Table of Contents
It’s 2025. Why Is It Still So Hard to Sell Things Online?
Justin Kan
Co-founder @ Rye
Sep 25, 2025
As a cofounder of Justin.tv and Twitch, and an experienced investor, I’ve seen a lot of different attempts at making money online. Selling physical things is still brutally, unnecessarily hard. Here's why, and what Rye's done about it.
I cofounded Justin.tv and Twitch and have spent years as an angel investor meeting founders trying to build the next big platform. I’ve seen a lot of different attempts at making money online. And I’ve come to the conclusion that people usually stick with ads and subscriptions, because selling physical things is still brutally, unnecessarily hard.
Years ago, when we wired up Twitch subscriptions, we used Stripe before most people had heard of it. Stripe made payments into a simple, easy to use API, and it was obvious how powerful that was.
Later, as an investor, I saw founder after founder pitch “Twitch for X.” They all wanted to monetize by selling stuff, but it was really difficult. They’d have to find suppliers, buy inventory, manage warehouses, integrate with ERP systems, etc.
After two decades in Silicon Valley building software companies, it’s long been clear to me that everything that can be an API should be an API. So from those conversations, there was a lightbulb moment: there should be an API for e-commerce that enables app/platform developers to sell things and make money.
It’s obvious that this API needs to work with (and utilize) agents. AI today is like Google before Google Shopping. People already use it to ask what they should buy and to compare products. But it’s blind to the things that actually make commerce work (inventory, final shipped cost, fulfillment) and it has no real way to transact.
Early versions of AI shopping show exactly how limited things are today. They either restrict themselves to the handful of marketplaces with deep API integrations (Amazon, Shopify) or they quietly fall back to humans behind the scenes to push the order through. Neither of those comes close to the real promise: AI that can buy anything from anywhere, instantly and autonomously.
Without structured, real-time data and checkout infrastructure, AI stays stuck as a shopping research tool. With those APIs implemented in a universal way, AI can actually act on behalf of users for the entire buying journey: query products, compare options, and check out.
Why don’t we have this yet? The main reason is that it’s really technically hard. Every merchant looks different, and even the platforms that do have APIs expose different schemas. Checkout flows and integrations change constantly, so something that works today can break tomorrow. Payments bring their own landmines like fraud detection and compliance. Users expect real-time pricing and inventory, so latency and reliability matter in a way that’s unforgiving.
At Rye, we’ve already spent several years tackling these problems, and it’s led us to develop an agent API that developers can call to buy anything from any store. Like my previous startups, we’ve taken a bunch of existing technologies, added some novel techniques to address specific issues, and made it into something useful. It’s kind of a crazy bet with many problems still to solve, but someone has to do it — or we’ll keep asking the same questions forever about why e-commerce is so hard.
I cofounded Justin.tv and Twitch and have spent years as an angel investor meeting founders trying to build the next big platform. I’ve seen a lot of different attempts at making money online. And I’ve come to the conclusion that people usually stick with ads and subscriptions, because selling physical things is still brutally, unnecessarily hard.
Years ago, when we wired up Twitch subscriptions, we used Stripe before most people had heard of it. Stripe made payments into a simple, easy to use API, and it was obvious how powerful that was.
Later, as an investor, I saw founder after founder pitch “Twitch for X.” They all wanted to monetize by selling stuff, but it was really difficult. They’d have to find suppliers, buy inventory, manage warehouses, integrate with ERP systems, etc.
After two decades in Silicon Valley building software companies, it’s long been clear to me that everything that can be an API should be an API. So from those conversations, there was a lightbulb moment: there should be an API for e-commerce that enables app/platform developers to sell things and make money.
It’s obvious that this API needs to work with (and utilize) agents. AI today is like Google before Google Shopping. People already use it to ask what they should buy and to compare products. But it’s blind to the things that actually make commerce work (inventory, final shipped cost, fulfillment) and it has no real way to transact.
Early versions of AI shopping show exactly how limited things are today. They either restrict themselves to the handful of marketplaces with deep API integrations (Amazon, Shopify) or they quietly fall back to humans behind the scenes to push the order through. Neither of those comes close to the real promise: AI that can buy anything from anywhere, instantly and autonomously.
Without structured, real-time data and checkout infrastructure, AI stays stuck as a shopping research tool. With those APIs implemented in a universal way, AI can actually act on behalf of users for the entire buying journey: query products, compare options, and check out.
Why don’t we have this yet? The main reason is that it’s really technically hard. Every merchant looks different, and even the platforms that do have APIs expose different schemas. Checkout flows and integrations change constantly, so something that works today can break tomorrow. Payments bring their own landmines like fraud detection and compliance. Users expect real-time pricing and inventory, so latency and reliability matter in a way that’s unforgiving.
At Rye, we’ve already spent several years tackling these problems, and it’s led us to develop an agent API that developers can call to buy anything from any store. Like my previous startups, we’ve taken a bunch of existing technologies, added some novel techniques to address specific issues, and made it into something useful. It’s kind of a crazy bet with many problems still to solve, but someone has to do it — or we’ll keep asking the same questions forever about why e-commerce is so hard.
I cofounded Justin.tv and Twitch and have spent years as an angel investor meeting founders trying to build the next big platform. I’ve seen a lot of different attempts at making money online. And I’ve come to the conclusion that people usually stick with ads and subscriptions, because selling physical things is still brutally, unnecessarily hard.
Years ago, when we wired up Twitch subscriptions, we used Stripe before most people had heard of it. Stripe made payments into a simple, easy to use API, and it was obvious how powerful that was.
Later, as an investor, I saw founder after founder pitch “Twitch for X.” They all wanted to monetize by selling stuff, but it was really difficult. They’d have to find suppliers, buy inventory, manage warehouses, integrate with ERP systems, etc.
After two decades in Silicon Valley building software companies, it’s long been clear to me that everything that can be an API should be an API. So from those conversations, there was a lightbulb moment: there should be an API for e-commerce that enables app/platform developers to sell things and make money.
It’s obvious that this API needs to work with (and utilize) agents. AI today is like Google before Google Shopping. People already use it to ask what they should buy and to compare products. But it’s blind to the things that actually make commerce work (inventory, final shipped cost, fulfillment) and it has no real way to transact.
Early versions of AI shopping show exactly how limited things are today. They either restrict themselves to the handful of marketplaces with deep API integrations (Amazon, Shopify) or they quietly fall back to humans behind the scenes to push the order through. Neither of those comes close to the real promise: AI that can buy anything from anywhere, instantly and autonomously.
Without structured, real-time data and checkout infrastructure, AI stays stuck as a shopping research tool. With those APIs implemented in a universal way, AI can actually act on behalf of users for the entire buying journey: query products, compare options, and check out.
Why don’t we have this yet? The main reason is that it’s really technically hard. Every merchant looks different, and even the platforms that do have APIs expose different schemas. Checkout flows and integrations change constantly, so something that works today can break tomorrow. Payments bring their own landmines like fraud detection and compliance. Users expect real-time pricing and inventory, so latency and reliability matter in a way that’s unforgiving.
At Rye, we’ve already spent several years tackling these problems, and it’s led us to develop an agent API that developers can call to buy anything from any store. Like my previous startups, we’ve taken a bunch of existing technologies, added some novel techniques to address specific issues, and made it into something useful. It’s kind of a crazy bet with many problems still to solve, but someone has to do it — or we’ll keep asking the same questions forever about why e-commerce is so hard.
Related Articles
Related Articles
Stop the redirect.
Start the revenue.
Stop the redirect.
Start the revenue.
Stop the redirect.
Start the revenue.
Product
Resources