On April 14, 2026, Anthropic shipped a research preview of Claude Routines, a new feature inside Claude Code that lets users schedule repeatable, automated AI tasks. A routine is a saved bundle — a prompt, one or more code repositories, and a set of connectors — that Anthropic runs in the cloud on a schedule or in response to events. The user's laptop no longer has to be online for the task to execute.
The company describes the model as "set it and forget it" for repetitive engineering work: nightly dependency audits, weekly security sweeps, scheduled refreshes of documentation, event-driven responses to new issues or pull requests on GitHub. Example use cases highlighted at launch include scheduled tasks, API workflows, and GitHub-triggered routines. The feature is available across Claude's paid tiers, with daily run caps that scale by plan — 5 runs per day for Pro, 15 for Max, and 25 for Team and Enterprise customers.
Routines ships alongside a redesigned Claude Code desktop app. The new UI lets users run multiple Claude sessions side by side from a single window, managed through a new sidebar — a clear response to the growing pattern of power users juggling several Claude instances at once. Together, the two updates move Claude Code toward something closer to an agentic development environment than a single-shot assistant.
For students and early-career developers, this is worth paying attention to for two reasons. First, the economics: automated, scheduled AI agents that maintain code, review PRs, and triage issues change what "junior engineering work" looks like at scale — the tasks that used to be on-ramps into a codebase are increasingly doable by a scheduled routine. Second, the craft: the people who will build, operate, and supervise these routines are a new skill layer in software engineering. Understanding how they're configured, where they fail, and how to audit what they changed will be part of day-one expectations sooner than most curricula are ready for.