Claude is becoming more consistent across different parts of its ecosystem.
Anthropic has unified Claude’s memory between Claude Chat and Claude Cowork, allowing information learned during conversations to carry over when users switch to Cowork for actual tasks. The goal is simple: users should no longer have to repeatedly explain the same project, preferences, or background information when moving from discussing something with Claude to asking it to do the work.
The change is more significant than a simple memory upgrade. It moves Claude closer to functioning as one continuous AI assistant, rather than separate experiences that happen to use the same underlying AI.
What Happened to Claude’s Memory?
Previously, Claude’s Chat and Cowork experiences could feel like separate environments.
A user might spend time discussing a project in Claude Chat, develop an idea, provide background information, and then move into Cowork to have Claude actually perform a multi-step task.
That transition could require the user to explain important details again.
With the new system, Claude’s memory is shared between Chat and Cowork. Information Claude has retained from one experience can be available when the user starts working in the other.
This effectively connects the conversation phase and the execution phase.
What Is Claude Cowork?
Cowork is Anthropic’s mode for delegating more complex, multi-step work to Claude.
Rather than simply answering a question, Cowork is designed to work through tasks using a user’s files, connected tools, browser and other resources.
Anthropic describes Cowork as a way to delegate real work to Claude, with workflows that can involve multiple steps and tools.
That makes shared memory particularly useful.
A normal chatbot can benefit from remembering a user’s preferences.
An AI agent can benefit much more from remembering how an ongoing project works.
What Does Shared Memory Actually Change?
The biggest improvement is the transition between planning and execution.
Imagine someone has been discussing an upcoming conference with Claude.
During earlier conversations, the user may have explained:
- The conference location
- Expected attendance
- Speakers
- The type of audience
- The event’s goals
- The preferred writing style
- Previous decisions
Previously, moving into Cowork could mean repeating some of that context.
Now Claude can use relevant saved information when performing the task.
For example, a user could ask Cowork to prepare a conference agenda without having to restate every detail that Claude has already retained.
The result is intended to make Claude feel more like one persistent assistant rather than multiple disconnected tools.
Claude’s Memory Is Also Updating Differently
There is another important change behind the scenes.
Anthropic says Claude can now add topics to its memory as the conversation progresses, rather than waiting until a conversation has ended to summarize what should be remembered.
That matters because memory becomes more useful when it can keep up with an ongoing project.
Suppose a user changes a project deadline halfway through a conversation.
A memory system that updates only after the conversation ends may not immediately reflect the change.
With more continuous memory updates, relevant information can become available across Chat and Cowork sooner.
This makes the memory system more closely resemble an ongoing knowledge layer.
Users Can See and Manage What Claude Remembers
Anthropic is also giving users more control over Claude’s saved memories.
Users can view memories organized by topic and can edit or delete information Claude has retained.
This is important because persistent AI memory creates a new problem:
What happens when the AI remembers something incorrectly?
A memory system is only useful if users can correct it.
If Claude remembers an outdated project requirement, an old preference, or incorrect information, users need a way to change the stored context instead of repeatedly correcting the AI during future conversations.
Topic-based memory controls are designed to address that problem.
What About Sensitive Information?
Privacy is one of the most important parts of the update.
According to reporting on Anthropic’s announcement, Claude does not automatically store personal or sensitive information such as health information, race, ethnicity, religious beliefs, political information and gender identity in memory by default.
Anthropic is also adding an “include sensitive topics in memory” option for users who want Claude to retain this type of information.
Users are notified when a sensitive topic is saved.
Certain information remains excluded from memory entirely, including government-issued identification numbers, Social Security numbers, criminal history and immigration status, according to the company’s stated rules reported by TechCrunch.
This distinction is important.
Claude’s memory is becoming more persistent, but Anthropic is not treating every piece of information a user shares as something that should automatically become permanent memory.
Who Gets the New Memory Feature?
The updated memory system is available across Claude’s consumer experiences.
