Claude Dashboards and the shift from chart builder to chart specifier
Short summary
Anthropic's new Claude Dashboards beta lets users specify dashboards in natural language rather than manually building them, which means BI professionals now need to focus on data modeling and visual communication instead of chart construction. The tool has strong features for auditing calculations but lacks enterprise governance features like row-level security and audit logs that traditional BI platforms provide.
- Claude handles the charting grunt work, so practitioner value shifts to understanding the data models behind dashboards (to verify accuracy) and knowing what to specify visually—skills that require years of experience to develop well.
- The Sources panel and commenting features let users interrogate metrics and calculations directly on the dashboard, improving trust by showing exactly how numbers were derived, though this only works if someone can read the underlying code.
- Claude Dashboards will likely work best as an authoring layer sitting in front of established BI platforms like Power BI or Tableau, since it lacks governance features (row-level security, audit logs, scheduled refresh, version history) that organizations need for sensitive data.
Full summary
Anthropic just released Claude Dashboards as a beta feature, and it marks a real shift in how business intelligence work gets done. Instead of practitioners building charts manually, they're now specifying what they want and letting Claude do the heavy lifting. The author tested it for about an hour using the classic Superstore dataset and found the tool genuinely capable—it generated KPI cards, filters, tooltips, and even a natural language summary that updated when filters changed. But here's the catch: knowing *what* to ask Claude to fix (like removing rounded corners that add visual noise) required years of data visualization experience. Someone without that background would've accepted the default output and called it done.
The real value Claude Dashboards brings is transparency around how numbers get calculated. Two features stand out. First, you can click anywhere on the dashboard and ask Claude to explain a metric or calculation—the answers appear in a chat context so you can ask follow-ups without emailing the person who built it. Second, a Sources panel shows exactly how each dashboard element connects to the underlying data, displaying the actual code Claude used alongside results in a Jupyter-like interface. This matters because trust is everything in business intelligence, and experienced practitioners obsess over showing the wrong number or the right number to the wrong person far more than newcomers do.
The tool has serious gaps if you work in an enterprise. Claude accepted a three-sheet Excel workbook and built a denormalized table from it, which worked fine until the author asked for a proper star schema with separate dimension tables. Claude then spotted data quality issues the flat structure had hidden—Product IDs that weren't unique, Canadian provinces mislabeled as states. That's useful, but it also shows Claude made structural decisions without human oversight. More critically, the beta lacks anything resembling enterprise governance: no row-level security, no scheduled refresh, no audit logs, no version history, no admin controls. Until those arrive, Power BI and Tableau keep their competitive moat, and organizations handling sensitive data should stay cautious about this tool.