The post-browser computing layer

Cotton is a bet that the web browser is the wrong interface for an era defined by artificial intelligence. For thirty years, the browser has functioned as a visual window for humans to navigate—a manual exercise in clicking, scrolling, and tab-switching. Cotton attempts to abstract this away by building an AI-native operating system that treats the web as a set of functions to be executed rather than pages to be viewed. The core thesis is simple: users do not want to 'browse' for a flight or a piece of research; they want the flight booked and the research summarized.

Founding engineer Zeyu Wang, who was the first employee at Perplexity AI, leads the team. This pedigree is significant. Perplexity spent its first two years disrupting Google by proving that users prefer a direct answer to a list of blue links. Cotton represents the logical next step in that evolution. If Perplexity solved the search problem, Cotton is designed to solve the execution problem. The company is based in San Francisco and operates at the intersection of consumer software and autonomous agent infrastructure.

Moving from search to action

Unlike traditional browsers that attempt to organize your tabs or offer sidebar AI assistance, Cotton is built around a goal-oriented interface. When a user inputs a command, the system does not just point them to the right URL; it interacts with the underlying elements of the web to perform the action. This requires a sophisticated visual and structural understanding of various websites, essentially creating a 'world model' of the digital environment.

This approach places Cotton in direct competition with 'Computer Use' frameworks from foundation model providers like Anthropic and OpenAI. However, Cotton’s advantage lies in its focus on the user interface. While Big Tech provides the raw capability for a model to move a cursor, Cotton is building the product layer that makes that capability usable for a non-technical audience. It is an attempt to create a 'Post-App' world where the specific software used to complete a task is secondary to the goal itself.

Competitive market and execution risks

Cotton sits in a crowded but high-stakes sector. They are competing for the 'attention layer' of the computer, currently held by Chrome and Safari. The Browser Company has already made significant inroads here with Arc, but even Arc remains fundamentally a browser. Cotton is more ambitious, aiming to be the primary execution layer that sits between the user and the web.

There are clear tradeoffs to this model. Total reliance on agents for web execution introduces latency and the risk of hallucinated actions—a much higher stakes error than a hallucinated fact in a search result. If an agent books the wrong hotel, the cost is real. Cotton's success will depend on its ability to prove that its autonomous execution is as reliable as manual clicking. As OpenAI prepares its 'Operator' agent and Anthropic refines its computer-use models, Cotton must move quickly to establish its interface as the default starting point for the agentic web.