A Lawyer Just Beat 500 Developers at Anthropic's Hackathon
The most valuable skill in AI right now isn't code, and you likely already have it.
The most valuable thing in AI right now isn’t code, an Nvidia GPU, or a computer science degree. It’s something you probably already have — and most people building AI products have no idea it even exists.
A lawyer, a cardiologist, and a road technician from Uganda. None of them has a software background. All of them built successful AI products liked by thousands in under a week.
If they can do it, so can you.
And I mean that literally — because the thing that made all three of them win is the same thing you’re probably sitting on right now. You just don’t know it yet.
Let’s rewind a little…
Last week, Anthropic — the company behind Claude — ran a competition. They wanted to see what people would build with their latest model in one week.
Thirteen thousand people applied to get in. Five hundred made it.
Most of them were developers. Engineers. People who’d spent years writing code. Those who you picture winning an AI hackathon. It was basically an unwritten rule: you don’t show up to these things unless you can code.
That rule has been cracking since people started vibe coding in 2023. By 2026, it's finally broken, for good.
Learning 1: Boring Means No Competition.
Mike Brown is an attorney in California. Not a software developer. Not a startup founder. A lawyer.
His friend builds backyard cottages and garage conversions — the kind of small homes that are supposed to help ease California’s housing shortage (yes, this issue is so well known there’s even a wiki page for that).
Every time this friend of his submits a permit application, it comes back rejected.
Not because the plans are wrong. Because one code citation is slightly off, or a local rule overrides a state rule in a way nobody documented clearly. The friend fixes it. Resubmits. Gets rejected again, for something completely different.
Months go by. The home doesn’t get built. The family doesn’t get housed because everything gets stuck in the bureaucratic process.
In San Francisco alone, the median time to get building approval is 627 days — and permit fees can add up to $74,000 per unit. Half of all projects are at risk of losing their financing before the first wall goes up.
So Mike built CrossBeam, he started the demo by clarifying the problem he’s solving:
Everyone thinks California has a housing crisis. We don’t. We have a permit crisis.
This app lets an owner upload their rejection letter and plans. And in 20 minutes, they’d get a precise, code-referenced action plan. Knowing exactly what to fix, exactly which regulation it references.
📺 Watch the original demo — this line lands harder when you hear it directly from him:
Connor Traut, Mayor of Buena Park testified:
I need over 3,000 new homes permitted by 2029. Last year we had under a hundred. We need this software.
A process that used to take weeks now takes a lunch break.
Built in six days. First place, here’s the link to his GitHub.
Yes, the permit office is one of the most unglamorous places on earth.
Unless you are in the trench, the kind of industries where real people are trapped in cycles of tedious manual misery. With few VCs funding this, no AI startup is interviewing permit applicants. And because nobody in AI is looking there, there is no competition.
That is the pattern.
Knowing an unglamorous industry was only part of Mike's edge. To see the next principle at its sharpest, meet someone who'd been standing inside the problem for a decade.
Learning 2: You Don’t Need User Research If You Are The User.
Michał Nedoszytko is a cardiologist in Brussels. His day job is to physically examine patients’ hearts to fix them.
Every single time, after they’ve just been told something is wrong with their heart. The patients feel frightened and confused. And within 15 minutes — before they’ve even reached their car — they’ve already lost the context.
Multiple studies show that patients can only recall about 49% of what their doctor tells them without prompting, and of what they do remember, around 15% is either incorrect or incomplete.
Michał has watched this happen thousands of times firsthand.
He knows which questions patients call back with the next day. He knows which parts of the discharge instructions nobody reads. He knows the gap between what gets said in the clinic and what patients can actually act on at home at 2am.
No product team could have known that. You cannot interview your way into ten years of bedside observation.
He heard about the hackathon on his morning commute. A week later he had a working product.
Postvisit.ai takes a patient’s clinical record and turns it into a personal AI companion — plain-language explanations, translated medical terms, follow-up answers, and tracking whether what’s happening at home matches what the doctor expected.
📺 Demo video:
His competitive advantage wasn’t technical skill, but proximity.
The best AI products right now aren't being built by the best coders; rather, they are being built by people like this doctor, who have personally watched the same painful (and literally life-and-death) gap play out for years.
But there's one last question: what kind of work is actually ready to be transformed?
The answer is hiding inside every “expert” job that, at the surface, looks impossible to automate.
Learning 3: A Task That Is Really Just Information Processing.
Kyeyune Kazibwe is a road technician in Uganda.
His job: drive roads, assess damage, estimate repair costs, and decide where money should go. It’s slow and physically exhausting. And it can’t be scaled.
There are far more roads that need assessing than there are technicians to cover the ground. Schools, markets, and clinics wait while the paperwork catches up.
In his own words:
Infrastructure is expensive. We need to make sure every dollar spent goes directly to a project which will have as big an impact as possible… We don’t have enough data and experts to do analysis for all the projects.
So he built TARA. You drive a road with a dashcam. You upload the footage.
In five hours — versus the weeks a traditional assessment takes — you get a complete report: surface condition by segment, roughness index, repair cost estimates, and an equity assessment identifying which communities actually benefit from the fix.
In his demo, he tested on a real road under construction near Kampala.
📺 Full demo with narration — the Uganda road footage is striking visual content:
Here is the principle that unlocks everything: strip away the jargon from any “expert” job and describe what’s actually happening, step by step.
What is Kyeyune doing? He looks at a surface. Compares it to a known standard, then assigns a number based on the guideline. That’s pattern recognition and standard information processing.
Once you can see a system or a workflow that way, you see it everywhere.
The insurance adjuster eyeballing a claim, the compliance officer reading a contract against a checklist, the auditor comparing line items to a regulation.
Every one of those expert jobs, at their core, is someone looking at something and comparing it to a standard.
That is exactly what AI does well.
Three Questions to Find Your Next Million-Dollar Solution
Remember that thing you didn't know you had? Here are three questions that will help you find the gold mine that only you can access:
Is the pain invisible from the outside?
If the industry (or even the task) looks boring — permits, patient follow-up, road inspection, insurance claims — great news!
That is an advantage.
Boring means fewer competitors and more desperate users. The people stuck in these systems have often been stuck for years. They will pay to get out.
Are you the user (or experience the problem firsthand)?
If you already lived the problem in your daily work, you have an advantage that none of the well-funded teams can buy.
You know which part actually hurts.
You know what users forget to tell researchers. You know the workaround everyone uses, but nobody writes down. That knowledge is the product.
Is this task following strict roles, comparing, and checking?
Strip away the jargon and describe the task step by step.
If the honest answer is “we look at something and compare it to a standard”, basically, plain information processing. That is compressible, exactly the task for AI.
One Principle.
All three questions are pointing at the same thing.
A lawyer who fixes permits. A cardiologist who watches patients forget. A road technician who can’t cover enough ground. None of them came from the software or AI industry.
None of them had a technical background.
However!
All of them knew the pain inside-out — because they live it every day.
So what is the AI opportunity that belongs to you and you only?
Try starting with the one most painful, slow, hated process in your day-to-day.