Gone Live — On Building Something From Nothing in 2026

It Exists Now

lamina.vet is live. Anyone with an internet connection can open it. A vet can create an account, start a consultation, record the conversation, watch clinical facts appear on screen in real time, and generate a SOAP note — a discharge summary, a prescription, a referral letter — in seconds. A pet owner gets a link they can open on their phone, no login required, and chat with an AI that knows exactly what happened at their animal's visit.

Before I started building it, none of this existed. There was no database. No authentication. No recordings. No documents. No companion. Nothing. Just an idea and a blank repository.

Now it's out there, in the ecosystem called the internet, just sitting there, real.

I find this genuinely hard to get my head around. The product is ones and zeros. It lives on servers I don't physically own, served over infrastructure I didn't design, displayed on screens I've never seen. And yet it's as real as anything. A vet anywhere in the world could open it right now. Something that didn't exist before I started building it now just — exists.

I'm still not sure what to do with that feeling.

What lamina.vet Is

The name is "animal" spelled backwards. I appreciate that.

The product is an AI-powered assistant for veterinary consultations. The core flow is simple: a vet records the consultation — the history, the exam, the assessment, the plan — and the system does the rest. Clinical facts are extracted in real time as the audio streams in. When the recording ends, the vet reviews and confirms the facts, adds their own diagnosis and treatment notes, and generates whatever documents they need: SOAP notes, discharge instructions, prescriptions, referral letters, follow-up plans, client summaries, lab requests.

What would have taken fifteen minutes of post-consultation admin takes seconds.

Then there's the Owner Companion — a shareable link the vet sends to the pet owner after the visit. The owner opens it on their phone and can ask questions: what does this medication do, when exactly should I bring her back, what warning signs should I watch for? The AI knows the visit. It answers from the consultation's facts. No login. No app to download. Just a link.

One person built all of that. It still surprises me when I say it out loud.

The Pieces

A product like this is really a composition of systems — the application layer, real-time data, authentication and organisation management, the AI layer itself, billing, deployment. Each of those is a deep domain in its own right. Each one has entire careers built around understanding it fully.

I am not an expert in any of them.

And that turns out not to matter as much as you'd think.

The Conductor

The best model I've found for how this works — and I'm still not sure it's right — is the conductor. Not the virtuoso.

A conductor doesn't play every instrument. They don't need to know every technique or have memorised every passage. What they need is a clear picture of what the whole thing should sound like — and enough musical understanding to hear when it doesn't.

That's roughly how I work, or try to. I hold onto a clear picture of what I want and understand enough of how each system works to see when something isn't right — to ask the right question, read the error, find the path. I test constantly: not just automated tests, but real manual testing, clicking through the actual flows, feeling whether things work in the way a vet would actually feel them. When I can't figure something out myself, I direct AI to find it. Sometimes that works on the first try. Often it doesn't.

For me, at least, the bottleneck is rarely technical knowledge. It's almost always been clarity about what I'm trying to build.

What AI Changed

I couldn't have built this without AI. Not in any reasonable timeframe, anyway. That's worth being honest about.

The combination that makes zero-to-one possible in 2026 isn't AI alone. It's AI plus a builder who knows what they want and has enough experience to evaluate what AI produces. AI without direction produces technically correct things that don't fit together. Direction without AI produces the right vision but can't move fast enough to make it real.

The way I use AI is less like asking an expert and more like working with a very fast, very capable collaborator who needs clear briefs and honest feedback. It can build the thing, but I have to know the thing needs to exist. It can wire up the integration, but I have to understand what that integration is supposed to do. It can generate the output, but I have to test it and feel whether it's right.

The judgement layer is still very much mine — and honestly, that's where most of the work happens. Everything else moves at a pace I wouldn't have thought possible a few years ago.

The Foresight Problem

There's something I can't quite articulate about the accumulated experience required for this to work. It's not technical expertise in any specific domain. It's more like foresight — the ability to see around corners.

While building lamina.vet, I'd often have a hunch about where the hard part would be before I'd written a line of code. Not because I'd solved the exact problem before, but because I'd seen enough systems to sometimes recognise the shape of the difficulty in advance. The feature that looked straightforward from the outside would have one specific seam that needed careful thought. Sometimes I could see it coming. Sometimes I couldn't, and it found me anyway.

I don't think you can shortcut this. Every project I'd worked on before quietly fed into this one — not in any specific technical way, but in the accumulated sense of having been wrong enough times to occasionally recognise what wrong looks like before it arrives. I'm still getting it wrong plenty. But less blindly than before, I think.

The Hard Parts

I want to be honest about this, because the version of this story where I describe the stack and the architecture and the live product makes it sound clean. It wasn't clean.

There were stretches where things didn't work and I couldn't immediately see why. Where something I'd wired together returned unexpected results and I had to rethink my assumptions from scratch. Where a subtle bug only surfaced at a specific sequence of actions I hadn't thought to test. Where I iterated many, many times before something finally felt right.

There was frustration. Real frustration — the kind where you've been looking at the same problem for hours and you're not sure if you're missing something obvious or if the system is genuinely broken. The kind where you go to bed not knowing if the thing you're building is going to work.

I think this matters to say, because the frustration is part of what made the eventual arrival worth something.

The Appreciation Loop

I wrote in another essay about how appreciation is earned — how difficulty creates the contrast that lets ordinary things feel extraordinary. I wrote about my first apartment without a dishwasher, and how that absence built a library of appreciation I still draw from.

I was writing that essay while building this product. And I realised, somewhere in the middle of it, that I was living the argument.

Because I've gone through the difficulty of building lamina.vet — the late nights, the iterations, the frustrations, the moments where something I thought was done turned out not to be — I have a deep appreciation for what it is now. Not in a performed, look-what-I-built way. Something quieter than that. When I watch the recording flow work — audio streaming in, facts appearing in real time, the vet reviewing and generating a document in seconds — I feel something. The appreciation is specific and earned. It's the appreciation of someone who knows exactly how much work is underneath that simple interface.

That wouldn't be there without the hard parts. The difficulty is what made it matter.

Ones and Zeros

It keeps coming back to this, and I keep finding it strange.

At the bottom of everything — every UI element, every database record, every streamed audio chunk, every generated document — it's just binary. Ones and zeros, encoded on physical media, transmitted over physical cables, interpreted by layers of abstraction built by thousands of people over decades. lamina.vet is an arrangement of ones and zeros that didn't exist before I created it.

And now it's out there. It lives in the ecosystem we built and called the internet. A vet in another country could open it right now and start using it. The Owner Companion link could be sitting in someone's WhatsApp, a pet owner asking about their cat's medications, getting answers from an AI that knows the visit.

That came from nothing. I don't have a tidy way to conclude that thought. It just seems worth sitting with — the fact that software lets you create something real from nothing, that the internet gives it a place to live and anyone to reach. That in 2026 one person, with enough clarity and enough accumulated experience, can actually do this.

I'm not sure I fully expected it to work.

Unanswered Questions

• Is the conductor model — high-level vision, AI execution, judgment layer — available to everyone, or does it require a specific kind of prior experience to work?
• What happens to the value of deep technical expertise in a world where broad directional understanding is often enough?
• Is there something lost when building is this fast? Does the speed undermine the appreciation that difficulty would otherwise create?
• What does "validation" actually mean for a product — when do you know that what you've built is genuinely useful, not just technically complete?
• Can you build something meaningful alone, or does the absence of collaborators create blind spots that only another person can surface?