Agent Horizon

Anthropic announced that Claude Cowork, its separate workspace for longer agentic tasks, is merging into the standard Claude chat interface, so users no longer have to decide upfront whether a request is a quick question or a bigger project. Alongside the merge, Anthropic launched two new beta tools, Claude Docs and Claude Slides, and moved Claude Design directly into the regular conversation view. Details are in Anthropic’s official post.

The rollout starts with Pro and Max subscribers across web, desktop, and mobile over the coming weeks, with other plans to follow and Enterprise admins getting advance notice and control over when the new features activate. Existing chats, projects, artifacts, connectors, and skills carry over, and documents or slide decks built with the new tools can be edited directly, presented from within Claude, or exported as PowerPoint or PDF files.

Why this matters

This is a consolidation move more than a capability leap, and it’s a sensible one. Cowork launched as a separate research preview back in January, and having two different Claude “modes” that users had to consciously pick between was always going to be a source of friction rather than a feature. Merging them mirrors a broader trend across the industry — OpenAI’s Agents API and ChatGPT’s expanding agentic features point the same direction — toward assistants that quietly decide how much autonomy a task needs rather than forcing users to declare it up front.

The addition of Docs and Slides also closes a real gap for Claude compared to rivals: Gemini has deep native ties into Google Workspace, and ChatGPT has had canvas-style document editing for a while, so Claude gaining first-party document and presentation creation makes it more competitive as a daily-driver tool rather than just a chat window you copy-paste out of.

The more interesting question is about defaults, not features. As Claude increasingly decides on its own whether a request should run as a quick reply or a longer background task, users are trusting the model to make that judgment call correctly more often, including on tasks that keep running after they’ve closed their laptop. Anthropic says permission prompts remain the default before Claude takes consequential actions, which is the right caution — but it’s also exactly the kind of default that becomes more important to get right as the system scales toward more autonomous execution.