Claude's chat and agentic work are moving into one product experience. That is a meaningful interface simplification, but it does not make every Claude workflow identical.
Projects, cloud memory, local files, older task history, usage, and cross-product handoffs still behave differently enough to plan deliberately. This guide explains what Anthropic has documented and what teams should check before changing a working process.
Product behavior and availability were checked September 20, 2026. Anthropic is rolling the experience out over time, so an account may not match every screen or capability described here.
Quick takeaways
- Anthropic now describes Cowork and chat as one Claude experience rather than two separate places to begin work.
- The merged experience removes the need to choose between separate Chat and Cowork starting modes, although the rollout is still account-dependent.
- A single interface does not erase the difference between cloud and local execution. Local-file work still needs the appropriate desktop workflow.
- Claude memory applies to cloud Cowork conversations, while local Cowork does not use memory in the same way.
- Projects can organize conversation and agentic work, but teams still need clear instructions, source boundaries, and approval rules.
- Longer, tool-using tasks can consume more usage than a short chat. “One Claude” is not “one fixed cost per request.”
- Older task history and search can have migration limitations. Confirm the record your team needs before depending on it.
What “one Claude” means
Anthropic's current help documentation says Claude Cowork and chat are one Claude. The product announcement makes the same point: users can keep conversational and agentic work in one Claude experience instead of choosing a separate starting mode each time.
For a business user, the important change is continuity. A quick question can lead into a more involved task without treating the work as a separate product. Conversations, projects, and settings are intended to carry into the updated experience.
The interface is not proof that every request runs the same way, however. A question answered in a cloud conversation and a task that needs access to a local folder still have different execution and governance implications. Treat the merge as a simpler front door—not permission to stop documenting where data lives or which tools a task can use.
Anthropic's rollout is account-dependent. Confirm what the specific account shows instead of writing a team procedure from an announcement alone.
Projects become the practical organizing layer
Claude Projects give related work a durable home. A project can hold context and instructions that should remain attached to one initiative rather than every Claude conversation.
That makes projects useful for recurring work such as a campaign, monthly report, research program, or operating process. A well-scoped project can answer four questions before the next task begins:
- What outcome is this project responsible for?
- Which source files and systems may Claude use?
- Which claims or decisions still require human review?
- What does a useful handoff need to contain?
Avoid turning a project into an unfiltered storage bin. A larger pile of context is not automatically a better brief. Keep the current source of truth easy to identify, retire stale instructions, and separate stable procedures from facts that change frequently.
For administrative examples that can live inside a project, see our guide to using Claude for admin work.
Cloud memory and local work are not the same
The memory distinction is one of the most important parts of the change. Anthropic's memory documentation explains that cloud Cowork conversations can use Claude memory. Local Cowork does not use that memory in the same way.
That creates a simple operational question: does the task need cloud continuity, local file access, or both at different stages?
Cloud work may be the more natural fit when a team wants a conversation to build on approved context already available in Claude. Local execution may be necessary when the task needs files on the computer or a local working directory. Moving between the two should be an explicit handoff, not an assumption that every detail automatically follows.
A reliable handoff package can be small:
- the current objective;
- the approved source files;
- decisions already made;
- ideas or claims that still need verification;
- open questions;
- the expected output format;
- decisions that still need a person’s approval.
This is the same principle behind our separate guide to switching between ChatGPT Work and Claude Cowork without losing context: the receiving system needs a compact, explicit state of the work.
Local files still require the right surface
A merged conversation list does not make a cloud conversation able to reach every local file. If the task needs a folder or file on the computer, use the supported desktop/local workflow and grant only the access the task needs.
Before a local task starts, record:
- the exact folder or files allowed;
- whether access is read-only or includes writes;
- which output location is approved;
- whether deletion, sending, publishing, or other consequential actions are prohibited;
- who reviews the result.
The safest default is narrow access and a reviewable output. “Claude can reach it” is not the same as “Claude should be allowed to change it.”
Check older task history before you rely on search
Anthropic's transition documentation notes limitations involving some older Cowork task history and search. That matters when an old task contains the only copy of a decision, prompt, or output the team expects to reuse.
Before changing a recurring workflow, locate the prior work by name and preserve the important files and decisions outside the conversation if they belong in the business record. Do not use search convenience as the only retention plan.
A simple migration check is enough:
- Identify the projects and older tasks the team still uses.
- Confirm that the important conversations and files are visible.
- Copy stable instructions into the appropriate project or documented procedure.
- Record any missing history instead of silently recreating it from memory.
- Test one low-risk task before moving a consequential workflow.
One product still has different usage patterns
Anthropic distinguishes usage limits from conversation or context limits. Usage can vary with the plan, selected model, effort, tools, context, and the work the system performs.
A short question and a long task that reads files, uses tools, and creates an artifact should not be budgeted as equivalent requests. The interface may now feel unified while the resource demand remains task-dependent.
For a team, the useful unit is not “messages.” It is a completed task the team can actually use:
- How many attempts did it require?
- How much source material did it need?
- Did tools or local files take part?
- How much review and correction remained?
- Did the result meet the team’s quality standard?
Track those questions for a representative week before increasing reliance on a recurring Claude process.
A practical checklist for the merged experience
Use this checklist before converting an existing chat or Cowork habit into a standard team workflow.
1. Confirm the account experience
Check the specific plan, device, and organization settings. Do not assume a rollout screenshot from another account describes yours.
2. Choose the project boundary
Keep each project's purpose, sources, instructions, and approval owner clear. Separate unrelated work even when the same people are involved.
3. Decide cloud versus local
Use cloud continuity where it is appropriate. Use local execution when the task needs local files. Write down the handoff when a process crosses the boundary.
4. Set the memory expectation
Tell users which context may carry forward and which local work will not use Claude memory. Never rely on assumed memory for a required fact.
5. Preserve the business record
Keep approved decisions, source files, and final deliverables in the system of record. A conversation list is helpful working context, not automatically the final archive.
6. Define consequential actions
Sending, publishing, purchasing, deleting, changing permissions, and modifying live systems should each have an explicit authorization rule.
7. Measure usable results
Record attempts, failures, correction time, and the usage or cost visible to the account. Review the process before expanding it.
What this means for the ChatGPT Work comparison
The merged Claude interface changes how users enter Cowork-style tasks, but it does not erase the larger product decision. Teams still need to compare data access, local execution, tools, permissions, administration, limits, and the quality of accepted work.
Our ChatGPT Work vs. Claude Cowork guide remains the place for that broader platform comparison. This page answers the narrower question: how should an existing Claude workflow adapt now that chat and Cowork share one product experience?
The bottom line
“One Claude” is a simpler way to start work, not a reason to collapse every workflow into one undifferentiated process.
Use projects to organize durable context. Keep cloud and local execution explicit. Verify memory instead of assuming it. Preserve older decisions outside the conversation when they matter. Measure usage at the workflow level. And require a person to approve high-impact actions such as sending, publishing, purchasing, or deleting.
