Running one AI agent at a time is starting to look like a bottleneck. Anthropic’s redesigned Claude Code desktop app, released April 14, treats parallel sessions as the default — not a workaround.
The new interface puts a session management sidebar front and center. From one window, you can kick off work across multiple repositories, watch them run, and pull finished results back when Claude surfaces them. Each session gets its own isolated copy of the project through Git worktrees, so parallel sessions don’t step on each other until changes are committed.
What changed underneath the cosmetics matters more than the layout itself. The app now includes an integrated terminal, file editor, rebuilt diff viewer, and HTML/PDF preview — all draggable into whatever arrangement makes sense for the task at hand. That eliminates constant context-switching between Claude and the tools developers were already using. The signal from Anthropic is clear: the desktop app is supposed to replace your workflow, not sit alongside it.
A new Side Chat (⌘ + 😉 lets you open a parallel conversation that borrows context from your main thread without feeding anything back into it. It’s designed for quick lookups or sanity checks that would otherwise pollute the session state — a small feature that suggests Anthropic has been watching how developers actually use the tool.
The bigger addition is Routines, currently in research preview. Routines package a prompt, one or more repositories, and connectors into a saved configuration that can run on a schedule, trigger via an HTTP API endpoint, or fire in response to GitHub webhook events. Critically, they run on Claude Code’s cloud infrastructure — your Mac doesn’t need to be online when a routine executes. That’s a meaningful architectural shift. It moves Claude Code from a tool you actively use toward something closer to a background process with judgment.
Whether Routines gets widely adopted depends on trust. Developers need to believe Claude will execute automated tasks reliably enough to not require babysitting. Anthropic is betting the answer is yes. The evidence for that bet is thinner than the product announcement implies — the research preview label is doing real work here.
Three view modes — Verbose, Normal, and Summary — give developers control over how much of Claude’s reasoning gets surfaced. That’s a practical acknowledgment that full transparency is overwhelming, but zero transparency is unsettling. Most developers will probably settle on Normal and forget it exists.
The redesign is available now for Pro, Max, Team, and Enterprise Claude users on macOS and Windows. Routines won’t work until you trust them. And trust, in agentic software, is earned line by line.
Sources
Anthropic Blog — 9to5Mac — SiliconANGLE
This article is AI-generated.
