Descriptions:
Felix Rieseberg, member of technical staff and product lead for Claude Cowork at Anthropic, joins Alessio and Swyx on the Latent Space podcast to discuss the product decisions and philosophy behind Claude Cowork — the desktop-native interface built on top of Claude Code for users who aren’t comfortable working in the terminal.
Rieseberg explains that Claude Cowork emerged from observing non-technical and semi-technical users adopting Claude Code for non-coding tasks: managing expenses, organizing Obsidian vaults, handling receipts. Rather than leaving those users in the terminal, Anthropic wrapped Claude Code in a desktop UI with guardrails and a more accessible entry point. He pushes back on the notion that “user-friendly” means less powerful, arguing Cowork is actually more capable for many workflows because it integrates directly with the user’s existing computer environment — browser, local files, and cloud systems simultaneously.
A substantial portion of the conversation addresses Anthropic’s contrarian view on local compute: Rieseberg argues that AI agents need access to the same tools a human uses — including the full desktop — to avoid being artificially constrained. He also discusses the deliberate choice not to build a proprietary browser, instead integrating with Chrome and other existing browsers to meet users where they already work. The episode offers a rare inside look at how Anthropic is thinking about the design space between power-user agent tooling and accessible consumer AI products.
📺 Source: Latent Space · Published March 17, 2026
🏷️ Format: Interview
