It's augmentation. Artists using AI to handle tedious mechanical tasks so they can spend more time on the creative decisions that actually matter.

This post is about how that works in practice. How 3D artists keep creative control while using AI and MCP. How the human-in-the-loop model works in real production. And an honest accounting of what AI can't do — because understanding the limitations is just as important as understanding the capabilities.

Why the "AI Replacing Artists" Narrative Is Wrong

Let's address the fear directly before talking about workflows.

What AI Art Tools Actually Do

Most AI tools in 3D production fall into a few categories:

Generation tools create initial content — text-to-3D models, procedural textures, generated concepts. These produce starting points, not finished assets.

Automation tools handle repetitive operations — batch processing, file conversion, property assignment, scene setup. These replace manual labor, not creative decisions.

Assistance tools provide suggestions or speed up specific tasks — auto-retopology, UV unwrapping, weight painting assistance. These are specialized helpers, not general replacements.

None of these categories replace artistic judgment. They replace the time artists spend on mechanical execution. There's a crucial difference between "deciding this character should have weathered leather armor with brass buckles" (creative judgment) and "unwrapping the UV islands on 30 buckle meshes" (mechanical execution).

The Real Anxiety

The legitimate concern isn't that AI tools are bad at art. It's that management will use "AI-generated" as a cost-cutting label regardless of quality. That's a business and labor issue, not a technology issue. Worse tools have been used as excuses to cut teams before — remember when "outsourcing" was the buzzword? Or "procedural generation means we don't need level designers"?

The answer then was the same as it is now: the best work comes from skilled people with good tools. AI is a tool. An unusually powerful tool, but still a tool that requires skilled hands to produce quality results.

What the 52% Survey Actually Means

When half of developers say AI is harming games, they're responding to real things: studios using AI-generated placeholder art and shipping it, reduced art budgets justified by "AI efficiency," a flood of low-quality AI-generated game assets on marketplaces.

These are problems. But they're not caused by AI tools themselves. They're caused by the same cost-cutting mindset that produced every previous wave of quality degradation in games. The tool is being blamed for the decisions of the people using it.

Our position: AI tools should make good artists more productive, not make bad art more acceptable.

The MCP Model: Artist Commands, AI Executes

MCP (Model Context Protocol) is fundamentally different from "AI generates art" tools. Understanding this difference is key to understanding how artists keep creative control.

What MCP Actually Is

MCP is a protocol that lets AI assistants control software through structured commands. When you connect an AI assistant to Blender through the Blender MCP Server, the AI doesn't generate 3D content from training data. It operates Blender's actual tools — the same tools you'd use manually.

Think of it like this: MCP turns an AI assistant into a very fast, very patient, voice-controlled operator of the software you already use. It can select objects, apply modifiers, adjust parameters, run scripts — but every operation is a Blender operation, producing the same results as if you'd clicked the buttons yourself.

The Critical Distinction: Generation vs. Operation

AI image generation produces output from statistical patterns learned during training. The AI's "understanding" of art is a probability distribution. The output is novel but it's shaped by training data in ways that are opaque and uncontrollable.

MCP-based automation produces output from deterministic software operations. When the AI applies a subsurface modifier in Blender, the result is exactly what Blender's subsurface algorithm produces. There's no black box. The operations are inspectable, undoable, and reproducible.

This distinction matters because it means:

Every operation is auditable. You can see exactly what the AI did.
Every operation is undoable. Standard undo/redo applies.
There are no surprise outputs. The AI can only do what Blender or Unreal can do.
The artist's file is the authority. Nothing happens outside the artist's project.

Artist as Director, AI as Assistant

The working model looks like this:

Artist decides what to create. A medieval castle gate with heavy iron-banded oak planks, rust weathering on the hinges, moss growth at the base.

Artist decides how to approach it. Model the gate frame first, then the planks, then the iron bands. Use a wood texture base with hand-painted detail. Geometry should be game-ready — under 5,000 triangles for the LOD0.

AI handles mechanical execution under artist direction. "Create a rectangular mesh, 3m wide, 4m tall, 0.3m deep. Subdivide the face into 8 vertical strips for the plank cuts. Add a loop cut horizontally at 0.5m and 3.5m for the cross braces."

