We built a single, connected workspace, conceived for the AI era. It had an infinite canvas with a browser that allowed users to access all their existing web tools, and an assistant that would understand the work going on in each of them. Users could ask the AI assistance for summarizing, drafting, categorizing, answering questions and more. An important part of the workspace was to go beyond the chatbot and explore new ways of interacting with AI.

The canvas resembled the experience of working alongside an AI. Like sharing screens in multi-player mode, the user gets a clearer grasp of what’s going on, can work on tasks in parallel, easily give context, and even edit results directly.

https://x.com/getperhaps/status/1767669622977810788

The workspace lived in a desktop app, which allowed it have a fully functional browser. This meant users could log into their accounts and use what was in there as context. A window into any tool for the AI.

https://x.com/ignaciosoffia/status/1789084341064372394

A novel interaction was selecting elements in the canvas to pass it as context to the AI.

A key technology we developed was translating agentic reasoning into canvas actions and creation of components, thus introducing a new interface layer for LLMs.

https://x.com/getperhaps/status/1767697286203576572

The Workspace had a powerful experimental feature, where the AI could directly control the browser.