Our $50M Markup Shows Why Small Business Is the Last AI Frontier.
Harper, the AI-native insurance brokerage we backed coming out of YC W25, raised a $47M Series A led by Emergence Capital.
The markup was significant, but more importantly, the thesis is working.
When asked what makes Harper, founder Dakotah Rice said: “We want to target middle America. The real-world businesses like daycares, manufacturers, car dealerships, local bars and restaurants.”
Silicon Valley has a blindspot… Is it building AI for itself while ignoring where most of the economy actually lives?
Two-Thirds of America Has Never Used ChatGPT
ChatGPT has 900 million weekly active users, which sounds massive until you realize only 34% of American adults have ever used it.
Two-thirds of the United States - the most technologically advanced country on the planet - has never even opened the most popular AI tool in history.
Even still, of those 900 million users, only 35 million are paying subscribers, which means 96% are on the free tier, probably asking it to write happy birthday messages and explain what a joke they heard on TV means.
Consider this tweet from Garry Tan.
If 20 year old software engineers haven’t heard of Claude… well… Now think about small business in that context.
I drive down any commerce corridor and see dry cleaners, restaurants, retail shops, bars - the actual fabric of the economy - and I ask myself… do they even know about AI?
You look at these businesses and there’s so much momentum in continuing to do things the way they’ve always been done, not because people don’t want to improve but because they’re busy running their business, trying to spin 100 plates and don’t have time to figure out new technology. Even if it might help.
But something’s about to happen.
AI is undeniably putting high-level tools in everyone’s hands - accounting, legal, marketing, really high-level information and actionable items that used to require hiring specialists.
Imagine that gap was bridged into the commerce corridor.
Think about running a restaurant where you’re cooking all day and now you have something that manages your social media, handles reservations, interacts with customers so you don’t have to take phone calls anymore.
The routine stuff like making reservations, changing reservations, noting dietary restrictions - all of it gets automated so you can focus on the actual craft of running a restaurant.
Yes, the headlines get grabbed by corporations talking to The Wall Street Journal about their AI transformation initiatives. But when you look at the GDP data, small business IS the economy.
However…
It’s easy to think sat on X all day that everything is moving so fast.
But really, tech moves quite slowly.
Data from early 2025 shows approximately 60% of corporate data is stored in the cloud… Yes it is up from 30% in 2015, which represents significant progress for large enterprises.
But that also means 40% of corporate data still sits on-premises or in legacy infrastructure, and small businesses aren’t even in these statistics.
40%!
So before we consider if your agentic AI startup is going to wipe Oracle off the map have you considered asking the corporations you are targeting if your agent can even read their files?
Startups targeting real America need to step outside their bubble.
Building for People Unlike You
What I mean is Silicon Valley has a structural blindspot.
We’re the 1%.
Building products for the tech elite because it’s easier to build for yourself. As VCs, we invest in companies we might actually use, which means we systematically miss opportunities outside our lived experience.
For example, Chime is worth $10B today, but they got 100+ rejections at Series A.
VCs said products for the underbanked could never be profitable. Those investors couldn’t relate to the customer, so they passed on what became a massive company serving a market they didn’t understand.
The standard advice for fixing this is to talk to people as far from the Valley as possible, understand their problems, then translate powerful new technology to solve them.
But there’s a counterargument that creates tension in how VCs think about opportunity:
Paul Graham popularized the idea that investors should focus on what hackers build on nights and weekends, because personal projects signal genuine passion and uncover breakthroughs before mainstream traction validates them.
One side says build for people unlike you.
The other says build what you’re personally obsessed with.
So does PG’s heuristic have the same blindspot?
Probably… It naturally favors hackers from top schools who have the luxury of spare time for side projects and the technical sophistication to see possibilities before they’re obvious.
But ironically, I think YC solves this problem through sheer volume and brutal selection.
With 500+ companies per batch now, YC funds plenty of founders building for markets the Valley doesn’t understand, simply because they can’t all be building the same consumer social app or enterprise SaaS tool.
What the Next Wave Looks Like
Harper fits this pattern. Insurance brokerage isn’t a product you’d build for yourself as a weekend project.
Most technical founders avoid it because it requires regulatory compliance, state-by-state licensing, undocumented industry norms, and earning credibility with underwriters who’ve spent 30 years in the business. It’s slow, unsexy, doesn’t fit the “build → launch → scale” venture script that most founders internalize.
But that’s exactly why it works as a business opportunity.
And critically, the customers Harper targets don’t care about AI. A daycare operator in Ohio or a car dealership in Nevada doesn’t want to hear about your large language models or your automation infrastructure. They care about getting insurance quotes faster and paying less for coverage. Harper delivers that using AI, but from the customer’s perspective it’s just better service at better prices, which is all that matters.
Look at the upcoming YC batches and you’ll see more companies breaking this mold. These are what I think of as “translators” - taking cutting-edge AI capability and making it work for people who don’t live in San Francisco and don’t care about the underlying technology, they just want their problem solved.
GrazeMate is building robot cowboys that muster cattle. Not AI agents for knowledge workers, but autonomous systems for ranchers managing thousands of acres who need help with physical labor in remote locations.
Verdex is helping to insure over 11% of all American farmland. Agriculture produces 10% of global GDP but runs on equipment from the 1980s and manual record-keeping because nobody’s built modern software infrastructure for an industry that looks boring to Silicon Valley.
OctaPulse uses computer vision to automate fish farming. Aquaculture is a $250B+ industry globally, growing faster than any other food production sector, and it’s almost entirely manual because the technology to monitor and manage thousands of fish at scale didn’t exist before AI made visual pattern recognition cheap and reliable.
Autositu is building an AI workspace for municipal development review. City planning departments are drowning in applications they can’t process fast enough, using systems that haven’t changed in 30 years, because local government moves slowly and the vendors serving them have no incentive to innovate. Now someone’s finally building the tool those departments desperately need.
These aren’t sexy Silicon Valley problems or what most hackers build on weekends.
But they are massive markets where AI can create real step-change improvements because the baseline is so low and the pain is so acute that even a mediocre solution would be better than the status quo.
Small business is the last AI frontier, and the companies that win there will look very different from the darlings in the Valley.
Visit lobstercap.com for more.