Artist evaluates and directs adjustments. "The plank widths need more variation. Make the second and sixth planks 15% narrower. Bevel the edges of all planks at 0.02m for visual softness."

AI handles batch operations. "Apply this weathered iron material to all objects in the 'Hardware' collection. Set roughness to 0.85. Add a vertex color layer for rust masking on all iron pieces."

At every step, the artist is making the creative decisions. The AI is executing them faster than manual clicking.

Practical Workflow 1: Batch Operations That Free Up Creative Time

The single biggest time sink for 3D artists isn't creative work. It's mechanical repetition. Here's where AI assistance through MCP provides the most unambiguous value.

Material Assignment Across Hundreds of Objects

Scenario: You've modeled a dungeon tileset with 45 unique pieces — walls, floors, columns, arches, stairs, doorways. Each piece needs 3-4 material slots assigned: stone, metal trim, wood accents, emissive (for magical elements).

Manual process: Select each object, open material slots, assign materials, verify UV mapping per slot. Time: 2-3 hours for 45 objects with careful attention.

MCP-assisted process:

"Assign the DungeonStone material to slot 0 on all objects in the Walls collection. Assign MetalTrim to slot 1. For objects with 'Door' in the name, add WoodPlanks to slot 2."

"For the Column objects, set the emissive material to slot 3. UV scale for the emissive channel should be 2x on all columns."

"For stair pieces, rotate the stone material UV 90 degrees so the grain runs along the tread direction."

Time: 10-15 minutes, including verification. The creative decisions (which materials go where, how UVs should orient) are all the artist's. The clicking is the AI's.

Naming Convention Cleanup

Every artist has inherited a scene with objects named Cube.001 through Cube.847. Renaming is important for pipeline organization but soul-crushingly tedious.

"Rename all objects in the scene using this convention: collection name + object type + sequential number. So objects in the 'Walls' collection become Wall_Mesh_001, Wall_Mesh_002. In the 'Props' collection, use Prop_Mesh_001. Maintain the current order within each collection."

Export Preparation

Getting assets from Blender to Unreal Engine involves repetitive export setup.

"For each object in the Export collection: reset origin to bottom center, apply all transforms, check for non-manifold geometry and report any issues, set smooth shading with auto-smooth at 60 degrees, export as individual FBX files to /Exports/ with the object name as filename."

The Blender MCP Server handles the entire export pipeline while the artist focuses on making sure the assets look right in the target engine.

Practical Workflow 2: Scene Assembly and Layout

Scene assembly is where artists spend enormous time on non-creative tasks: positioning, aligning, distributing objects according to a design that already exists in their head.

Interior Set Dressing

You're dressing a tavern interior. You've modeled or sourced all the props — tables, chairs, mugs, plates, candles, wall decorations. Now you need to place them.

Creative decisions (artist):

The tavern has a busy area near the bar and a quieter area near the fireplace
Tables should feel organically placed, not grid-aligned
A few overturned chairs near the bar suggest a recent brawl
The hearth area should feel warm and inviting

Mechanical execution (AI via MCP):

"Place 8 tables in the main room area. Vary their rotation by plus or minus 15 degrees from grid alignment. Cluster 5 of them in the left half near the bar counter, space the remaining 3 more widely in the right half near the fireplace."

"For each table, place 2-4 chairs around it. Chairs should face roughly toward the table center but with 10 to 20 degrees of random rotation variation. Pull one chair out from each table slightly, as if someone just stood up."

"Near the bar, take 3 chairs and rotate them to various fallen positions. One on its side, one tilted at 40 degrees leaning against a table leg, one upside down 2 meters from the nearest table."

"Place candle props on 5 of the 8 tables. Place mug groups — 2 to 4 mugs in loose clusters — on every table."

The artist reviews the layout, makes adjustments ("move that table closer to the wall," "that fallen chair looks wrong — rotate it"), and the AI executes the changes. The creative vision is entirely the artist's. The 45 minutes of drag-and-drop positioning is reduced to 5 minutes of conversation and review.

Exterior Environment Dressing

"Along the forest path, place rock meshes from the ForestRocks collection at irregular intervals. Vary scale between 0.3 and 2.5. Larger rocks should be more common off the path edges, smaller ones can be on the path itself. Partially embed all rocks into the terrain — lower their Z position by 20-30% of their height. Add random rotation on all axes."

