The next big AI startup might be hiding in a job OpenAI isn’t even testing for.
Mercor just mapped 29 major AI benchmarks against 923 occupations in the US economy. What they found is a massive blind spot:
90% of American working hours are spent on tasks with little or no benchmark coverage.
Think about that.
AI labs spend billions making their models better. But much of that improvement is measured on a surprisingly narrow set of jobs.
Lawyers: 18 of 22 tasks tested.
Legal secretaries: 1 of 14.
Court clerks: 0 of 30.
Freight forwarders: 0 of 31.
Medical records specialists: 0 of 17.
And here’s why this matters for founders.
AI labs use benchmarks to find where models fail, build training data around those failures, and make the next generation better.
That creates a feedback loop: the jobs that get tested attract more targeted improvement. The ones that don’t are easier to overlook.
And that’s where vertical AI startups have an opening!
While the big labs optimize for the same benchmarks, founders can build specialized agents for the workflows nobody is measuring: customs declarations, medical approvals, court filings, construction permits, insurance claims.
Some are already turning these overlooked tasks into million-dollar businesses.
📚 In this newsletter we mapped 8 overlooked job categories, the startups already winning in them, and the specific AI businesses you could build next.
Inside you will find:
Why AI Only Gets Better Where It Gets Tested See the loop that makes models improve on some jobs and stand still on the rest.
Freight and Customs: How Evos Hit $1M ARR in Four Weeks Find out why the people who move goods across borders are almost untested.
Healthcare Paperwork: How Forus Went From $1B to $3B in Four Months Learn why the faxes, approvals and medical records around the doctor are wide open.
Courts and City Halls: The Legal Jobs With Zero Coverage See why clerks and city staff get none of the attention lawyers get.
Loans, Claims and Appraisals: How Vesta Grew Revenue 12x Find the money jobs the labs skip once you go below the analyst.
Engineering and Permits: How Solvera Booked $1M in Three Months Learn how startups sell finished drawings and designs to the people who build things.
Special Education: The 12-Month Wait Startups Are Going After See why special education has zero tasks tested and what founders are building there.
The Trades: How UpSmith Makes Money for Plumbers and Electricians Find out why the office of a trade business is the easy part to automate.
This issue is available exclusively to paid subscribers!
Paid subscribers unlock the full library: 200+ decks, 10,000+ investors, AI Agent playbooks, and fresh high-signal content every week.
Not subscribed yet? Try it free for 7 days and unlock this guide, plus our full premium library:
We’re already 80,000+ founders and operators inside!
Let’s begin!
1. Why AI Only Gets Better Where It Gets Tested
