Claude Dashboards: AI-powered dashboards for SMBs

Anthropic has made its move in the field of AI-assisted analytics. The company has just released Claude Dashboards in beta, a feature that allows building live dashboards on enterprise data sources and, moreover, seeing the exact query behind each number. The news comes just a month after OpenAI introduced its data agent within ChatGPT Work, confirming that the race to colonize the enterprise productivity layer is in full swing.

What exactly Claude Dashboards does

The operation is simple to describe: the user connects their data platforms (Snowflake, Databricks, Amazon Redshift, or ClickHouse, among others) and describes in natural language what they want to see. Claude generates the dashboard. By clicking on any figure, the SQL query that produced it appears; and if the user doesn't know how to read SQL, they can ask Claude to explain it. Existing connectors, such as Salesforce, also work.

Anthropic insists that dashboards refresh as data changes, with a timestamp on each chart indicating when it was last updated. This is not a minor detail: in an SMB environment, where many decisions are made with last week's data, knowing when an indicator was calculated avoids pointless discussions.

It doesn't replace your BI tool (yet)

It's worth reading the nuance carefully: Anthropic does not position Dashboards as a replacement for specialized business intelligence tools. Its feature set does not cover what a well-set-up Looker, Tableau, or Power BI does. What it does allow is exporting the dashboard to services like Grafana, Hex, Mixpanel, monday.com, Omni, PostHog, or Sigma. Support for Looker, Perplexity, and Tableau will come later.

The takeaway for an SMB is clear: if you already have a consolidated BI tool, Claude Dashboards can serve as a layer for quick exploration and prototyping, not as a substitute. If you have nothing, it's a reasonable way to start looking at your data without setting up a months-long project.

Permissions: here's the important conversation

The point that interests us most from a risk perspective is how access is managed. According to Anthropic, a dashboard starts out private to its creator. When shared, each viewer runs queries with their own connections, so they only see the data they already had access to. In Enterprise plans, the feature is disabled by default until an administrator enables it in the organization settings. And in Team and Enterprise plans, dashboards remain within the organization unless the owner explicitly enables external sharing.

This permission-inheritance design is sensible, but it directly connects to a pattern we have already analyzed in this blog: when an AI agent acts on behalf of a user, it inherits their credentials and, with them, their exposure surface. That the dashboard respects the table, row, and column permissions of the connected account is the minimum condition; the question an IT manager should ask is what happens if that account has more permissions than it should.

Claude Motion and the rest of the package

Alongside Dashboards, Anthropic has launched Claude Motion, a feature that turns material such as a quarterly report or a board chart into a short animated clip. The curious part is the approach: since Anthropic doesn't have a video model, Claude generates code that animates the text, charts, and images. That makes it more flexible and editable, and allows exporting to MP4. It is in beta for Team and Enterprise customers.

Additionally, Docs, Slides, and Design are leaving beta and will be enabled by default in Enterprise organizations. Anthropic claims that more than 45 million documents, presentations, and designs have been created since their arrival. Artifacts now support customer-managed encryption keys (CMEK), and administrators can control which templates users see. Users of the standalone Claude Design site have until December 14 to migrate, and their chats and comments will not be transferred.

What we recommend from ForgeNEX

1. Don't enable it recklessly. If you're on Enterprise, the feature comes turned off. Before turning it on, review which service accounts and connectors you have registered and with what permissions.
2. Start with a scoped use case. A sales or support dashboard with data that is already in your warehouse. Don't expose the entire ERP on day one.
3. Demand traceability. The ability to see the query behind each figure is the best auditing feature the tool brings. Use it.
4. Define an external sharing policy. Just because the option exists doesn't mean you should allow it. Decide it in writing.
5. Don't ditch your BI. If you already have Tableau or Power BI, this is an exploration add-on, not a replacement.

The move by Anthropic and OpenAI in the same direction confirms a trend: model providers want to move up the enterprise productivity stack and stop being mere token providers. For a Seville-based SMB, the opportunity is real, but so is the risk of giving broad access to sensitive data without first sorting out permissions. As always, technology arrives before policies.

Source: The New Stack. Analysis and adaptation: ForgeNEX.