# endpoint privacy

When AI Moves In: Desktop ‘Agents’ from Claude to QoderWork Promise Productivity — and New Risks

Desktop AI agents—software that gives large language models permission to read, write and operate a user’s local applications—have transitioned from experiments to commercial products. Anthropic’s Claude Cowork and open-source projects like OpenClaw have catalysed a market response in China, where Alibaba’s QoderWork, MiniMax’s Agent2.0 and Step’s desktop partner offer competing approaches. The shift promises productivity gains and clearer monetisation but raises substantial security, privacy and regulatory challenges.