Data, analytics and AI platforms are converging around semantics and agents

Introduction

On August 28, 2026, Snowflake recommended that customers using Cortex Analyst move toward Cortex Agents.

The revealing part was what Snowflake said did not need to change: the semantic views underneath could carry over unchanged. Cortex Agents would use the same governed semantic definitions that had already supported natural-language analytics while adding unstructured retrieval, tool use, conversation context and multi-step orchestration. Snowflake’s transition guidance makes that continuity explicit.

Data platforms, business-intelligence systems and enterprise AI stacks have historically exposed meaning through different consumption layers. Analysts worked through dashboards and semantic models. Data engineers worked through tables, SQL and notebooks. AI teams built retrieval systems, prompts and agents around whatever enterprise context they could assemble.

During 2026, those boundaries have become harder to maintain. Semantic models, metric definitions, ontologies and governed business concepts are increasingly being exposed to both analytical interfaces and AI agents, while agents are moving inside the same platforms that already store, model, govern and query enterprise data.

The emerging converged data analytics AI platform semantics pattern is therefore more significant than adding an AI assistant beside a dashboard. Business meaning is beginning to operate as reusable platform infrastructure across human analysis and machine reasoning.

At a glance

• Semantic objects are increasingly being reused across dashboards, SQL, conversational analytics and AI agents, rather than remaining confined to BI.
• Databricks made Unity Catalog Business Semantics generally available in April 2026 for use across analytical and AI surfaces.
• Google, Microsoft, Snowflake and Databricks are all bringing agent interfaces closer to governed analytical and semantic assets rather than forcing AI teams to rebuild business context separately.
• Convergence does not mean query engines, permissions and data-processing systems have collapsed into one layer.
• Semantic changes now have a larger production blast radius because the same governed definition may influence both conventional analytics and agent-generated answers.

The semantic object is starting to survive the interface change

For years, the semantic layer was closely associated with business intelligence. Its job was to give analysts stable measures, dimensions, joins and terminology so that revenue, margin or active customer did not have to be redefined inside every dashboard.

Natural-language analytics initially added another interface over that environment. Users could type a question instead of constructing a report, but the interaction still looked recognisably like BI.

Agentic systems give the semantic object a broader role. Snowflake’s August transition guidance is useful precisely because it makes the continuity visible: Cortex Analyst relied on semantic views to translate analytical questions into SQL, and Snowflake now recommends Cortex Agents as the broader interaction layer while retaining those semantic views. The agent adds orchestration, unstructured retrieval, tools and conversational context without requiring the business definitions underneath to be rebuilt.

The user interface changes.

The semantic asset survives.

That continuity is one of the clearest signs of convergence. Enterprise meaning is becoming less dependent on a particular analytical interface and more like a reusable contract within the platform.

Analytics artifacts are becoming platform artifacts

Databricks made that shift explicit on April 2, 2026, when it announced the general availability of Unity Catalog Business Semantics. Metric Views can define governed measures, dimensions and semantic metadata centrally and expose those definitions across AI/BI dashboards, Genie, notebooks, SQL applications and other consumers.

Placement matters here. A metric definition created primarily for a dashboard can still be treated as downstream presentation logic. Once dashboards, agents, notebooks and applications all consume the same definition, it becomes a shared platform dependency.

Google is approaching the same problem from the BI side. At Google Cloud Next ’26, the company described Looker as a governed foundation for conversational agents, dashboard agents, downstream actions and a managed MCP server, all grounded in the Looker semantic layer. Google’s agentic BI announcement shows BI semantics being exposed beyond the conventional dashboard experience.

Microsoft’s Fabric IQ, currently in preview, takes a broader contextual approach. Fabric IQ brings semantic models, ontology, graph, data agents and operations agents into a shared business-context architecture. Semantic models provide trusted analytical definitions, while ontology represents business entities and relationships that agents can use when interpreting requests.

These platforms are not implementing the same semantic abstraction. They are, however, moving analytical meaning closer to the centre of the data-and-AI platform.

