DeepSeek Superagent: Permission to Work Is Part of the Architecture
DeepSeek’s research prototype separates messages, mandates and execution. Its architecture is worth studying, including the limits of its permission model.
Categories
AI
2 posts in this category
Lightpanda and the Browser an AI Agent Actually Needs
Lightpanda removes page rendering while retaining JavaScript and the DOM. Its architecture makes a useful case for matching browser infrastructure to an agent’s actual task.