"At the path junction, create a small campsite: place the tent mesh facing southeast, add the campfire ring mesh 3 meters in front of the tent entrance, scatter 4-6 log seat meshes around the campfire in a loose semi-circle, add bedroll meshes inside the tent."

The Key Point

In all these workflows, the artist is doing the creative work — deciding what the space should look and feel like. The AI is doing the placement work — positioning, rotating, scaling, and distributing objects according to the artist's direction. The artist's creative control isn't diminished; their creative time is increased because less of their day is spent on mechanical positioning.

Traditional 3D work has a slow feedback loop. Make a change, render a preview, evaluate, navigate back to the parameter you want to change, make another change. MCP enables a conversational refinement loop that's dramatically faster.

Material Iteration

"The stone material is too smooth. Increase roughness from 0.4 to 0.7."

"Better. Now the normal map isn't strong enough to read at this roughness. Scale the normal strength from 1.0 to 1.8."

"The color is too uniform. Add a noise texture driving a color ramp between the current stone color and a slightly darker, greener variant. Noise scale around 3.0."

"The green is too saturated. Pull the secondary color toward gray-green. And reduce the noise contrast — the boundary between colors should be softer."

Each of these adjustments is 10-15 seconds through MCP versus 30-60 seconds of navigating node graphs manually. Over hundreds of iterations across a project, that adds up significantly.

"Add a bevel modifier to the table mesh. Width 0.005, 3 segments."

"The bevel is catching the bottom edge — add the bottom face to the bevel weight exclusion."

"The table leg profile needs to be more tapered. Scale the bottom loop cut to 0.8 on X and Y."

"Now apply the bevel and add a weighted normal modifier. Auto-smooth at 45 degrees."

Lighting Iteration

"Add a point light above each table in the tavern. Warm color temperature — around 2700K. Intensity 800 lumens. Attenuation radius 4 meters."

"The lights are too uniform. Vary intensity between 600 and 1000 lumens randomly. Make one light near the back corner flicker — enable the flicker property with rate 0.3."

"The shadows are too hard. Increase source radius on all point lights to 0.15 meters."

"Add a single directional fill light from the window direction. Very low intensity — 100 lumens — just enough to keep the shadow areas from going pure black."

This conversational lighting workflow is particularly powerful because lighting is inherently iterative. You're constantly making small adjustments and evaluating the visual result. Reducing the mechanical overhead of each adjustment means you can try more variations in the same amount of time.

What AI Cannot Do: An Honest Assessment

Credibility requires honesty about limitations. Here's what AI tools — including MCP-based automation — cannot do, and what we don't expect them to do in the near future.

Final Artistic Judgment

AI has no taste. It cannot tell you whether your scene "feels right." It cannot judge whether the color palette evokes the emotion you're going for. It cannot identify when a composition is unbalanced in a way that serves the narrative versus when it's just wrong.

This is the core of what makes artists irreplaceable. Artistic judgment is a synthesis of aesthetic training, emotional intelligence, cultural context, and project-specific understanding that current AI systems fundamentally lack.

When you tell the AI to "make this look better," it has no meaningful way to fulfill that request. It can apply common techniques — increase contrast, add rim lighting, balance color temperature — but it doesn't understand "better" in the context of your specific creative intent.

Style Coherence Across a Project

One of the hardest problems in game art is maintaining a consistent visual style across thousands of assets created over months or years by multiple artists. AI tools can help with mechanical consistency — ensuring all assets use the same material naming convention, the same polygon budget, the same texture resolution. But they cannot enforce artistic style coherence.

They can't tell you that the new building model's proportions feel different from the existing buildings. They can't identify when a texture's hand-painted style doesn't match the semi-realistic style of adjacent textures. They can't flag when an environment's color usage drifts away from the established palette.

Style enforcement requires human eyes and human judgment. Always has, and currently still does.

Emotional Storytelling Through Environment

A good environment artist doesn't just place props. They tell stories. The empty chair at the set table tells a story. The child's toy next to the overturned wagon tells a story. The single flower growing through cracked concrete tells a story.

