Anthropic Gives Claude Code Automated Routines and a New Desktop

Anthropic shipped two changes to Claude Code on April 14 that together make a case for how agentic development will actually look in production. The first was a full redesign of the Claude Code desktop application, built around parallel sessions. The second was "Routines," a cloud-execution system that lets saved automations run on Anthropic's infrastructure rather than on the developer's local machine. Both address the same underlying constraint: developers working with AI agents need to run multiple workstreams simultaneously, and they cannot be tethered to their laptop to do it.

The new desktop opens with a session sidebar that collects every active and recent Claude Code session in a single view. Each session operating inside a Git repository gets its own isolated copy of the project via Git worktrees, so two agents can work in the same codebase without stepping on each other's changes until the work is deliberately merged. Sessions can be filtered by status, grouped by project, or searched by name. For teams managing more than one agent at a time, this removes a friction point that had previously limited how aggressively developers would parallelize work.

Beyond the sidebar, the redesign pulls several tools that developers previously had to context-switch away from the app to access. An integrated terminal sits alongside the conversation pane. An in-app file editor handles spot edits without opening a separate application. A rebuilt diff viewer renders large changesets faster than the previous implementation. And the preview pane now handles HTML files and PDFs in addition to local servers, giving design-adjacent and document-heavy workflows a place in the tool. Every pane is drag-and-drop, letting developers arrange the workspace to match their preferences.

Routines: Agents That Don't Need a Babysitter

The more architecturally significant change is Routines. A Routine bundles a prompt, a repository, and any relevant connectors into a configuration that executes on Anthropic's cloud infrastructure, not on the developer's machine. The practical effect is direct: a nightly bug triage, a weekly dependency scan, or an automated pull-request review can now run at 2 AM without the developer's laptop being on.

Routines come in three forms. Scheduled Routines fire on a cron-style schedule, suitable for recurring tasks like documentation drift scanning or backlog grooming. API Routines expose a dedicated endpoint and authentication token, letting external services trigger Claude via HTTP from alerting platforms like Datadog or CI/CD pipelines. GitHub Routines connect to repository events directly: a Routine can kick off automatically when a new pull request opens, when a commit lands on a branch, or when a CI check fails. The combination covers the three patterns engineering teams encounter first when they try to automate repetitive work: run this on a schedule, run this when an external system says to, and run this when this repository changes.

Routine quotas scale with plan tier: five per day on Pro, fifteen on Max, and twenty-five on Team and Enterprise plans. Routines launched in research preview, which means Anthropic may iterate on the underlying implementation as adoption grows. Teams building critical workflows on Routines during the preview period should account for that.

"The goal is that a Routine should feel like a cron job you'd trust to run in production, not a fragile script that needs hand-holding." Anthropic documentation, April 2026

What This Changes for Engineering Teams

Enterprise buyers evaluating Claude Code in 2026 are not asking whether an AI model can write code. They are asking whether it can slot into existing engineering workflows without creating new supervision overhead. Routines answer that question directly. A team can define what a good pull-request review looks like, attach the task to a GitHub Routine, and let Claude handle the first pass on every PR without a reviewer needing to queue it manually. The quality postmortem Anthropic published in April showed that output consistency was a major concern for enterprise users; Routines with defined rubrics address exactly that.

The session isolation provided by Git worktrees is equally significant for teams running more than one agent at a time. In earlier versions of Claude Code, running two sessions on the same repository risked file conflicts or inconsistent state. Worktree isolation removes that risk at the architecture level, making it practical to have one agent handling a feature branch while another works through the main branch for regression risks. The Code with Claude developer conference in May highlighted parallel agentic execution as the primary direction for Claude Code's enterprise growth, and the desktop redesign reflects that commitment in the product itself.

Claude Code's growth trajectory makes these additions timely. The tool's annualized revenue run rate crossed $2.5 billion in early 2026, and it has added daily active installs at a rate that put it alongside GitHub Copilot and Cursor as one of the three most widely used AI coding platforms. Routines and the redesigned desktop are built for the enterprise customers driving that adoption: teams that need agents to work on schedules, respond to events, and coordinate across repositories rather than waiting for a developer to open a terminal. Whether Routines remain in research preview through the summer or graduate to general availability on a faster timeline depends largely on how those enterprise teams put them to work.