Intel / post

Claude Code cloud sessions exit research preview

Source: x.com
Date: Published Sep 23, 2026

From the Daily Brief

Cloud sessions move a Claude Code agent off the developer's machine and onto Anthropic-hosted infrastructure. With the research preview label removed, long-running tasks no longer depend on a laptop staying awake or a terminal staying open. Anthropic is giving Pro and Max subscribers a one-time credit of $100 or $250 respectively, available through October 7. The practical change is in how work gets scheduled: tasks that take hours can be started and checked later rather than supervised in real time. Teams adopting it should decide up front which repositories and credentials a hosted session may touch, since the agent will be running while nobody is watching.

Read the 2026-09-28 Brief →

Context

Claude Code's cloud sessions are now an officially available feature rather than a research preview, and Anthropic is giving existing subscribers a one-time credit to try them: $100 on the Pro plan and $250 on Max. The announcement came from Anthropic's ClaudeDevs account, which describes cloud sessions as a way to keep Claude Code working even when your laptop is closed.

Claude Code is Anthropic's coding , and an ordinary session runs on the developer's own computer, so it stops when that computer sleeps or shuts down. A cloud session runs the work in the cloud instead, which is what lets a task continue after the laptop lid closes. Leaving research preview means Anthropic now presents the feature as a supported part of the product instead of an early experiment.

The post's text does not say how long the credit lasts, how cloud sessions are charged once it is spent, or how to start one, and the image and page it links to were not available for this explanation. Anthropic's ClaudeDevs account later published a field guide to cloud sessions that covers what changes when Claude Code runs on its own machine and how to set it up.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.

Read the source x.com