LoginSign Up
← Back to Analytics
Weekly · Analytics · Week of September 21, 2026

AI assistant GA, a dbt-free semantic layer, and the first AI observability 1.0 made analytics the week's most structurally active sector.

ai-interaction-layer-gasemantic-layer-decouplingdocumentation-in-analytics-graphai-observability-emergenceai-crawler-analytics
Y
Generated 1h agoDrawn from 8 products

The week in analytics

Three distinct shifts defined analytics this week. First, AI-native interaction layers formalized: Keboola's Kai hit GA, dbt Core 2.0 added AgentSkill loading, and OpenObserve shipped v1.0 with AI observability as a headline surface. Second, the semantic layer debate advanced: Lightdash shipped GitHub and Bitbucket support for native YAML, removing dbt as a prerequisite. Third, documentation and context are being integrated directly into the analytics graph — Holistics made Markdown files first-class objects linked to models and dashboards.

The sector is bifurcating between tools building AI as a new interaction layer on existing pipelines, and tools rethinking what the underlying graph should include in the first place.

Leaders

Keboola's Kai AI assistant reached general availability for contracted multi-tenant customers. Kai handles data querying, visualization building, and pipeline explanation through conversation — and GA means Keboola is formally committing AI as its primary interaction layer, backed by production SLAs. For a data orchestration platform, this is a significant positioning shift: the product's surface is now the conversation, not the pipeline editor.

dbt Core 2.0 shipped stable with the official OSS/proprietary split formalized: Fusion becomes 'dbt' (proprietary) while 'dbt-oss' becomes the open-source fork. The most directional signal in the release is AgentSkill loading from packages — dbt projects can now surface themselves as agent tools, making data transformation accessible to AI coding agents in the same flow as other agent skills. The naming clarity also matters: teams can now make an explicit choice between the two tracks.

Lightdash connected its native YAML project directly to GitHub and Bitbucket: model changes pull from your branch, custom metrics and dimensions write back as YAML PRs, and an AI-proposed model changes workflow means non-dbt teams get a Git-backed semantic layer without the prerequisite stack. This is a meaningful step toward Lightdash as a standalone semantic layer — not a dbt companion tool.

Holistics made Markdown files first-class analytics objects: frontmatter renders as a Properties panel, and doc links resolve to actual models, datasets, and dashboards. Documentation is now a node in the analytics graph — not a sidebar wiki. The second spark this week from Holistics continued expanding AI governance controls, suggesting a parallel investment in content and compliance.

OpenObserve shipped v1.0 GA with AI Observability as the headline surface: end-to-end monitoring for LLM and agent workloads, trace/session evaluations, annotations, and comparison across test scenarios. SLOs with burn-rate alerting and composite alerts also ship in 1.0. Reaching GA with AI observability as the primary differentiator — not a late addition — sets OpenObserve apart from incumbents retrofitting AI monitoring.

Wildcards

Microsoft Clarity shipped Scrape-to-Referral Insights: the ability to trace whether a specific AI crawler's visit to your site led to a real user session from an AI platform. Measuring the bot-to-human conversion funnel for AI search is a genuinely new analytics category. Microsoft Clarity is the free tool with the broadest site coverage to offer it, which gives it an unusual data density advantage.

TimescaleDB 2.30.0 shipped DeferredChunkAppend, changing last-point query cost from O(n chunks) to O(1). The change is pure execution engine work — no query rewrites required, no schema changes. Most time-series analytics platforms treat last-point queries as a known limitation; TimescaleDB just eliminated it at the executor level.

Themes that compounded

  • AI interaction layers formalized across the analytics stack: Keboola Kai reached GA, dbt Core added AgentSkill loading, and OpenObserve shipped AI observability as a 1.0 surface.
  • Semantic layers are decoupling from dbt: Lightdash's native YAML with Git integration removes the dbt dependency, a pattern that will pressure the broader analytics-as-dbt-companion category.
  • Documentation is entering the analytics graph: Holistics linked Markdown docs to models and dashboards, while Omni Apps reaching GA brought governed data models into embedded analytics products.
  • AI crawler analytics emerged as a new category: Microsoft Clarity closed the loop between bot crawl and human referral, a measurement gap every site owner with SEO exposure faces.

Watch this week

Watch Lightdash's native YAML adoption among non-dbt teams — if it gets traction without dbt as a prerequisite, it signals the semantic layer market is larger and more contested than the current dbt-centric framing suggests. Also watch OpenObserve v1.0 positioning against observability incumbents: launching GA with AI observability front-and-center is a bet that monitoring for LLM workloads will drive platform decisions, not the other way around.