Claude can scan inboxes
Claude's new computer‑use feature can scan inboxes, pull metrics, and auto-generate reports — framing the model as a 'junior ops hire' that automates A/B test summaries and dashboard updates. Proponents say this shortens analyst busywork and speeds creative iteration; the feature was called out in social discussion on March 24. (x.com)
Anthropic published a research-preview rollout of its “computer use” capability on March 23, 2026, and said the feature is available inside Claude Cowork and Claude Code for Pro and Max subscribers on macOS. ( ) The capability includes screenshot capture plus direct mouse-and-keyboard control so Claude can open apps, navigate web pages, complete forms and populate spreadsheets without manual clicks. ( ) Anthropic demonstrated persistent scheduling and memory for multi-step flows—examples in their materials show Claude taking a single instruction and then running that task on a recurring timetable without repeated prompts. ( ) The company says each run executes inside an isolated virtual machine with user-set folder-level access controls and network allowlists, and that users can observe actions in real time and pause or stop the agent at any moment. ( ) Anthropic added automated monitoring for prompt‑injection attempts and defaults to blocking high‑risk or sensitive apps, while advising customers to start with trusted applications and to avoid exposing sensitive data during the preview. ( ) Developers can build with the computer‑use beta via the Anthropic API and partner platforms including Amazon Bedrock and Google Cloud Vertex AI, and Anthropic tied the capability to its Claude 3.5 Sonnet model upgrade. ( ) Industry coverage and social posts about the feature appeared on March 23–24, 2026, and an X post from Aadit Sheth on March 24 highlighted workflows that automate routine reporting and ops tasks (original post linked). ( )
