Skip to main content

Graphics Programming Plus AI

5 min read
Low Level

Low Level

AI can draft shaders and suggest optimizations. Performance and visual correctness are yours to verify.

Graphics Programming Plus AI

TL;DR

  • AI can draft shaders, suggest rendering techniques, and help with optimization. It will also produce code that's slow, wrong, or platform-incompatible.
  • Use AI for inspiration and first passes. You own performance, correctness, and cross-platform behavior.
  • Graphics is highly platform-specific. AI doesn't know your GPU, your engine, or your constraints.

Graphics programming sits at the intersection of math, hardware, and art. AI has seen a lot of shader code and rendering docs. It can generate GLSL, HLSL, and explain techniques. It will also give you code that runs at 5 FPS or produces visual artifacts. Your job: use AI to explore, then lock down with your expertise.

What AI Can Help With

Shader drafting:

  • "Implement a basic Phong shader" or "Add normal mapping." AI can scaffold. You tune for your pipeline.
  • Good for learning and iteration. Always profile. Always check output.

Technique explanation:

  • "How does PBR work?" "What's a shadow map?" AI can summarize papers and common implementations.
  • Use as a starting point. Verify against authoritative sources.

Optimization ideas:

  • "How can I make this faster?" AI might suggest LOD, batching, or different algorithms.
  • Test. Graphics performance is counterintuitive. What works on one GPU can fail on another.

Porting and translation:

  • GLSL to HLSL, or vice versa. AI can help. You verify semantics and extensions.
  • Shader dialects differ. AI mixes them. Check.

What AI Gets Wrong

Platform specifics:

  • Mobile vs. desktop. Vulkan vs. OpenGL vs. Metal. Driver quirks. AI doesn't know your target.
  • You do. Validate on real hardware.

Performance:

  • "This should be faster" — maybe. Branching, texture fetches, register pressure. AI doesn't profile your scene.
  • Benchmark. Always. On your target hardware.

Visual correctness:

  • Gamma, color space, precision. AI can produce code that "looks" right in a screenshot and breaks in edge cases.
  • You have the eye. You know the spec.

Math and physics:

  • AI can implement formulas. It can also get them wrong. Especially in edge cases or when combining techniques.
  • Verify the math. Compare to references.

The Workflow

  1. Generate — Use AI for a first pass. Shader, technique, or optimization idea.
  2. Profile — Does it run? At what cost? On what hardware?
  3. Correct — Fix platform issues, precision, artifacts. Your standards.
  4. Integrate — Into your engine, your pipeline. AI doesn't know your stack. You do.

Your Moat

  • Performance intuition. You know when a technique will work and when it won't. You've been burned. AI hasn't.
  • Cross-platform mastery. You know the differences. You know the workarounds. That's years of experience.
  • Aesthetic judgment. "Looks good" vs. "looks right" — you have the eye. AI has averages.

Manual process. Repetitive tasks. Limited scale.

Click "With AI" to see the difference →

Quick Check

What remains human when AI automates more of this role?

Do This Next

  1. Generate one shader with AI — A simple effect. Profile it. List what you had to fix. That's your QA process.
  2. Document your target constraints — Hardware, engine, budget. When you use AI, include these in your prompt. Better context = better output.
  3. Build a reference bench — A scene or test that represents your typical load. Use it to validate every optimization. AI or not.