AI can place objects. It cannot understand the emotional weight of placement. When you tell it to "create a scene that feels abandoned," it can apply surface-level heuristics — overturned furniture, broken windows, vegetation overgrowth. But the specific choices that make a scene emotionally resonant rather than just technically "abandoned" require human creative intelligence.

Innovation and Surprise

AI tools work from patterns — either training data patterns (generative AI) or established operations (MCP automation). They are excellent at executing within established patterns. They are fundamentally limited at producing genuine creative innovation.

The decision to make a game's art style look like wet oil paint on parchment isn't something that emerges from AI assistance. The choice to use impossible architecture to convey a character's mental state isn't something AI suggests. Creative leaps come from human imagination. AI can help execute them once they're defined.

Contextual Understanding

AI doesn't understand your game. It doesn't know that this is a horror game where every shadow matters. It doesn't know that the player is supposed to feel safe in this area before the betrayal in Act 2. It doesn't know that the client specifically asked for "corporate but not soulless."

You can provide context through conversation, and modern AI assistants are good at working within stated constraints. But they're applying rules you give them, not understanding the underlying creative reasoning. When an unexpected edge case arises — and they always do — the AI doesn't have the deep project understanding to make the right call autonomously.

Building a Human-in-the-Loop Workflow

Given what AI can and cannot do, here's how to structure a workflow that maximizes AI productivity benefits while maintaining full artistic control.

The Three-Phase Model

Phase 1: Human defines creative intent. Before touching any AI tools, the artist establishes the creative direction. This might be concept art, reference boards, written descriptions, rough blockouts, or just a clear mental image. The key: the creative direction exists before AI gets involved.

Phase 2: AI executes under direction. Using MCP tools, the artist directs the AI to execute their creative intent. Scene assembly, batch operations, iterative refinement — all under conversational control with the artist evaluating results at each step.

Phase 3: Human reviews, refines, and finalizes. The artist reviews all AI-executed work with fresh eyes. This is where artistic judgment applies final polish: adjusting the specific prop placement that tells the story, tweaking the material that doesn't quite feel right, catching the inconsistency that the AI couldn't recognize.

Review Checkpoints

Build explicit review points into your AI-assisted workflow:

After initial placement: Does the overall layout match my creative intent? Is the spatial composition working?

After material assignment: Do the materials read correctly at gameplay camera distance? Is there sufficient variation without breaking consistency?

After lighting setup: Does the mood match the intended atmosphere? Are the light-dark patterns guiding the player's eye correctly?

After optimization: Did the optimization passes change anything visually significant? (Sometimes LOD reduction or instance culling affects the art.)

Final art pass: The artist does a manual final pass on everything. This is non-negotiable. No AI-touched asset ships without human review.

The 80/20 Split

In our experience, the ideal split is roughly:

80% of time: creative decisions, evaluation, and refinement. This is the valuable work. This is why you have artists.
20% of time: mechanical execution. This is the work AI handles.

Without AI tools, the split is often reversed: 20-30% creative work, 70-80% mechanical execution. Flipping that ratio doesn't reduce the artist's role. It amplifies it.

Practical Integration with StraySpark Tools

Here's how this philosophy plays out across our tool lineup.

Blender MCP Server: The Artist's Automation Layer

The Blender MCP Server with its 212 tools across 22 categories is designed as an automation layer for artists, not a replacement for them. Typical artist-controlled workflows:

Batch mesh operations: Apply modifiers, clean up topology, standardize settings across asset collections
Material setup: Create and assign material templates, batch-adjust properties, set up texture nodes
Scene management: Organize collections, rename objects, configure export settings
Iteration support: Rapid parameter adjustment during the creative refinement loop

The artist maintains full creative control at every step. The MCP Server just makes each step faster.

Unreal MCP Server: Production Pipeline Automation

The Unreal MCP Server serves a similar role in the engine. For environment artists working in UE5:

Level dressing: Place and configure actors through natural language rather than manual drag-and-drop
Material instance creation: Batch-create material instances with varied parameters for asset variation
Lighting setup: Position and configure lights conversationally, iterate on atmosphere quickly
Quality assurance: Check for common issues — overlapping meshes, missing collisions, incorrect LOD settings

Procedural Placement Tool: Rule-Based with Artist Override

The Procedural Placement Tool embodies the human-in-the-loop philosophy directly. It scatters objects based on rules (density, slope, layer masks) but gives the artist per-instance override control. You can procedurally place 10,000 trees and then manually adjust the 50 that matter most.

