Had a good catch up this week with a former colleague who used to build developer communities and spread the word about APIs with me. He raised an important and fundamental topic that fits the AI rush and is a good test for any founder’s problem statement.
Back in 2009, shoppers in Korea couldn’t buy or bid on most Amazon and eBay listings. So I built a shopping agent, years before generative AI, that bought and bid for them. The business goal was to open a market closed to a whole country. The engineering goal was an agent that could act for each shopper on platforms never built for them.
A decade later, I worked on routing. The traveling salesman problem asks for the best route through many stops and back home. Add many vehicles, each with its own capacity, shift, and delivery windows, and you get the multi-vehicle routing problem. Leading a 70+ person engineering department that built for this, which was a challenge of its own. We planned and routed hundreds of millions of requests on the fly, scaling both out and up, with observability so each round of planning was checked against business metrics.
In 2021 I shared with FreightWaves the data needed for real time decisions crossed point to point connections millions of times a day, and most of it sat in silos. That was harder than the math.
These problems are still NP hard in theory, yet decades of solvers handle real cases so well that few people notice.
AI is going the same way. Startups are building harnesses that send each user request to the right model or provider, weighing cost, speed, quality, and GPU capacity. That’s a scheduling and assignment problem, close kin to vehicle routing.
When tools for a known problem get this good, selling one more fix gets harder. A new routing or supply chain vendor now competes with agents that do much of that work for far less.
AI also lets a small team take on problems that used to be too complex to try, with more moving parts, messier data, and people whose needs change by the hour. That’s where most valuable outcomes are created today, and a better shot at a lasting business.
Questions I’d ask as a founder… would be something like, could an agent solve this next year for a fraction of the cost? And if that’s the case, what’s still hard after that?