Traditional BI tools are being disassembled into three separate concerns — extraction, semantic foundation, and rendering. The 12–18 month window to own this transition is now.
This isn't one market shift. It's two happening simultaneously — and where they cross is the opportunity.
Agentic data engineering is the synthesis of two simultaneous disruptions — not one. Understanding both is what makes the methodology argument commercially precise rather than generic AI hype.
The implication: Every client on Databricks or Fabric is sitting at this intersection — agentic DE arriving from above, BI disruption arriving from below. The engineering discipline that connects them is CE + IE. That is your white space.
The tools that bundled extraction, semantic logic, and rendering together for 20 years are losing that advantage. Each layer is now a separate market — and Impetus plays across all three.
Business logic buried inside Cognos Framework Manager, SAP BO Universes, Tableau calculated fields, and MicroStrategy metrics — extracted by AI reader agents, mapped to a unified business glossary, lineage reconstructed, formulas documented. The KPI inventory that makes Layer 2 possible.
KPIs unlocked in Layer 1 land here — defined once in YAML or MetricFlow as revenue, churn, risk exposure, regulatory capital — and served to any renderer via REST, GraphQL, or MCP. The semantic layer is the durable asset. Metric governance is ongoing, not a one-time project. This is where the client relationship deepens.
CE encodes not just what a KPI is, but what it means to the business — units, grain, filters, edge cases, business calendar rules. IE adds intent: when an agent should trust a metric, when to escalate, what stop rules apply. Without CE + IE at this layer, NL-to-SQL produces confident wrong answers on real enterprise estates.
Natural language interfaces for executives and ad hoc analytics. No dashboard, no analyst in the middle. Prompt → visualisation → insight.
Regulated reporting — BCBS 239, COREP, SOX, IFRS — stays as controlled, certified output. Same semantic foundation underneath. Governance and auditability intact.
BI development is not being eliminated — it's being redirected: from building dashboards to building semantic models that AI surfaces automatically. An enterprise that deployed AI analytics without a governed semantic layer saw adoption collapse to 12% in six months as conflicting metric definitions destroyed user trust. The semantic layer is the most valuable BI investment of the next three years.
This is no longer a roadmap conversation. Clients are being pitched these tools right now.
MCP-connected live dashboards from natural language. Sources: Databricks, Snowflake, BigQuery, Salesforce, Amazon Redshift, ClickHouse. Can export to BI tools. Governance flagged as a concern for enterprise production use.
Chat + dashboard + NL query, fully integrated. Powered by Databricks Metric Views (GA 2026) and Genie Ontology. Agents can now monitor metrics and trigger downstream actions.
NL-to-SQL grounded in YAML Semantic Views (GA 2026). Internal benchmark: >90% SQL accuracy with a governed semantic model. Without it: similar low accuracy as any raw NL-to-SQL approach.
Grounded in LookML semantic layer. Native MCP server for agent access. Agentic Workflows that monitor metrics and surface recommendations autonomously. Published to Gemini Enterprise.
Dedicated semantic layer with deterministic reasoning, aggregate awareness, and governed business definitions. ThoughtSpot now positions itself as an "Agentic Analytics Platform."
"A report author you collaborate with." Deep integration with Power BI semantic models. Generates DAX, builds visuals, explains measures. Strongest in Microsoft-stack enterprises.
Launch OSI (Apache Ossie) — open standard for portable semantic layer definitions. First version of the specification published. No product ships native import/export yet.
Spotter Semantics launched — deterministic reasoning layer, aggregate awareness, governed definitions. Platform rebranded as "Agentic Analytics."
Looker BI Agents with native MCP server. Agentic Workflows monitoring metrics autonomously. Agents draw from up to 5 explores simultaneously, published to Gemini Enterprise.
Databricks Metric Views and Snowflake Semantic Views both reach General Availability. Warehouse-native semantic layers are now production-grade — but not portable between platforms.
Claude Dashboards public beta. MCP-native, connects to Databricks, Snowflake, BigQuery, Salesforce, Redshift, ClickHouse. Generates live, interactive dashboards from natural language.
The term is new. The pattern is accelerating.
Coined by MIT Sloan's Michael Schrage (MIT Sloan Management Review, 2025). The core idea: eliminate the traditional translation layer between business questions and technical analysis. The user describes what they want to an LLM — the tool builds the visualisation, exploration, or app. No SQL. No analyst queue. No waiting three days for the wrong chart.
It's the analytics version of vibe coding: the user's intent becomes the interface.
"A Southeast Asian telecom surfaced more financially relevant insights in 90 minutes than it typically generates in 90 days."
"MCP tools gave agents a standard way to discover schemas and run queries — vibe analytics as MCP-native analytics."
Most vibe analytics demos work on a clean star schema with five tables. Enterprise estates have 50,000 tables, three conflicting definitions of "revenue," and a semantic layer nobody has maintained since 2019. A governed semantic model is the only thing that makes Genie, Cortex Analyst, and Fabric Copilot safe to put in front of an executive. Almost none are. That is the entry point.
Gartner's 2026 Market Guide draws a clear line: not all of BI is going conversational.
Most vendors offer an as-is tool swap — move the dashboards from Tableau to Power BI, same problem, new invoice. The opportunity is the semantic modernisation play: extract the KPIs, govern the foundation, and serve both output lanes from it.
Clients are getting pitched conversational BI by Databricks and Snowflake right now. Some are saying yes before they understand what it takes to make it work on a real estate. When it breaks — and on a raw, ungoverned estate it will — someone gets called in to fix the semantic layer. That call is the entry point. The window to be that someone is 12–18 months before the market learns to ask for it upfront.
The lock-in that Tableau, Power BI, and Qlik rely on lives in their semantic layer. Move that layer out, make it portable, and the BI tool becomes a rendering surface you can swap. OSI is early. The opportunity to own the portability layer is now.
Define metrics once in YAML alongside dbt models. MetricFlow generates SQL for every downstream tool. Flows to Tableau Cloud, Power BI, Looker, Genie via REST, GraphQL, and SQL. Requires paid dbt Starter or Enterprise tier.
GA — Most AdoptedGA in 2026. Integrated with Genie and Unity Catalog governance. Deep Databricks-native. Not portable to Snowflake or other platforms.
GA 2026GA in 2026. YAML-based, powers Cortex Analyst. Snowflake-only. Not portable. LookML carries the same warning — meaningful vendor lock-in.
GA 2026Launched January 2026 by dbt Labs, Snowflake, and Salesforce/Tableau. Targets portable authorship of semantic definitions across platforms. No product ships native import/export yet. The standard exists — the ecosystem is catching up.
Emerging · Jan 2026