Is AI actually good enough to build real things?
Yes — with the right expectations. AI excels at:
- Scaffolding: Getting you from zero to a working draft in minutes
- Iteration: "Make it smaller, add a chamfer, change the color palette"
- Translation: Converting between formats, languages, mediums
- Research: Synthesizing information across domains
AI struggles with:
- Precision engineering: Tolerances, material science, safety-critical design
- Artistic nuance: It can imitate style but not create genuine emotional resonance
- Long-term consistency: Multi-day projects require careful context management
- Real-world physics: It approximates but doesn't simulate
The sweet spot: AI generates 70-80% of the work, you contribute judgment, taste, and domain knowledge for the remaining 20-30%. The combination is stronger than either alone.
Do I need expensive hardware?
Depends on the medium:
| Activity | Minimum Hardware | Ideal Setup |
|---|
| Coding | Any laptop from the last 5 years | 16GB RAM, SSD, decent display |
| 3D Printing | $200 printer (Bambu A1 Mini) | $750 (Prusa MK4) |
| AI Image Gen | 8GB VRAM GPU (RTX 3060) | 12GB+ VRAM (RTX 4070+) |
| Video Editing | 16GB RAM, dedicated GPU | 32GB RAM, fast SSD, decent GPU |
| Music Production | Any computer + headphones | Audio interface + monitors (~$300) |
| Writing/Notes | Literally anything | A good keyboard |
You can start every single category for under $500 if you already have a computer. GPU is the main bottleneck — if you don't have one, use cloud AI services (many have free tiers) instead of running locally.
What about running AI models locally vs. cloud?
Local (your computer):
- No ongoing cost after hardware purchase
- Complete privacy — your data stays on your machine
- No internet required
- Limited by your GPU
- Setup can be technically involved
Cloud (API or service):
- Works on any hardware
- Latest models immediately available
- Pay per use — can be cheaper for light use
- Requires internet connection
- Your data goes to third-party servers
Recommendation: Start with cloud (ChatGPT/Claude free tiers). Move to local once you know what you need and have the hardware. For image generation specifically, local is dramatically cheaper after the initial GPU investment.
How do I organize projects that span multiple mediums?
This is the real skill that separates productive builders from chaotic ones. Use a consistent structure:
~/projects/project-name/
├── README.md # What, why, how
├── DECISIONS.md # Key choices and reasoning
├── prompts/ # Prompts that worked
├── src/ # Source files (any medium)
├── output/ # Generated artifacts
├── reference/ # Inspiration, specs, research
└── .git/ # Version control everything
Key rules:
- One project, one folder — never scatter related files
- README in every project — future-you forgets fast
- Version control text — Git handles code, markdown, prompts, SVG
- Separate source from output — source is what you edit, output is what you generate
- Name files descriptively —
headphone-mount-v3-stronger-hook.stl not final_final_v2.stl
Which AI model should I use for what?
| Task | Best Model | Why |
|---|
| Code generation | Claude (Sonnet/Opus) | Best at understanding codebases and writing clean code |
| Code explanation | Claude or GPT-4 | Both excellent at teaching |
| 3D design (OpenSCAD) | Claude or GPT-4 | Solid parametric code generation |
| Image prompts | GPT-4 (for writing prompts to use in SD) | Good at descriptive language |
| Writing | Claude | Strongest prose quality |
| Research synthesis | GPT-4 or Claude | Both handle large context well |
| Quick questions | Gemini Flash / GPT-4o Mini | Fast, cheap, good enough |
Don't get locked into one model. They have different strengths and they improve monthly. The best builders use 2-3 models depending on the task.
Is AI-generated content copyrightable?
The legal landscape is evolving (as of 2026):
- Code you prompt and edit: Generally treated as your work in practice
- AI-generated images: Copyright status varies by jurisdiction. US Copyright Office requires meaningful human authorship
- AI-assisted work (human-directed, AI-augmented): Strongest copyright position
- Pure AI output (no human creative direction): Weakest position
Practical advice: Always add meaningful human creative input. Edit, curate, combine, and refine AI outputs. Keep records of your creative process (prompts, iterations, decisions).
What's the cheapest possible way to start making things with AI?
$0 budget:
- Sign up for ChatGPT free tier or Claude free tier
- Install VS Code (free), OpenSCAD (free), GIMP (free), Audacity (free)
- Use DaVinci Resolve free for video
- Use Obsidian for notes
- Use GitHub free for version control and storage
- If you have a GPU: install ComfyUI for image generation
You can code, design 3D models, edit audio, edit video, create images (if you have a GPU), take notes, and manage projects — all for exactly zero dollars.
$200 budget (adds physical making):
- Everything above + Bambu Lab A1 Mini ($200) + a roll of PLA ($20)
- You can now go from prompt to physical object