Anthropic Pushes Workflow Lock‑In

Anthropic is moving Claude deeper into office software and managed automation just as some of its most active users say the product has become less reliable. (venturebeat.com, aol.com) The company launched Claude Managed Agents in public beta on April 8, pitching a hosted system that handles orchestration, sandboxing, and governance for enterprise agents instead of leaving that work to customer engineering teams. VentureBeat reported the product gives companies a “one-stop shop” for deploying agents while shifting more of the stack inside Anthropic’s platform. (venturebeat.com) Anthropic has also been wiring Claude into Microsoft workflows. Its help center says the Microsoft 365 connector can search SharePoint, OneDrive, Outlook, Teams Chat, and Teams Calendar, and requires a Microsoft Entra Global Administrator to grant tenant-wide consent before employees connect. (support.claude.com) Inside Claude itself, Anthropic is pushing reusable “Skills,” which the company describes as packages of specialized knowledge and workflows that Claude loads when needed. On Team and Enterprise plans, owners can upload skills organization-wide, and Anthropic’s directory now shows shared skills, connectors, and plugins in one place. (support.claude.com, support.claude.com, support.claude.com) That setup moves Claude from a chatbot into something closer to workplace plumbing. Anthropic’s own documentation says Claude can work across Excel, PowerPoint, and Word add-ins so it can read from one Microsoft app and make changes in another without users copying material between them. (support.claude.com) Anthropic has been building toward that office role for months. On February 5, the company said Claude Opus 4.6 would focus on coding, search, finance, and document work, and said Cowork could use those abilities to multitask autonomously on a user’s behalf. (anthropic.com) At the same time, complaints about Claude’s quality have spread across developer and user communities. AOL, citing Fortune’s reporting, said Anthropic was facing backlash over reports that Claude had “regressed” on complex engineering tasks and that users were pressing the company for clearer explanations. (aol.com) Anthropic’s public materials still emphasize performance gains. The company said Sonnet 4.6 became the default model for Free and Pro users on February 17, and said Opus 4.6 led several coding and reasoning benchmarks while keeping the same token pricing. (anthropic.com, anthropic.com) The tension is that Anthropic is asking companies to trust Claude with more of their documents, workflows, and internal know-how while some customers are asking whether the underlying product has changed in ways they cannot see. Anthropic has separately said trustworthy agents need to be “safe, reliable and trustworthy,” language that now reads like a direct test for its own rollout. (venturebeat.com, anthropic.com) For enterprise buyers, the immediate question is no longer just which model writes the best answer. It is whether the company embedding itself into Word, Outlook, SharePoint, and custom team workflows can keep that trust as the software becomes harder to swap out. (support.claude.com, support.claude.com, venturebeat.com)