Building AI-to-Physical Products

The gap

AI generates content easily. Claude can write a project status update in seconds and ChatGPT can summarise a meeting transcript instantly. The content is there.

But turning that content into something professional, something you can print, present, hand to a client, is still tedious. You copy from Claude, paste into PowerPoint, fight with formatting, adjust margins, fix overflow, wonder why the columns don't balance.

That gap is where I build.

AI generates content easily. Turning it into professional physical output is hard. That's the gap.

The model

Every product in this portfolio follows the same model:

User generates content      →    Tool processes          →    Physical artifact
(in Claude/ChatGPT)              (deterministically)          (slides, PDF, etc.)
     ↓                                  ↓                           ↓
  They pay for AI               You pay nothing              You charge for value
  (or free tier)                (no AI costs)                (convenience, quality)

I call this BYOAI (Bring Your Own AI). The user already has Claude or ChatGPT, so I don't need to bundle AI and charge for it. I just handle the formatting.

Why this works

Firstly, there are no AI costs on my side. I'm not calling OpenAI or Anthropic APIs. The user's AI generates the content and my tool just formats it, so margins stay high.

Secondly, the output is deterministic. Same input equals same output, every time. No prompt engineering, no model drift, no "why did it do that?" debugging. I think that predictability is underrated in a world where everyone's chasing generative everything.

And the thing that ties it together is that physical correctness is genuinely hard. Fold lines that actually align when printed, columns that balance automatically, PPTX files that are editable rather than images. This is the kind of thing AI tools do poorly, and the kind of thing traditional tooling does well.

The portfolio

TriFold PDF was the first product. You paste AI content and get a printable tri-fold reference card with proper fold alignment. One-time purchase, and it validated the model for me.

DeckEngine is the current focus. It takes markdown and produces editable PowerPoint with automatic two-column balancing. The layout engine uses Typst, which gives me 40 years of typesetting algorithms that no competitor can easily replicate.

I'm also exploring church bulletins (52 times a year recurrence) and name badges (Avery sheet layout math). Same model: AI content plus physical formatting equals value.

The criteria

Not every idea fits. Before I build anything, I think about four things. Does it have a recurring need? Monthly reports beat one-time spice jar labels. Is there a B2B or prosumer buyer? Businesses have budget, while consumers are price-sensitive. Is there a technical moat? Column balancing is hard, but text on rectangles is easy. And can I ship an MVP in two weeks? If I can't validate it fast, it's probably not worth the investment.

The goal isn't one big product. It's a portfolio of small, focused tools that each do one thing well and share infrastructure underneath.

If you're building something similar, or have thoughts on this approach, I'd like to hear from you.