on vertical AI, Motown, and the bitter lesson https://x.com/i/article/2116175062772453376
It’s Better to Be Different Than Better
The history of the human economy is the history of specialization of labor.
It’s the first chapter of Adam Smith’s The Wealth of Nations: one man making pins alone might not make twenty a day, but a team of ten working on specialized steps could produce tens of thousands. America’s 20th-century industrial economy was scaled around the assembly line, and it trickled down to our culture. Berry Gordy worked the Lincoln-Mercury line before he started Motown. We built professions, licensing boards, org charts and entire cities around the idea that a person builds a career getting very good at one increasingly narrow thing, then trading for everything else. Our educational system, once built to accommodate an agrarian society, has adapted. The pathways we offer young people into society are increasingly narrow: every state in the country offers career and technical education in high school.
I built my first company to solve the pain of managing complex academic pathways at the K12 level. We started in Houston ISD, where high schoolers can graduate with a pilot’s license at Sterling High School, next to Hobby Airport; we grew it for six years before a strategic acquisition. A vertical SaaS company is an organizational expression of the specialization of labor: an industry’s knowledge and shared workflows captured as best a web interface and cloud-hosted back end can.
It’s easy to look at the AI paradigm shift as a threat to this vertical specialization. The “bitter lesson” of scaling models is that general-purpose models trained with more compute and data outperform specialized, narrower ones. Imagine Adam Smith’s digital pin factory, where Claude Code draws the wire, straightens it and cuts it. The pin assemblers join the permanent underclass. But if every model keeps getting generally smarter, intelligence is the thing everyone has. My old board chairman, Kent Hudson, used to say that it’s better to be different than better. Better is a commodity, people are drawn to what’s different. Kent didn’t mean quality doesn’t matter, you still have to be right. He meant that “better” is nebulous, and for customers who don’t feel understood by large horizontal players, differentiation drives trust. K12 is one of these markets; we competed against multi-product suites as a point solution and never lost a deal to them.
There are plenty of markets well suited to large horizontal players, but the AI world underestimates how much differentiation it will take to deploy AI across the economy. AI won’t arrive in the real economy through a few labs. It will arrive through thousands of new companies, and a new generation of operators will build generational wealth that’s largely illegible to the Twitter investor class. Our digital tribes have split off from the rest of our lives. If you listen to the same podcast as someone across the country, you probably feel closer to them than to the family zoned for the same public school. Fully digital environments can be run autonomously with enough compute and reinforcement learning, but general-purpose models cannot deploy themselves into the real world. A model can write a home builder’s change orders, but needs an FDE riding in the truck to learn that the real schedule lives in his text messages to the framing crew. An agent can buy you a beer but can’t get a VP drunk enough to tell you where there’s budget left over. A large language model can’t make a handshake deal.
America’s industrial century ended with a national network of invisible millionaires with local service monopolies: car dealerships, franchises, endocrinologists. At Lafayette we’re building a factory for these businesses. We started with eleven young operators, almost none of them engineers. The group included high school founders and college athletes, a youth powerlifting champion, a bowhunter and a playwright. Motown’s assembly line worked the same way. The Supremes were teenagers from the Brewster-Douglass projects when Gordy signed them. We gave each operator a vertical AI deployment business in a box and the runway to make it self-sufficient. On our platform, they do the diffusion work that the labs can’t: learning, one customer at a time, how a home builder in Houston or an agricultural retailer in eastern North Carolina actually runs. Each one landed paying customers, and the portfolio's booked revenue exceeded our total investment within five months of our summer kickoff in NYC. We want a Motown-sized run of hits.
Almost all businesses are capacity constrained. Over the next ten years the successful ones will use AI to change the way they work. They will run their back offices on agents and spend the time they get back with customers. Managed IT providers earn about $80 billion a year running IT for other companies. The technology is generic, but businesses want to buy it from someone in their network. There are 50,000 MSPs in the country, and 40,000 of them have less than $10 million in annual revenue. AI service markets will mature in similar ways and become much bigger than IT services at scale, a trillion-dollar annual market built on vertical differentiation rather than commodity intelligence. It’s better to be different than better.
Thanks to Will Manidis, Jeremy Giffon, Packy McCormick, Bill Janeway, Eron Lutterman, Alex Athansiou, Eli Dukes, Sri Komanduri, Matthew Lewis, Zach Ware, Zac Rollins