Claude Sheets and Whiteboards: AI moves into shared work

With Sheets, Whiteboards, Claude Data and its connectors, Anthropic is moving Claude beyond chat and into shared documents and team tasks. Here’s what these tools can do, and what remains to be verified.

A shared spreadsheet, a meeting whiteboard and dashboards connected to company data: Anthropic wants to make Claude a space where people create and share work, not just an assistant they ask questions. The difference lies less in a new conversation than in the document it produces, the connected applications and the access rules governing these operations.

From dialogue to working document

With Claude Sheets, several people can create and edit a spreadsheet with Claude, organise data, write or adjust formulas and work across multiple tabs. Whiteboards applies the same approach to collaborative preparation: the team has a space to organise ideas, then can ask the assistant to produce a summary. These tools have been announced in beta, meaning they are still in development, and are available on paid plans.

The change is tangible: AI no longer produces text only in a chat window. It works within an object that colleagues can take over, add to and share. Anthropic also describes context being carried between these formats: an idea developed on a whiteboard could be reused in a spreadsheet or dashboard without the user having to explain everything again. This may reduce repetition, but the quality of the result still depends on the information available and how the team has organised it.

This approach extends features Anthropic already offers for creating documents and presentations. It brings Claude closer to a work suite, while remaining distinct from a complete replacement for established office tools. File formats and available features matter as much as the generation itself: for advanced spreadsheet needs, the company points users in particular to exporting to Excel or CSV, or to Claude for Excel.

Connectors bring Claude closer to applications

Another part of the announcement concerns interactive connectors. They allow certain third-party applications to be integrated into Claude’s interface, so tasks can be carried out without switching between tabs. Anthropic cites Adobe and Atlassian in particular. The company also says that extensions for Salesforce, NetSuite and Workday combine tools with predefined functions, for example to prepare a sales call or analyse financial data.

A connector is not proof that the assistant fully understands a business system. It provides defined access to data or actions, depending on the permissions granted. Anthropic says Claude operates within the permissions already assigned to the user and that administrators can limit a connector’s capabilities: allowing it to read a space, for example, without permitting it to send messages. For an organisation, this distinction determines whether AI is merely consulting or preparing information, or actually taking action.

Context is central. Information useful for teamwork is often scattered across emails, files and applications. Connectors, built in particular around the MCP protocol, are designed to link Claude to these environments. Anthropic says its marketplace brings together more than 9 000 connectors and extensions, including community-created ones. This total describes a catalogue claimed by the publisher; it does not measure their quality or relevance to any given company.

From dashboards to meetings

Anthropic also presents Claude Data, the new name for its Dashboards tool, with connections to services such as Snowflake and Salesforce. The company says it can draw on definitions certified by data teams to produce dashboards, visual narratives and short videos with commentary. These definitions play an important role: a metric can mean different things depending on the period, scope or the way the company calculates it. Generating a presentation therefore does not replace the work of setting and validating these reference points.

Claude Meetings, meanwhile, has been announced as a forthcoming feature. The assistant is expected to be able to join a video conference to take notes and update team documents. Anthropic has not given a launch date. This is therefore an announcement, not a feature whose conditions of use or results can already be assessed.

The examples put forward by the publisher show the tasks being targeted: preparing briefing packs, assembling forecast presentations, turning negotiated clauses into contracts or drafting follow-ups after a call. Anthropic reports that the finance team at its client Advantage Solutions cut the time taken to assemble materials, comments and checks for a forecasting cycle from nearly ten hours to under thirty minutes. This is an account provided by the company, not an independent measurement applicable to all teams. The time saved depends in particular on the work included in the calculation and the level of human oversight retained.

Collaboration does not remove responsibility

When several people and an assistant edit the same document, it is important to know who can access the data, what actions are permitted and who checks the result. Anthropic says administrators on Enterprise plans can manage access to products and connectors, set spending limits and consult audit logs. The company also says that Enterprise data is encrypted in transit and at rest. These mechanisms are controls announced by the provider; each organisation still needs to assess them against its own security and compliance requirements.

The arrangements vary by subscription: the Sheets and Whiteboards beta is limited to paid plans, the features are enabled by default in Team, while an Enterprise administrator must enable them. Access rules and deployment therefore vary according to the organisation’s setup, and a feature being available does not mean it is authorised for all employees.

The challenge for users is to distinguish three operations that are often confused: producing a draft, editing a shared document and taking an action in a connected service. The first mainly requires checking the content; the other two raise additional questions about versions, permissions and responsibility. Claude can make it easier to turn an idea into a shared resource. The reliability of the work still depends on the data selected, the access granted and checks by the people who know the subject.