ProbateDesk is software that automates probate paperwork in British Columbia. I built it when I was 14. A law firm bought it. It's still running.
That sentence is on my homepage and it's the single hardest thing on there to check, because the product carries no founder credit anywhere. I sold it, so I can't add one. This page is the record.
What probate paperwork actually is
When someone dies, their estate usually has to go through probate — a court process that confirms the will is valid and gives the executor legal authority to deal with the assets. In BC that means a set of Supreme Court forms, filed in a particular order, with figures that have to agree across every document in the package.
The work is not intellectually hard. It's exacting and repetitive, and the failure mode is brutal: a number that doesn't reconcile between two forms, or a name spelled two different ways, gets the whole filing sent back. Then you wait weeks and do it again. Firms handle this by having a person check the package by hand, several times, which is expensive and still doesn't catch everything.
So: rules that are written down publicly, structured input, structured output, high cost of small errors, and everyone doing it hates doing it. That's about as close to an ideal target for software as you get.
The bottleneck was never the forms. It was reading a shoebox of documents and getting every detail right.
Why 14, and not 12
I'd already tried legal tech two years earlier and stopped. I'd shelved it at 12 because the models couldn't be trusted to read a document and extract the details accurately, and in probate a detail that's nearly right is wrong. Instead of forcing it, I wrote down what would have to change and left it.
What changed by 14 was document reading. Models got good enough to pull structured facts out of unstructured paperwork reliably enough that the remaining errors could be caught by checks rather than by hoping. That was the one condition on the list. So I opened the notes and built the thing.
I want to be clear about what that means, because "I waited for the tech" sounds like luck. It partly is. But the reason I could move fast in the window was that I'd done the domain work two years before and had it written down. The waiting was the work.
What it did
The short version: you gave it the raw material for an estate, and it gave you back a filing package that was internally consistent.
The part I'd defend as the actual product wasn't the form-filling. Anyone can fill a form. It was the consistency layer — the checks that made sure the same fact appeared the same way everywhere it appeared, and that flagged the places where it didn't. Filling forms saves time. Catching the mismatch before it reaches the registry saves weeks.
What broke
The general shape of the problem, which I'd expect anyone building in this area to hit: real documents are much worse than the examples you test on. They're scanned crooked, they use inconsistent naming, and they contain edge cases that the rules technically cover but nobody writes down. Every hour I spent on the happy path was worth less than an hour spent on a document that didn't parse.
The other thing I got wrong early was trusting a model output because it looked right. Formatted, confident, plausible output is the most dangerous kind in a domain where the reader can't easily verify it. Most of what I ended up building was the machinery for not trusting it.
The sale
A law firm bought it. That's the part people react to, and honestly the least interesting part to me — the software was the work, the sale was a conversation.
What I'd tell anyone in the same position: firms don't buy software because a demo is impressive. They buy it because it removes a specific cost they can name. Every conversation that went anywhere started with them describing the hours, not me describing the product.
It's now owned and run by ProbateDesk Technologies Inc. and it's still in production. I don't have any ongoing involvement.
What I'd do differently
I'd have shipped something narrower, sooner. I spent too long trying to cover the whole filing package before putting any of it in front of someone who does the work for a living. The first real piece of feedback moved the product more than the previous month of building had.
I'd also have kept better records. Building at 14 in the gaps around school, I didn't write down the decisions as I made them, and reconstructing them now for a page like this is harder than it should be. That's part of why the now page exists.
One thing this page is not
This is a build retrospective, not legal advice. Nothing here tells you how to file for probate in BC, and you shouldn't treat it as guidance for your own estate matter. If you have one, talk to a lawyer.
Where this went next
ProbateDesk is the reason I take timing seriously. It's a fairly plain business application that happened to sit exactly on the line where a capability had just arrived — and the whole return came from having been early to the problem and late to the build.
That's the same bet I'm making now. Browser agents can already do real work unsupervised. Almost nothing sends them work at the moment a person actually decides to do something. That's Anticipy.
If you've built in this area — legal tech, document extraction, anything where the cost of being confidently wrong is high — I'd like to hear from you: omar@anticipationlabs.com