Reporting indicates that memory is enabled by default for Free, Pro and Max users across web, desktop and mobile. iPhone and Android users need the latest version of the app.
For Team and Enterprise environments, availability and controls can differ depending on organizational settings.
That distinction matters for businesses because persistent memory can involve organizational data and privacy requirements that are different from individual consumer use.
Why Shared Memory Matters for AI Agents
This update becomes more interesting when viewed alongside the broader shift from chatbots toward AI agents.
A chatbot generally answers a question and waits for the next instruction.
An agent is expected to work through a larger objective.
That means an agent benefits from knowing:
- What the user is working on
- What decisions have already been made
- Which preferences matter
- What previous tasks were completed
- What information is relevant to the current assignment
Without memory, every task can begin with a lengthy briefing.
With persistent memory, an AI agent can potentially start much closer to the point where the previous task ended.
That is why connecting Claude Chat and Cowork is more significant than simply making Claude better at remembering conversations.
It connects knowledge with action.
Chat → Cowork Is the Important Workflow
The most useful way to understand the update is as a workflow:
Discuss → Remember → Delegate → Execute
A user can first use Claude Chat to explore an idea.
Claude can retain relevant project information.
The user can then move into Cowork.
Cowork can use the retained context while carrying out the task.
This removes one of the biggest friction points in AI-agent workflows: repeated context transfer.
It also changes how users may interact with AI.
Instead of treating every conversation as a separate session, users can increasingly treat Claude as an ongoing workspace.
What Doesn’t Change?
Shared memory does not mean Claude remembers everything a user has ever said.
Memory is selective, controlled and subject to Anthropic’s stated privacy restrictions.
It also does not mean every piece of conversational context is permanently available to every Claude interaction.
Users still have control over saved memories, and sensitive information has additional restrictions.
This distinction is important because memory is not the same thing as unlimited conversation history.
Memory is intended to retain useful information across interactions, while individual conversations still contain their own immediate context.
How This Could Change Cowork
The biggest potential impact is on longer-running projects.
Consider a marketing campaign.
A user could discuss the brand’s audience, messaging, tone and campaign goals in Claude Chat.
Later, they could ask Cowork to prepare campaign materials, analyze information or complete operational tasks.
If the relevant details are already stored in memory, Cowork can potentially begin with a much better understanding of the assignment.
The same principle could apply to:
- Business planning
- Research
- Content production
- Software projects
- Event planning
- Marketing operations
- Data analysis
- Administrative workflows
Anthropic is already demonstrating Cowork workflows for areas such as marketing operations and business development, including recurring reports, account research and automated workflow tasks.
The Bigger Competition Is About Context
AI companies are increasingly competing on more than model intelligence.
The next competitive layer is context.
A model may be extremely capable, but if it has no understanding of a user’s ongoing work, the user may still spend significant time explaining what needs to be done.
An assistant that understands the user’s projects, preferences and previous decisions can potentially provide much more useful output.
That makes persistent memory an important part of the broader AI-agent race.
The competition is shifting from:
“Who has the smartest chatbot?”
toward:
“Who can build the most useful AI that understands and acts within a user’s ongoing workflow?”
Privacy Becomes More Important as Memory Improves
The same feature that makes AI more useful can also make it more sensitive.
A chatbot that forgets everything after a session has limited knowledge about the user.
An assistant with persistent memory could accumulate information about:
- Projects
- Preferences
- Work habits
- Personal goals
- Professional relationships
- Decisions
- Long-term plans
That makes transparency and user control essential.
Anthropic’s decision to expose stored memories and allow users to edit or delete them is therefore not just a convenience feature.
It is part of the trust model required for persistent AI assistants.
Limitations and Open Questions
The update solves an important problem, but several questions remain.
How Accurate Will Memory Be?
A persistent AI memory can become problematic if it stores outdated or incorrect information.
Users will need to periodically review important memories.
How Selective Is the System?
Not every conversation detail should become permanent memory.