This is the correct model for AI-artist collaboration: automated systems for the bulk work, manual control for the meaningful work.

Practical Workflow 4: Asset Optimization and Technical Prep

Beyond creative workflows, there's an entire category of technical preparation work that artists spend hours on and that benefits enormously from MCP automation. This is work that has zero creative content — it's pure technical compliance — and it's a perfect candidate for AI handling.

Texture Resolution Standardization

You: "Check all texture assets in the current project. Flag any textures that aren't power-of-two resolution. For each flagged texture, resize to the nearest power-of-two that doesn't lose more than 10% of the original resolution. Report which textures were resized and from what dimensions."

Polygon Budget Enforcement

You: "Check all meshes in the scene against these polygon budgets: hero props under 10,000 triangles, standard props under 5,000, background props under 2,000, foliage under 3,000. List every mesh that exceeds its category budget with the current count and the target."

The artist reviews the list and makes decisions: does this asset genuinely need more triangles (artistic override), or should it be optimized (apply decimation)? The AI does the auditing; the artist makes the judgment calls.

Scene Cleanup Before Handoff

When handing a scene file to another department (from environment art to lighting, for example), cleanliness matters.

You: "Clean up the current scene for handoff to lighting. Delete all hidden objects that aren't in the 'Reference' collection. Purge all orphaned data blocks — unused materials, textures, and meshes. Verify all objects have applied transforms. Rename any objects still using default Blender names (Cube, Sphere, etc.) based on their collection and position. Generate a scene manifest listing all objects, their collections, material assignments, and polygon counts."

This kind of technical housekeeping takes 30-60 minutes manually and is universally dreaded by artists. Through MCP, it's a 2-minute automated process. The artist's time is freed for work that actually uses their artistic skills.

Addressing Industry Concerns Directly

"AI Will Reduce Team Sizes"

Maybe. But good tools have always changed team structures. Photoshop reduced the number of artists needed for some tasks. Game engines reduced the number of programmers needed for rendering. The question isn't whether tools change workflows — they always do — but whether they make the end product better.

Our bet: studios that use AI tools to make their existing artists more productive will produce better games than studios that use AI tools to fire artists and accept lower quality. The market will reward quality. It always has, eventually.

"AI-Generated Content Is Generic"

When used as a final output, yes. AI-generated models, textures, and layouts tend toward the median of their training data. That's a real limitation.

But MCP-based workflows don't generate content from training data. They execute artist-directed operations. The output is exactly as generic or unique as the artist's creative direction. A talented artist using MCP automation produces distinctive work. The tool doesn't constrain the artist's voice; it amplifies their productivity.

"Clients Will Expect Faster Delivery at Lower Budgets"

This is a legitimate business concern. When tools increase productivity, there's pressure to pass those gains to clients as lower prices rather than better quality.

This is a negotiation issue, not a technology issue. Artists and studios need to articulate the value of quality work and negotiate accordingly. AI tools don't change the fundamental economics of "good work takes skilled people and costs money." They change how skilled people spend their time.

"Junior Artists Won't Learn Fundamentals"

This concern has merit. If junior artists use AI automation for tasks they should be learning manually, they'll miss foundational skill development.

Our recommendation: juniors should learn the manual process first. Understand UV unwrapping by hand before automating it. Learn material node graphs by building them before having AI set them up. Use AI automation after you understand what it's automating. The tools are productivity multipliers — they multiply whatever skill level you bring to them.

The Future We're Building Toward

We believe the future of 3D art production looks like this:

Artists are directors. They make creative decisions, evaluate quality, maintain style coherence, and tell stories through their work. Their artistic skill and creative judgment are the irreplaceable core of the process.

AI tools are assistants. They handle mechanical execution, batch processing, repetitive setup, and routine operations. They operate within the boundaries the artist defines. They're fast, consistent, and tireless at mechanical work.

The collaboration is controlled. The artist initiates, directs, evaluates, and approves. The AI executes, suggests, and automates. The artist can always override, undo, or reject. The AI never acts autonomously on creative decisions.

Quality goes up, not down. Because artists spend less time on tedious work and more time on creative work, the overall quality of output improves. The same artist produces better work with good tools than without them.

