Cowork has handled flight booking before, but with Opus 4.7 it 1-shotted the entire process for the first time, a meaningful step up from previous attempts that required more iteration.
Agent View is a new Claude Code native interface for managing multiple sessions, designed as a purpose-built alternative to tmux for AI coding workflows.
Agent view is purpose-built to run more sessions simultaneously while keeping cognitive overhead low.
Running `claude agents` opens a control plane in your terminal for managing all active Claude Code agents in one place. From any CLI session, hitting the left-arrow key registers that session with the control plane, making it easy to oversee multiple agents from a single root directory.
With flight preferences set in Cowork instructions, Opus handled the full booking flow unattended: opening browsers, navigating travel sites, and completing reservations for 8 flights and 5 hotels while the user worked on something else entirely.
Fast mode for Opus 4.7 is now in research preview, available on both the API and in Claude Code.
The default fast model has been updated to Claude 4.7. Not 4.6.
The team is actively working on quality improvements and wants specific friction points from users across CLI, Desktop, and VSCode. A direct invitation to report small annoyances, not just big bugs.
A user is burning through usage limits unusually fast. The team suspects a plugin may be consuming quota silently, and is asking for /usage output to debug. Work is underway to make this kind of diagnosis more self-serve.
A fix for the issue raised by the user is shipping in v2.1.140 the following day.
Auto mode in Claude Code is the suggested way to safely delegate permissions to Claude, per a team member responding to a user question.
In the file picker or insertion UI, hitting space gives a preview while enter commits the selection inline.
Common workflows can be packaged as skills and dispatched in parallel. Claude runs autonomously until a skill signals it should stop and ask, at which point a quick reply unblocks it without needing to open the full transcript.
Running many agents in parallel is now cleaner, eliminating the need to cycle between terminal tabs. The linked resource appears to show a workflow for scaling from one agent to many.
Running 'claude agents' from a high-level directory containing all your repos lets it track which sessions need input and makes resuming work significantly easier.
A new 'claude agents' command is now live and available to run today.
The agent view panel surfaces background sessions. Running /bg inside any existing session sends it to the background, where it appears in that list.
Passing '--bg' to each Claude invocation lets you kick off multiple background tasks simultaneously. Claude won't spawn parallel tasks on its own; for orchestrated multi-agent work, the Agent Teams feature handles that.
You can change a session's color in the list by running /color either via peek or from inside the session itself.
