The thing that jumps out to me is not the “Docs” or “Slides” part. It’s Anthropic quietly admitting that the old split between chat and Cowork was a UX tax. If you had to stop and decide which mode you were in before you even asked the model to do anything, that was already a sign the product was fighting itself.
I think the merge is the right move, but also a revealing one. Anthropic is basically saying: the model can now infer enough about intent that the user should not have to pick between “simple answer” and “agentic workflow” up front. That sounds obvious in hindsight, yet lots of AI products still force people into brittle mode switches. Those switches tend to become little landmines. You choose wrong, the output feels off, and then you blame the model.
What I’d actually want to try here is the messy middle: start with a normal question, then push it into research, file work, and drafting without leaving the conversation. If that really works, it’s more useful than having a shiny separate “agent workspace” nobody remembers to enter. The part that matters is not the label on the tab; it’s whether Claude can preserve context while it escalates from chatty helper to do-the-thing assistant.
The new Docs and Slides features sound useful, but I’m a little skeptical of how much of this is genuinely new versus packaging. “Make a document,” “make a deck,” “edit it directly,” “export to PowerPoint or PDF” — that’s all good, but it’s also the exact sort of capability every serious assistant product now claims. The real test will be whether the generated artifacts are actually pleasant to revise, not just decent on first pass. If Claude can produce something that looks okay and also survives human editing without turning into a fight, then that’s meaningful. If not, it’s just another demo that looks better than the workflow feels.
I also noticed the move toward auto-selected tools: web search, file reading, code execution, all chosen by Claude unless you intervene. That’s where the ergonomics and the trust problem collide. Removing the old web toggle makes the experience cleaner, sure, but it also removes a very visible control. Maybe that’s fine. Perhaps most users never wanted the switch anyway. But for people building with Claude, it does mean you’ll care more about how often the model decides to reach for tools on its own, and how explainable those choices are when it goes sideways.
The other interesting detail is that Anthropic kept Manual and Auto as explicit permission modes. That suggests they know people want a spectrum, not just a fully autonomous agent with a glossy name. That’s probably the right instinct. The hype version of agents says “let it run.” The actual developer version is usually “let it run, but not too far.” I’d expect a lot of teams to stay conservative with Auto until they see predictable cost and failure behavior.
And cost is the part I’d watch most closely. The article says long-running agentic tasks consume more of your plan limit than short questions, which is fair, but also a reminder that these products still have a habit of hiding the real price of autonomy behind usage caps. That’s not a complaint unique to Anthropic; it’s the whole category. The more work the model does for you, the more you need to know what it burned along the way.
If anything, this feels like Anthropic trying to make Claude feel less like a chat app with a secret advanced mode and more like one continuous workspace. That’s a better product shape. Whether it becomes a better developer tool will depend on whether the system stays predictable once people stop treating it like a demo and start using it like infrastructure.
Reference: Anthropic、Claudeの「チャット」と「Cowork」を統合 資料作成の「Docs」「Slides」も
