LoginSign Up
← Back to all sparks
L

Lightdash

ANALYTICS
Velocity7.5

Lightdash is handing the analyst's job to agents and keeping the semantic layer as referee.

agentic analyticssemantic layerdata appscontent as codemcpbi governance
◆Current state
Lightdash has spent the last two months rebuilding around agents rather than around its own web editor. Data apps can be scaffolded and iterated locally with Cursor, Claude Code or Codex and uploaded for the instance to build; chart types can be generated from a prompt; verified content and AI agent answers now share one store that the Lightdash MCP serves to outside tools. The conventional BI surface is still being maintained — SQL Runner big numbers, filter groups, timezone handling — but it is no longer where the new capability lands.
◆Where it's heading
The pattern is a deliberate split: authoring and interrogation move outward to whatever agent the user already runs, while the governed metrics, permissions and build stay inside Lightdash. Deep Research extends that from generating artifacts to conducting analysis — exploring data, testing competing explanations, validating numbers. Content as code now covers charts, dashboards, permissions, automations, users and roles, which makes the whole instance addressable by an agent through a repository rather than a UI.
◆Prediction
Expect the next releases to make agents first-class operators of the instance itself — driving the content-as-code surface to refactor resources and access, and extending Deep Research from answering questions to monitoring for the anomalies it currently only explains.

◆Recent moves

  1. 5d ago

    Deep research

    ⚡ SPARK

    Deep Research is the analysis half of the agent thesis Lightdash has been building toward: having moved authoring out to external coding agents, it now moves investigation to an in-product agent that reasons over the governed layer. It targets root-cause and anomaly work rather than chart production.

    View source ↗
  2. 14d ago

    🤖 Build data apps locally with your favorite agent

    ⚡ SPARK

    This is the release that set the current direction: authoring left the browser entirely, and Lightdash kept the build and the governed data. Everything since, including Deep Research, reads as the same bet applied to a different part of the workflow.

    View source ↗
  3. 18d ago

    📦 More content as code

    Content as code widened from charts and dashboards to permissions, automations, virtual views, AI agents, users, groups and custom roles. That makes the instance configuration reviewable in pull requests and, more to the point of the current arc, editable by an agent working in a repository rather than a UI.

    View source ↗
  4. 19d ago

    SQL Runner: Big Number

    Big Number charts in SQL Runner close a gap against the standard BI feature set — one headline value with an optional label, comparison or trend. Maintenance of the conventional surface continues alongside the agent work rather than being displaced by it.

    View source ↗
  5. 22d ago

    🎯 Ask for one filter, not every filter

    Requiring one filter from a group, rather than all of them, is a targeted fix for slow dashboards that previously forced viewers through every selector before anything rendered. The note field lets the builder explain the constraint instead of leaving viewers guessing.

    View source ↗
  6. 1mo ago

    🌍 Timezones that just work

    Project-level timezone handling now applies across filters, date grouping and displayed timestamps, with per-viewer and pinned-zone options. Rolled out to all Cloud organizations with UTC preserved as the default, so it lands as correctness work rather than a migration.

    View source ↗