The quality of the selection process will influence how useful the feature becomes.
How Will Businesses Manage Shared Memory?
Enterprise users may require stricter controls around what Claude can retain and how that information is used.
Will Memory Improve Agent Performance?
Shared context should reduce repetitive instructions, but memory alone does not guarantee that Cowork will execute a task correctly.
The quality of the underlying model and its ability to interpret context still matter.
How Much Memory Is Enough?
As AI assistants accumulate more information over time, retrieving the right memory may become more important than simply storing more of it.
Affitronix Analysis
Claude’s new shared memory is easy to dismiss as a quality-of-life update.
It is actually part of a much larger change in how AI assistants are being designed.
The early generation of AI products treated each conversation as a relatively isolated interaction.
The emerging generation is different.
AI systems are becoming persistent, contextual and increasingly capable of taking action.
Claude Chat represents the conversational side of that experience.
Cowork represents the execution side.
Connecting their memories effectively creates a bridge between the two.
That could become especially valuable as Cowork handles more complicated workflows.
The user no longer needs to think:
“What do I need to tell Claude before it can do this?”
Instead, the experience moves toward:
“Claude already understands the project; now I can ask it to work on it.”
That is a meaningful step toward the long-term vision of an AI assistant that follows a user’s work rather than existing only inside individual chat sessions.
But memory also creates a new responsibility for Anthropic.
The more Claude remembers, the more important it becomes to make that memory accurate, understandable, controllable and private.
If Anthropic gets those four things right, shared memory could become one of the most important foundations of Cowork’s usefulness.
What Happens Next?
The next stage will likely be about making memory more useful across increasingly complex workflows.
The key things to watch are:
- How accurately Claude retrieves relevant memories.
- How Cowork uses long-term project context.
- Whether memory improves multi-step task completion.
- How enterprise administrators control persistent context.
- How Anthropic handles privacy and sensitive information.
- Whether users begin treating Claude as a long-term workspace rather than a chatbot.
As Cowork takes on more autonomous work, persistent memory will become increasingly important.
The combination of memory + tools + context + execution is what could turn AI assistants into genuinely persistent digital workers.
Final Takeaway
Claude’s new memory system connects Chat and Cowork, allowing relevant information learned during conversations to carry into task-oriented workflows.
The update also gives users more control over stored memories, including the ability to view, edit and delete information by topic, while Anthropic has added additional controls around sensitive information.
The bigger significance is that Claude is moving closer to a continuous AI assistant.
Instead of repeatedly briefing Claude before every task, users can increasingly discuss a project, build context over time and then ask Cowork to act on that context.
For AI agents, that may be just as important as raw model intelligence.
FAQ
What is the new Claude memory update?
Anthropic has unified Claude’s memory across Chat and Cowork, allowing relevant information retained in one experience to be available in the other.
Can Claude remember information from Chat when using Cowork?
Yes. The shared memory system allows relevant saved information from Claude Chat to be used when working with Claude Cowork.
Can users see what Claude remembers?
Yes. Users can view memories organized by topic and can edit or delete saved information.
Does Claude automatically remember sensitive information?
No. Sensitive topics are excluded from memory by default, according to Anthropic’s stated approach. Users can choose to enable the option to include sensitive topics in memory.
Is Claude memory available on the Free plan?
Yes. The updated memory feature is available for Free, Pro and Max users, with availability across web, desktop and mobile. Mobile users may need to update the Claude app.
What is Claude Cowork?
Claude Cowork is designed for delegating multi-step tasks to Claude, allowing the AI to work with files, tools, browser resources and other connected information.
Why is shared memory important for AI agents?
Shared memory allows an AI agent to retain useful context about ongoing projects and preferences, reducing the need for users to repeatedly explain the same information before starting new tasks.
Can Claude remember everything I tell it?
No. Claude’s memory is selective and subject to Anthropic’s memory and privacy controls. Some sensitive and highly restricted information is excluded from memory.
