For a long time, using AI tools looked roughly like this: you ask a question, get an answer, copy it into a document, open another program to make a presentation, switch to yet another system for design – and in the end you’re still managing the whole process yourself.
AI helped us with parts of the work, but the work environment itself stayed fragmented.
Anthropic is now trying to change that model. The company announced that Claude Chat and Claude Cowork are merging into one experience. At the same time it launched Claude Docs and Claude Slides, and brought Claude Design directly into conversations.
On the surface it sounds like an interface change. In practice, it’s another step in turning Claude from a chat that answers questions into a system that takes on a task, uses the right tools and returns finished work.
What changed in Claude?
Until now, Cowork was a separate area meant for bigger, more complex projects. Users had to decide in advance whether they wanted a regular conversation or to hand Claude a broader piece of work.
The problem is obvious: in the real world, the line between a question and a task isn’t always sharp.
A conversation can start with a small request like “what changed in sales this week?” and quickly grow into an organised report, a document with recommendations and a presentation for a management meeting.
After the merger, users no longer have to choose in advance where a task belongs. They go to Claude in one place, and Claude is meant to recognise whether it needs a quick answer, longer work, files, Connectors, or a document and a presentation.
The change includes several key capabilities:
- Working on long tasks even after the user has closed their computer
- Using the projects, files, Skills and Connectors already linked to the account
- Creating documents with Claude Docs
- Building presentations with Claude Slides
- Using Claude Design from within the conversation
- Editing the outputs directly inside Claude
- Sharing the outputs with one link and opening them on a phone too
- Downloading presentations as PowerPoint and outputs as PDF
- The option to turn processes into recurring, scheduled tasks
How is it supposed to work in practice?
In Anthropic’s example, a user asks Claude to check what changed in the sales pipeline during the week, write a report in the organisation’s usual style, flag anything that stalled and prepare five slides for a management meeting.
Instead of just getting a text summary, Claude is meant to collect the information, prepare a document, build a presentation from it and keep the two consistent, because they were created from the same conversation and the same context.
If information is missing, Claude can ask. If the user wants to fix a sentence or change a slide, they can do it directly. And if it’s a regular weekly report, it can be scheduled so the process starts every Monday without typing the same request again.
By default, Claude is meant to ask for approval before taking actions. Users who want to give it more independence can let it keep working and stop only when a more significant decision is needed.
The real innovation isn’t Docs or Slides
It’s easy to look at the announcement and think Anthropic simply added an alternative to Google Docs and PowerPoint. But that isn’t the most important point.
The main innovation is the connection between the conversation, the context, the tools and the outputs.
Instead of the user deciding which tool to use at each stage, Claude becomes the management layer of the work. The user defines the goal, and Claude chooses which capabilities to use to get there.
In other words, we’re moving from using AI at the level of a single action to using AI at the level of a process.
Once, we asked it to write a paragraph.
Then we asked it to analyse a file.
Now we’re starting to ask it to complete a whole process that includes gathering information, analysis, writing, design and preparing a final output.
What does this mean for businesses and marketing agencies?
For businesses and agencies, the potential isn’t just in saving a few minutes of writing. The real value is in recurring processes that require moving between several systems.
For example:
- Collecting campaign data and preparing a weekly report
- Spotting drops in performance and flagging issues to check
- Building an executive summary from several data sources
- Preparing a client presentation based on the report
- Turning long research into an organised document and slides
- Preparing a content plan and passing the brief to design
- Creating documentation, procedures and meeting summaries
The big advantage is that all the outputs are created from the same context. If the presentation is built from the report, it doesn’t start from scratch and shouldn’t lose the data, the decisions or the writing style along the way.
Can you let Claude work on its own?
Not completely, and certainly not on every task.
A system that can read information, use tools and take actions can save a lot of time, but it can also quickly take the wrong action. So the question isn’t only what Claude can do, but what it makes sense to let it do without approval.
You can let AI prepare a draft report, summarise data, suggest a presentation or point out anomalies. Publishing content, changing an ad budget, sending a message to a client or making a business decision, on the other hand, usually require human review and approval.
The quality of the result also depends on the quality of the context. Claude can’t know the business’s procedures, metrics and preferences unless they are defined, documented or connected to it.
So moving to work with AI agents doesn’t remove the need for management. It actually makes good management more important.
The next skill: not writing prompts, but managing work
In recent years there’s been a lot of talk about prompt engineering – the ability to phrase the right request to get a better answer.
When AI tools start carrying out whole processes, a broader skill is needed:
- Defining a clear goal
- Providing the right context
- Deciding which information sources are reliable
- Setting limits and permissions
- Deciding when approval is required
- Checking the result against business metrics
The employee or manager is no longer just the person who asks AI to write something. They become the person who manages a system that can do work.
Who is it available to?
According to Anthropic, the merger of Claude Chat and Cowork is rolling out gradually to Pro and Max subscribers on web, desktop and mobile. Team and Free plans are expected to get the change later. Claude Docs, Claude Slides and Claude Design are in beta on paid plans, and on Enterprise accounts admins will be able to choose when to turn them on.
As with any gradual rollout, the options may not appear in every account right away.
The bottom line
Anthropic’s announcement isn’t just about merging two products. It reflects a bigger change in how AI companies see their role.
The goal is no longer just to build a smarter chat. The goal is to build a place where the user hands over a goal and the system carries out the sequence of work needed.
Claude still isn’t an independent employee you can hand the business to and forget about. But the direction is clear: less switching between tools, less manual copying and more delegating whole processes to one system.
The important question is no longer just “what can we ask AI?”
The new question is: “What work are we willing to let it manage?”
Source: Claude Cowork and chat are now one Claude, Anthropic, 16 September 2026.