Analytics tools crossed from answering questions to building workspaces, with semantic layers going native and AI observability reaching its first GA product.
The week in analytics
Analytics shipped two structural moves this week. The first is the semantic layer going native: both Basedash and Lightdash shipped formal semantic layers that the AI can reference by name, not just SQL it generates fresh each time. When the AI has a named definition for "monthly active user," it stops hallucinating metric definitions. That's a meaningful accuracy improvement, not a UI feature.
The second move is AI observability becoming a standalone product category. OpenObserve shipped its v1.0 GA with AI Observability as the headline feature — end-to-end trace and session evaluations for LLM applications. What was a monitoring plugin two years ago is now a GA product line.
Leaders
Basedash shipped two significant releases this week: Models (a semantic layer where teams define SQL-backed business concepts reusable by both UI and AI) and full dashboard generation from chat. The combination is what matters — the AI builds dashboards by referencing named semantic definitions, not ad-hoc SQL. This is the clearest architecture for reliable conversational analytics that any BI tool has shipped.
Lightdash shipped a dbt-free path: native YAML semantic layer with direct GitHub and Bitbucket sync, no dbt intermediary required. Custom metrics and dimensions can be defined and version-controlled without touching dbt. In the same week, Lightdash let teams scaffold data apps locally with Claude Code or Cursor and deploy to Lightdash — a coding agent workflow for BI tool customization.
OpenObserve hit v1.0 GA with AI Observability as its defining new surface: LLM trace and session evaluations, annotation queues for human review, and database monitoring. The jump from v0.92 to v1.0 in a single quarter reflects a product that found its focus — AI workload observability — and shipped to it.
Hex shipped CLI and API access for Hex Agent alongside direct editing of AI-generated app components. The CLI/API move is the more significant one: Hex Agent stops being a browser-only feature and becomes a callable service. Teams can now invoke Hex Agent programmatically and integrate it into their own workflows, not just work inside the Hex UI.
Wildcards
Chord shipped two moves in the same week: persistent team memory across conversations (the AI writes durable context back into its knowledge base mid-conversation) and the ability to build and save CDP audiences mid-conversation in plain language. A CDP tool where the AI can create production audience definitions from a chat session is a qualitatively different product from a query interface.
Themes that compounded
- Semantic layers went from optional integration to native architecture: Basedash's Models and Lightdash's native YAML both ship reusable, AI-accessible metric definitions this week
- AI observability matured from feature to product: OpenObserve v1.0 GA makes AI trace evaluation a first-class offering, not an add-on
- Coding agent workflows entered BI tooling: Lightdash lets teams iterate data apps with Claude Code or Cursor before deploying — a new development pattern for analytics
- Conversational analytics gained write access: Chord's AI can now create CDP audiences and Basedash's chat builds complete dashboards — not just read, but build
Watch this week
Watch dbt Core 2.0.0 reaching release candidate: its Rust-based Fusion parser and full microbatch materialization support will determine how smoothly teams can migrate the dbt pipelines that feed tools like Lightdash and Basedash. A clean 2.0 migration path would accelerate semantic layer adoption across the stack. Also watch whether Tinybird's JSON-by-default change (with Classic plan sunset on September 15) triggers migration friction — it's the kind of forced upgrade that reveals which customers were actually using the product in production.