Claude Cowork and Chat Merge: Docs, Slides, Design, and Background Tasks

Anthropic is merging Claude Cowork and chat into one Claude experience beginning September 16, 2026. Instead of deciding whether a request belongs in ordinary chat or a separate agent workspace, users can ask for the outcome in one conversation and Claude can determine whether the task needs research, connected apps, files, code, background work, or a finished document.

The launch also introduces Claude Docs and Claude Slides, while bringing Claude Design directly into conversations. A single task can produce a living document, a presentation, and visual material that share the same context. More involved cloud tasks can continue after the user closes the laptop, and recurring work can be scheduled.

This is more than a navigation change. It moves Claude closer to a general work interface where a quick answer, a long-running assignment, and the files produced from that assignment all begin in the same place.

Source note: This guide reflects Anthropic’s official product announcement and Help Center documentation available on September 16, 2026. The rollout is gradual, capabilities remain plan- and platform-dependent, and Docs, Slides, Design, and computer use include beta features. Verify current availability and administrator controls before adopting the workflow.

Claude Cowork and chat merge at a glance

Existing users are not expected to migrate projects manually. Anthropic says chats, tasks, projects, artifacts, connectors, skills, and settings carry into the unified experience. Once an account receives the new interface, it cannot switch back to separate Chat and Cowork options.

Why Anthropic removed the separate Cowork decision

Cowork was designed for larger assignments and Claude Design for visual work. Anthropic says users found it frustrating to decide where a task belonged and discovered that context did not always carry naturally between separate surfaces.

The new interface changes the first step. A user describes the intended result, and Claude chooses an appropriate execution path. A short factual question can still receive a short answer. A larger request can become a multi-step task involving research, connected data, calculations, code, files, and review.

That is a better mental model for business work because many assignments change shape while they are being completed. A request that begins as “What changed in the pipeline?” may become a written report, a leadership deck, a follow-up list, and a scheduled Monday workflow.

The important product question is no longer which mode should I open? It is what outcome, sources, boundaries, and approval points should I define?

What Claude can now do in one conversation

Handle a complete assignment

Anthropic says Claude can search the web, read files, run code, use connected apps, and combine the results without requiring the user to break the assignment into every individual step.

This can help with work such as:

• synthesizing interview notes into themes and supporting quotations;
• researching options and producing a sourced comparison;
• reviewing a set of business files and creating an executive summary;
• analyzing operational data and returning a spreadsheet with working formulas;
• creating a report and turning its findings into presentation slides.

A clear outcome still matters. Delegating the steps does not remove the need to define the audience, approved sources, sensitive data rules, acceptance criteria, and final reviewer.

Continue cloud work in the background

More involved tasks can keep running in the cloud when the user closes the browser or laptop. The result remains in the conversation for later review. Several tasks can run at the same time.

Anthropic distinguishes cloud execution from local computer access. If the assignment needs files stored only on the computer or must operate desktop applications, Claude Desktop needs to remain open.

Teams should make this difference visible in operating procedures. A background task should not silently depend on an employee’s laptop, local session, expiring login, or mounted folder without a clear status and recovery path.

Use existing connectors and skills

Connected apps can provide the context needed to complete a task. Anthropic lists examples including Google Drive, Gmail, Microsoft 365, and Slack. Existing skills and project context also carry into the unified experience.

Connectors create value because they reduce manual copy-and-paste work. They also expand the data and actions available to the assistant. Every connector should be reviewed for its access scope, account identity, read/write capabilities, retention, and audit visibility.

Claude Docs, Slides, and Design

The unified conversation is also becoming a production surface for common business deliverables.

Claude Docs

Claude Docs creates living documents that a user and Claude can edit together. A document can begin from the research and context already in the conversation rather than from an exported answer pasted into another application.

Potential uses include project briefs, operating procedures, client reports, research summaries, launch plans, and internal policies. The user can revise content directly, ask Claude for changes, and share the result through one link.

Claude Slides

Claude Slides creates presentations from the same conversation and evidence. Users can edit a slide, comment on it, present directly, share it, or download it as a PowerPoint file.

This reduces the handoff between analysis and presentation, but the deck still needs human review. Check figures, citations, chart scales, dates, brand requirements, accessibility, speaker context, and whether the slides accurately reflect the underlying report.

Claude Design

Claude Design now works within conversations while remaining available as a standalone product. It can produce visuals and mockups, including assets guided by an organization’s design system where configured.

