Anthropic is making its Claude artificial intelligence assistant more persistent by introducing a shared memory system across its chat and Cowork experiences.
The update is designed to address one of the most frustrating problems associated with AI assistants: having to repeatedly explain the same project, preferences or background information.
Anthropic said Claude will now be able to carry relevant information between different parts of its ecosystem.
The company is also allowing users to see, edit and delete information retained in memory.
Why memory matters
AI assistants are becoming increasingly capable.
They can write documents.
Analyse information.
Develop software.
Conduct research.
Manage projects.
But their usefulness has often been limited by the need to repeatedly provide context.
A user might spend several hours discussing a project with an AI chatbot and then need to start again when moving to another tool.
Shared memory changes that experience.
One assistant instead of separate tools
Anthropic’s goal is to make Claude feel more like one continuous assistant.
For example, a user could discuss a conference with Claude.
Later, they could ask Cowork to help prepare an agenda.
Instead of starting from scratch, the system could already understand information such as the event’s location, speakers and expected number of attendees.
User control
Memory introduces privacy concerns.
AI systems can potentially retain highly personal information.
Anthropic says users can inspect what Claude has remembered.
They can also edit or delete stored information.
The company says sensitive information is not stored by default, although users can choose to enable some categories.
Why this matters for businesses
Memory could become particularly valuable for professional users.
Businesses often work on long-term projects.
Employees may spend weeks or months interacting with AI systems.
An assistant that remembers previous decisions could become much more useful than one that treats every conversation as new.
Competition in AI assistants
The move also increases competition between AI companies.
AI developers are increasingly trying to make their systems feel like personal digital assistants rather than simple chatbots.
Memory is a major part of that transformation.
A useful assistant needs to understand the user’s context.
Risks of AI memory
There are also risks.
Users need to understand what information is being stored.
Companies must protect that information.
Incorrect memories could also create problems.
If an AI remembers something inaccurately, future responses could be based on false information.
The bigger shift
The development represents a broader change in artificial intelligence.
The first generation of chatbots focused on answering individual questions.
The next generation is moving toward persistent assistants that can understand long-running relationships with users.
That could fundamentally change how people interact with AI.
Instead of repeatedly telling an AI who they are and what they are working on, users may eventually expect their assistant to already know.
