Published 10 July 2026
From commerce to AI: what 20 years taught us
PATH’s first projects were online shops. The latest are autonomous agents that answer support tickets and screen candidates. The technology could not be more different. The lessons are mostly the same.
The customer does not care about the model
In 2008 nobody bought from a shop because of its database. In 2026 nobody keeps an AI product because of its model. They keep it because it did the job on Tuesday and again on Wednesday. Reliability beats capability.
Operations is the product
The best commerce companies we worked with won on returns, delivery times and stock accuracy, not on the storefront. The AI products that stick are the ones with boring parts done well: permissions, audit logs, fallbacks when the model is wrong.
Integration is where the margin is
A shop that talked to the ERP, the courier and the bank was worth more than a prettier one that did not. An agent that lives inside the tools a company already uses is worth more than a chat window beside them.
Start narrow
Every marketplace that tried to launch with every category struggled. Every one that started with one category and owned it grew. The same is true for agents. One function, one team, one measurable outcome.
What we look for now
Founders applying AI to a workflow they know well, in a market where the operational detail matters more than the demo. If that is you, we would like to hear about it.