Designs you can talk to
in your own terms.
I build software and physical systems. I'm looking for people who think the next interface isn't chat or menus — it's designs you edit in your own vocabulary, with the machine holding the physics, not the property panel.
Shipping ops platforms today. Treating generative UI as a field to join, not a buzzword to ship.
What I'm exploring
There's a spectrum of how people should interact with complex tools. I'm interested in the middle band — but not the version that's “ChatGPT with a form attached.”
I want interfaces where:
- —You describe your design dimensions, not the tool vendor's — “2.5 bricks wide at the firebox corner,” not hunting for firebox.opening.width in a tree.
- —The system understands construction, structure, physics, and materials deeply enough that “make the wall longer” just works.
- —You iterate layer by layer — structure first, aesthetics second: “now increase the arch radius slightly,” “random river stone on the oven face.”
- —The same conversation moves across what are separate apps today — spatial design, structural reasoning, rendering, docs — without forcing your idea through each editor's primitives.
Where this came from
The masonry heater
I recently finished a masonry heater — aircrete shell, jutt-brick walls, bell chamber, firebox, the whole stack. Real build, real corner of a real room.
Before I touched brick, I tried to design it with AI image tools: a photo of the empty corner, then incremental changes. It was miserable. Each generation was a one-shot with no stable model to edit. Dimensions drifted. I couldn't say “aircrete walls, then jutt brick perpendicular, leave a ~2-brick gap at the inner corner for the firebox — actually make that 2.5” and have the next step inherit the last.
I knew a proper 3D modeler would help, but learning a discipline-specific tool for one project wasn't worth the ramp — even though parametric editing was exactly what I needed. So I built it anyway, mostly from head and hand. The heater exists. The design conversation I wanted doesn't.
That gap is what pulled me into generative UI as a field to join, not a buzzword to ship. I want sliders for the aspects I specify, backed by domain knowledge broad enough to edit across disciplines without making me become five kinds of expert first.
The bet
The winners in the middle band won't be “Photoshop but faster.” They'll let the user name the editable dimensions, unify the domain model, make edits semantic, and get structure right before beauty.
Level 1
AI generates a one-off screen — a viz to explain relativity. Useful. Not enough.
Level 2
Familiar base UI plus adaptive regions driven by rules and a component catalog. Chesky scrapped Airbnb's chat booking; this is the direction that survived. This is where I live.
Level 3
The product is the rules. Rare, high stakes.
Why this isn't armchair
I've shipped the platform, built the heater, and hit the wall that motivates the product. Each of these is one face of the same problem: how do parts of a design relate when one thing changes?
Replaced spreadsheet hell at Transparent Digital Services with extensible client/domain/GTM management, RBAC, audit logging, real-time updates. Next.js, Supabase, TypeScript.
Platform variables: the business defines its own fields and validation. The closest thing I've shipped to user-named dimensions of control instead of a fixed form.
React/Redux overlay editor for broadcast venues; fixed multi-element drag perf by inverting the store mapping. First deliverable was the iframe embed server over postMessage.
Relationship-aware updates when one piece moves — and untrusted UI across a postMessage boundary, the same family as MCP apps' double iframe.
Designed and laid an aircrete/brick thermal-mass heater in a real room corner — firebox, bell, arch — after AI image tools failed to hold a parametric conversation.
Layer-by-layer semantic edits across thermal, structural, and aesthetic concerns. The domain no single app owns. This is the product I wish existed.
A loud, deeply customizable floating toolbar built around my own mobile workflow — buttons compose tap / long-press / swipe commands from primitives, with saveable config slots.
Composing UI from primitives you control versus the UI a host forces on you.
Co-editing credibility: I ran my own CouchDB replication for Obsidian — PouchDB on each device syncing to my own server — because the product couldn't move the files I cared about. I know what you're paying for when sync works, and what breaks when two writers touch the same thing without a merge story. Co-editing is that, plus an agent writer and generated UI as a projection.
The physical side isn't a sidebar
Yurt, two campers, a masonry heater, PCB work. The generative UI I want isn't for people who only live in Figma — it's for people who build things that have to stand up, heat correctly, and pass inspection. Every one of these builds wanted the same tool I didn't have:
About
Full-stack engineer and builder across software and physical systems — they/them. Home Depot (React/Go, then the UI-library hydration battles that taught me what fake abstraction costs), Meta (front-end, Proton team), Airbnb and Peloton, then freelance work with Tash: Lupe, LoveLace, and more.
I was a college tutor and mentored interns on React architecture. I like finding the gap in a conversation and helping someone see light — which is the social shape of this work: conversations about the spectrum, not solo manifestos.
What I want next is paid work on the middle band — internal tools, ops platforms, creative and agent tooling where the job is “menu/CAD/spreadsheet hell → a system with the right mix of fixed UI and generated regions.” Teams I'm watching: Zoo/KittyCAD, 3E8 Robotics, and the agent layers going onto Figma, Canva, and Autodesk.
Need a freelancer for something more down-to-earth right now? Here's the freelance version of this site.
Having this conversation?
I'm looking for builders, designers, and researchers working on generative UI, agent harnesses, component catalogs, and cross-domain design models — internal tools, creative tooling, agent platforms. Paid work first; good conversations a close second.
“I built a masonry heater and tried to design it with AI images first. It couldn't hold a parametric conversation — I had to one-shot everything. That's the product I wish existed, and I'm looking for the people building toward it.”