Anthropic says Docs, Slides, and Design are available in beta on paid plans. Enterprise administrators decide when to enable them. Beta status should be part of any rollout decision involving client-facing, regulated, or business-critical materials.

A practical example: the weekly leadership report

Anthropic describes a workflow in which a user asks Claude to review pipeline changes, write the report in the organization’s usual style, flag slipped items, and produce five leadership slides.

A production version of that workflow could follow this sequence:

1. retrieve approved pipeline data and the previous report template;
2. validate the reporting period, owners, and data freshness;
3. calculate changes using reproducible rules;
4. identify exceptions and missing data;
5. draft the written report with links to evidence;
6. generate slides that match the reviewed report;
7. pause for a named reviewer before sharing;
8. record what was approved and distributed;
9. schedule the next run with the same controls.

The useful automation is not simply “make slides every Monday.” It is a repeatable reporting system with traceable inputs, exception handling, review, and delivery confirmation.

What the rollout means for each Claude plan

Anthropic says the unified experience is rolling out over the coming weeks to existing and new Pro and Max users on web, desktop, and mobile. Accounts on the same plan may receive it at different times, and there is nothing to enable manually.

Team and Free plans are expected to follow. Anthropic has not provided one universal activation date for every account.

For Enterprise, Anthropic says administrators will receive at least 30 days’ notice before the experience changes. Admins will control whether Docs, Slides, and Design are enabled for the organization.

If a Pro or Max account still shows separate Chat and Cowork choices, the rollout has not reached that account. Users can continue with the existing surfaces until the change arrives.

Safety and governance for a unified work agent

Removing the mode decision makes the product simpler. It can also make the boundary between answering and acting less obvious. Teams should define that boundary explicitly.

Keep permissions narrow

Use the least-privileged connector and account for each workflow. A task that needs to read a report folder should not automatically gain access to every drive, mailbox, workspace, or customer record.

Separate reading from changing

Researching a pipeline and updating the pipeline are different authorities. Reading an email and sending a reply are different authorities. Creating a draft and sharing it externally are different authorities. Require explicit approval when a workflow crosses from analysis into consequential action.

Verify downstream results

A tool accepting a request does not prove the final result. Confirm that the file exists, the calculation is correct, the message reached the intended destination, or the scheduled task ran with the expected data.

Defend against untrusted instructions

Web pages, emails, documents, comments, and connector results can contain misleading or malicious instructions. Treat retrieved content as data, not as authority to disclose information, expand permissions, install software, or contact someone.

Preserve review and recovery

Keep source links, task history, file versions, approvals, and a way to stop or reverse recurring work. A scheduled agent should stay quiet when nothing meaningful changed and notify a person when it fails, needs authorization, or finds an exception.

How to evaluate the new Claude experience

1. Choose one recurring deliverable

Start with a report, research summary, project brief, spreadsheet, or presentation that has clear inputs and an identifiable reviewer.

2. Create an acceptance checklist

Define required sections, data sources, calculations, citations, brand rules, privacy boundaries, and delivery format. Include examples of unacceptable output.

3. Begin with read-only access

Let Claude retrieve and draft before granting it permission to edit external systems, send messages, reorganize folders, or run scheduled actions.

4. Test interruption and failure

Disconnect a connector, remove a file, introduce stale data, close the laptop, and reject an approval. Confirm that the task reports what failed and does not invent a successful result.

5. Measure the completed outcome

Compare preparation time, reviewer corrections, factual errors, missed exceptions, delivery reliability, and total usage. The goal is an accepted deliverable—not the number of autonomous steps Claude performed.

6. Schedule only after the manual flow is reliable

Recurring automation multiplies both value and mistakes. Schedule the work only after inputs, approvals, notifications, and rollback behavior are dependable.

How I can help build a reliable Claude workflow

I provide workflow automation services for organizations turning reports, research, approvals, connected apps, and recurring operations into reliable AI-assisted systems.

I also provide AI consulting and custom AI development, SaaS product engineering, website development, and mobile app development. The XReporter operations and reporting system is an example of combining structured information, accountable workflows, and usable reporting.

Book a free strategy call to identify one Claude workflow worth automating, define its permissions and review points, and test it against a measurable business outcome.

Official sources

• Claude: Claude Cowork and chat are now one Claude
• Claude Help Center: Claude Cowork and chat are one Claude