Agents are becoming another consumer of the analytical layer

The older enterprise AI pattern often began outside the analytics stack. An application retrieved records, schemas or documents, supplied enough context to a model and then recreated business interpretation through prompts, retrieval logic or custom code.

A competing pattern is now visible: let the agent consume a governed analytical surface that already exists.

Google’s Conversational Analytics API became generally available for BigQuery and Looker on June 23, 2026, giving developers a programmatic data-agent interface rather than keeping conversational analytics inside a traditional BI product.

Microsoft Fabric data agents can work across lakehouses, warehouses, Power BI semantic models, KQL databases and ontologies. A published data agent can also be exposed as an MCP server, allowing an external MCP-compatible AI client to call the governed agent rather than building a separate direct-access path into every underlying data source.

Databricks has tied Genie to Unity Catalog Metric Views so that natural-language questions can resolve against centrally defined measures and dimensions rather than requiring the agent to infer the business calculation independently.

Agents are therefore becoming another first-class consumer of analytical meaning. The analytics layer is no longer concerned only with reports consumed by people; it can also provide vocabulary and calculation logic to software reasoning over enterprise data on a user’s behalf.

The interface is converging faster than the execution layer

A “converged data and AI platform” can easily be interpreted as one engine replacing all the layers underneath it. Current product architectures do not support that conclusion.

Microsoft’s ontology integration provides a useful example. Fabric ontology can supply business entities, relationships, definitions and mappings to a data agent, but Microsoft documents ontology as a context source rather than the query-execution endpoint. The agent can use that context to interpret a question and then generate SQL, KQL or DAX for execution by the underlying source.

Snowflake Cortex Agents similarly use semantic views as a governed interpretation layer while the analytical query is still executed against the data platform. Snowflake’s Agentic Analyst integration preserves that separation between agent interaction and analytical execution.

A single conversational experience can therefore depend on several distinct systems: the agent interprets intent, a semantic asset supplies business meaning, the platform identifies the relevant data, an execution engine runs the query, and the requesting identity still needs the appropriate permissions.

The front door can converge while execution remains plural.

That distinction matters operationally because platform convergence does not remove the need to understand where query execution, access control and authoritative state actually reside.

Semantics now have a larger blast radius

A semantic-model change once had a relatively visible downstream effect. An altered measure could change a dashboard, report or scheduled analytical output, and BI teams could usually identify most of those dependencies inside the analytics estate.

The same semantic object can now sit beneath a wider range of experiences. Databricks Metric Views can serve dashboards and Genie. Looker’s semantic layer can support conventional BI and agentic interfaces. Snowflake semantic views remain relevant as users move from Cortex Analyst toward Cortex Agents. Microsoft is connecting semantic models and ontology to conversational and operational agents.

A change to business meaning can therefore influence a dashboard and an agent-generated answer at the same time.

That expands the job of semantic lifecycle management. Versioning, lineage, testing and change review no longer protect only analytical consistency; they increasingly protect the context machine consumers use when interpreting enterprise data.

A definition can remain syntactically valid while changing the meaning of an agent’s answer. Unlike a visibly broken dashboard, the agent may continue responding fluently after the semantic dependency underneath it has shifted. Platform teams therefore need visibility into which human and machine consumers depend on a governed semantic asset before changing it.

The convergence is visible in authoring as well as consumption

Agents are also beginning to participate in the creation and maintenance of semantic assets.

Databricks’ 2026 Business Semantics release includes agent-assisted Metric View authoring, with suggestions for measures, dimensions, synonyms and documentation. The same Business Semantics announcement describes Genie-assisted creation and refinement of those definitions.

Microsoft’s ontology agent, currently in preview, can inspect Fabric workspace sources and propose or apply ontology changes through natural-language interaction. Microsoft separates planning from action so that proposed structures can be inspected before changes are applied.

The resulting feedback loop is new enough to deserve attention. Semantic assets are becoming inputs to AI agents, while agents are beginning to help teams produce those same assets.

