Claude ships computer use, Dispatch

Anthropic is pushing Claude from chat interface into a true desktop operator, which is the right move if AI is going to do more than draft text. The hard part now is trust: once the model can touch your machine, permissions, prompt injection, and safe app boundaries become the product.

• –Computer use fills the gap when APIs and connectors stop short, especially for browser workflows, IDE chores, and file handling.
• –Dispatch is the more durable wedge because cross-device task handoff maps to real work habits, not just demo wow-factor.
• –The macOS-only research preview is a sensible brake on hype, but it also means Anthropic still has to prove reliability before this feels mainstream.
• –The safety warnings are not boilerplate; GUI agents are harder to secure than text-only copilots, so adopters will need to sandbox sensitive workflows.