We have a home music studio. It has mirrored walls, heavy velvet curtains, a sump pump riser in the corner, an arched doorway, and a family that plays everything from progressive metal to classic soul. It also had 116 acoustic panels, 4 ceiling panels, and 12 corner bass traps sitting in a pile, waiting for someone to figure out where they go.

Acoustic treatment planning isn't rocket science, but it's not nothing either — first reflection points, bass trap positioning, working around architectural features, calculating panel placement across four distinct walls. For a non-professional, it felt like the kind of project that would either take weeks of research or require hiring someone. I gave it an afternoon with Claude instead.

What I actually did

I used two tools. ChatGPT handled the initial research phase — identifying materials, evaluating options, establishing the supply list. Claude handled the design and planning phase, which is where the spatial work lived.

I gave Claude six photos of the space, room dimensions (149"W × 128"D × 93"H), a full gear inventory, the supplies I already had on hand, and my mounting constraints. What I got back was a scaled SVG floor plan and wall elevations — one artboard per surface, with all fixed elements accurately placed — plus an acoustic treatment strategy and a materials inventory with sourced product links.

That's the clean version. The real version looked like this:

  • ~12 prompts answering clarifying questions about the room
  • ~8 prompts providing corrections and refinements to placement
  • ~6 prompts requesting unit conversions (inches to pixels for Illustrator handoff)
  • ~4 prompts requesting product dimension lookups
  • ~4 prompts requesting document deliverables
  • ~2 prompts sharing photos

36 prompts total. Not magic — a real working session.

What I fixed myself

Claude's SVG output was a strong starting point, not a finished product. In Illustrator I cleaned up label placement where text overlapped, refined acoustic panel positioning, corrected bass trap and sump pump cover dimensions, and converted the arched doorway from two shapes to one. I also placed the decibel meter — Claude didn't think to include it.

This is the part that gets lost in most AI demos: the human doing the finishing work isn't a workaround. It's the job. I brought spatial thinking, design instincts, Illustrator skills, and a clear vision of what the room needed to feel like. Claude brought the math, the drafting, and the patience to iterate.

What I learned

I would do this again without hesitation. A few things that made it work:

  • Start with a clear problem statement and be prepared to answer a lot of clarifying questions. The more specific your inputs, the more accurate the outputs. This isn't a limitation — it's the process. Expect to refine across multiple prompts rather than get it right in one.
  • Know what you're bringing. I had spatial reasoning, design skills, and Illustrator experience. Someone without those would have gotten a different result from the same prompts. AI-assisted work is still human work — the human's capability shapes the ceiling.
  • Choose your tools deliberately. I used ChatGPT for research and Claude for design iteration. They're not interchangeable. Knowing which tool fits which phase of a project is its own skill.

The output

Six wall elevations, a ceiling plan, a floor plan, and a materials table that let me identify exactly what mounting supplies I still needed. Total time: six hours including this writeup — and lunch, and supervising other family members working on other parts of the project.