Anthropic ships team automation
- Anthropic released team‑focused workflow features including Cowork, Routines, and Managed Agents. - Routines support scheduled or webhook runs roughly 5–25 times per day, and Rakuten reported biweekly releases plus a 97% error drop using Managed Agents. - The updates aim to automate routine team tasks and reduce production errors across enterprise workflows ( ).
Anthropic has rolled out a set of team automation tools that let Claude run recurring work on its own, not just answer prompts. (anthropic.com) The new lineup spans Cowork for desktop knowledge work, Routines for scheduled or event-driven jobs in Claude Code, and Managed Agents for cloud-hosted agents on Anthropic’s platform. Managed Agents entered public beta on April 8, 2026, and Anthropic said the service is built for “long-running” sessions that can keep working for hours. (anthropic.com) Routines arrived in research preview last week as saved Claude Code setups that can run on a schedule, from an application programming interface call, or from GitHub events on Anthropic-managed infrastructure. Anthropic’s documentation says plan limits range from 5 runs a day on Pro to 25 runs a day on Team and Enterprise. (claude.com, code.claude.com) Cowork is Anthropic’s pitch to non-technical teams that want an agent to move through files, folders, and desktop apps and return a finished document, summary, or spreadsheet. Anthropic says it built the product after internal marketing and data teams started using Claude Code for multi-step tasks that did not fit a chat box. (anthropic.com) The shift is toward software that handles a whole assignment after a goal is set, instead of waiting for one prompt at a time. Anthropic says Cowork is meant for repetitive desktop work, while Managed Agents handles the harder part for companies that want secure execution, permissions, checkpointing, and tracing in production. (anthropic.com, claude.com) Anthropic’s engineering team has framed that production problem as an infrastructure problem as much as a model problem. In a post published this week, the company said it separated the agent “brain” from the execution layer so the system can keep working even as model behavior changes. (anthropic.com) Rakuten is Anthropic’s clearest early case study. In a customer interview published this week, the company said it moved from major software releases once a quarter to every two weeks and cut initial critical errors by 97% after adopting Claude Managed Agents. (claude.com) Rakuten also said it is deploying specialist agents across engineering, product, sales, marketing, and finance, with each agent going live within a week. The agents plug into Slack and Microsoft Teams and return outputs such as apps, proposal decks, and spreadsheets. (claude.com, claude.com) Anthropic is packaging these products around a broader workplace argument: chat is useful for answers, but teams also want software that can finish recurring work without a person babysitting every step. The company is now selling that idea to both developers wiring up cloud agents and office teams delegating file-heavy desktop tasks. (anthropic.com, anthropic.com) The test now is whether companies trust those agents with more of the work that sits between a prompt and a finished deliverable. Anthropic’s latest release is built around that handoff. (claude.com, anthropic.com)