This isn't idealistic projection. It's what we see in practice with our users today. Artists who use the Blender MCP Server and Unreal MCP Server report spending more of their day on the work they enjoy — creative decisions, artistic refinement, problem-solving — and less on the work they tolerate — clicking through menus, batch-processing files, setting up repetitive configurations.

That's what good tools do. They don't replace the craftsperson. They free them to do more of the work that matters.

Case Study: Environment Art Pipeline With and Without AI

Let's make this concrete with a real comparison. We tasked two approaches with the same goal: dress a medieval tavern interior with approximately 200 props — tables, chairs, tankards, food items, wall decorations, lighting fixtures, floor debris.

Without AI Assistance

The artist opened the level, loaded the prop library, and started placing. Each prop required: selecting the mesh, dragging it into the scene, positioning with transform gizmo, adjusting rotation for natural randomness, setting scale variation, checking collision, verifying material assignment.

Placement rate: approximately 15-20 props per hour for careful, quality-conscious placement. Some props (wall-mounted items, ceiling fixtures) required more precise positioning. Total time: approximately 12 hours of focused placement work.

The creative quality was excellent. Every prop placement was intentional. The artist's vision was fully realized. But two full working days on a single room is a significant time investment.

With MCP Assistance

The artist described the room in sections. "The bar area needs to feel busy — clustered tankards, stacked plates, a few spilled drinks. The fireplace area should feel cozy — armchairs, a rug, books on the mantle. The dining area is the functional middle — tables and chairs in a loose grid."

Placement rate: approximately 50-80 props per batch command, with review and adjustment after each batch. Total time: approximately 3 hours, including all review and manual adjustment passes.

The creative quality was nearly identical after the adjustment passes. The initial AI placement was about 70% of the way to the final result. The remaining 30% was artist-directed refinement: "move that tankard off-center on the table," "tilt the broom against the wall at a steeper angle," "the candle on table 3 should be melted down more — swap it for the short candle variant."

The Verdict

The final tavern interior was indistinguishable between the two approaches after the adjustment pass. The AI-assisted approach took 75% less time. But — and this is important — both approaches required the same artistic vision. The AI didn't decide what the tavern should look like. The artist did. The AI just placed the props faster.

The time saved was reinvested in additional polish: the artist added a second room that wouldn't have fit in the schedule otherwise, and spent extra time on the lighting, which improved the overall scene quality.

This is the core value proposition of AI-assisted creative workflows: same artistic quality, more artistic output, because the mechanical overhead is reduced.

Getting Started Without Losing Control

If you're a 3D artist considering AI tools but worried about losing creative control, here's a practical starting path:

Start with Automation, Not Generation

Don't start with text-to-3D tools. Start with automation tools like MCP that control the software you already know. The operations are familiar, the results are predictable, and you maintain full control. This builds confidence that AI assistance doesn't mean AI replacement.

Define Your Boundaries

Decide in advance which tasks you want AI to handle and which you do manually. A reasonable starting point:

AI handles: Naming, organization, batch operations, export prep, repetitive property assignments Artist handles: All modeling, texturing, lighting, layout decisions, and final quality review

Expand the AI's role as you get comfortable, but always maintain the boundary: creative decisions are yours.

Use Undo Liberally

MCP operations go through standard undo systems. If the AI does something you don't like, undo it. This simple safety net means there's no risk to trying AI-assisted workflows. The worst case is pressing Ctrl+Z.

Review Everything

Never ship AI-touched work without reviewing it. Not because the operations are unreliable — Blender's tools produce the same results whether you click the button or MCP invokes the command — but because review is where artistic judgment happens. The review step is where you catch "technically correct but artistically wrong."

Keep Learning

AI tools don't make manual skills obsolete. They make manual skills more valuable, because the person directing the AI needs to understand what they're asking for. Invest in your artistic fundamentals. Learn new techniques. Study composition, color theory, anatomy, architecture. The deeper your knowledge, the better your AI-assisted output will be.

The 52% of developers concerned about AI in games aren't wrong to be concerned. But the answer isn't to reject AI tools. It's to demand that AI tools serve artists rather than replace them. That's what we're building, and that's what the MCP model delivers: tools that artists command, not systems that command artists.