Machine assistance can reduce the manual work involved in discovering tables, proposing relationships or documenting concepts. It does not settle whether those concepts accurately represent the business. As semantic authoring becomes more automated, domain ownership becomes more consequential rather than less.

Convergence has not produced one semantic standard

The direction of travel is becoming clearer without producing a single common semantic implementation.

Power BI semantic models, Fabric ontology, Databricks Metric Views, Snowflake semantic views and LookML are different abstractions. Some concentrate on measures and dimensions; others model richer entities and relationships. They also differ in authoring, permissions, execution behavior and lifecycle tooling.

There are signs that vendors want semantic assets to travel farther across interfaces. Databricks says it is open-sourcing the core Metric Views implementation in Apache Spark as part of its Business Semantics work. Google’s Looker direction exposes governed semantics through APIs and MCP-based access. Snowflake semantic views can be addressed as database objects as well as consumed by Cortex services.

These developments broaden where semantic assets can be used, but they do not make the underlying systems interchangeable. For enterprises, platform convergence is not the same as semantic standardisation.

That is an important boundary for architecture decisions. The current shift shows that semantics are becoming shared dependencies across analytics and AI; it does not yet provide one universal model of enterprise meaning.

The operating model has to catch up with the product architecture

Data teams, analytics teams and AI teams have historically owned different parts of enterprise meaning. Data platforms managed schemas and pipelines. Analytics engineering defined measures and dimensions. BI teams maintained semantic models. AI teams frequently recreated context through retrieval, prompts and tool definitions.

A converged platform weakens those boundaries.

If an agent and a dashboard consume the same metric object, maintaining separate “BI logic” and “AI logic” for that definition becomes difficult to justify. If an ontology grounds both analytical questions and operational agents, changing its entities or relationships is no longer metadata housekeeping.

Platform engineering does not need to become the owner of every business definition, but it does need to make shared semantic dependencies governable. Teams need to know who may change them, what consumes them, how changes are tested, which version is active and how an incorrect answer can be traced back to a semantic change.

That is where the platform consequences become most visible. Product architecture is converging faster than many enterprise ownership models.

FAQs

What does convergence of data, analytics and AI platforms mean?

It means capabilities previously separated across data engineering, BI and AI stacks are increasingly sharing platform services and governed objects. A prominent example is semantic definitions being reused across dashboards, natural-language analytics and AI agents instead of being recreated for every interface.

Why are semantic layers important to AI agents?

Agents need more than raw tables and column names to answer business questions consistently. Semantic assets can provide governed measures, dimensions, relationships and terminology. This article focuses on the current platform convergence around those assets rather than the implementation mechanics of semantic-layer architecture.

Are BI platforms turning into AI platforms?

Some BI products are adding agentic capabilities while data-and-AI platforms are adding richer analytical and semantic layers. The result is increasing functional overlap rather than the disappearance of product categories. Execution engines and operational responsibilities can remain distinct even when the user experience converges.

Are data agents replacing dashboards?

No. Dashboards remain useful for repeatable visual monitoring and shared analytical views. Agents add conversational exploration, orchestration and, in some products, action. The more significant development is that both interfaces can increasingly consume the same governed semantic definitions.

Does platform convergence mean enterprises need one semantic layer?

No. Current platforms expose different semantic abstractions, and large enterprises may already depend on several meaning layers. Convergence means semantics are becoming more central to both analytics and AI consumption; it does not prove that one universal semantic model should replace every existing layer.

The lasting platform asset may be the meaning between data and interface

The visible interface is changing quickly. A dashboard can become conversational; conversational analytics can become an agent; an agent can acquire retrieval, tools and the ability to initiate downstream work.

The more durable asset may sit underneath those interfaces.

Snowflake can move users from Analyst toward Agents while retaining semantic views. Databricks can let dashboards and Genie consume the same Metric Views. Google can extend Looker’s governed semantics into agentic BI and external conversational experiences. Microsoft can place semantic models, ontologies and agents inside the same Fabric context architecture.

That is what makes the current convergence more consequential than another round of AI features added to analytics products.

The interface can keep changing.

The harder platform question is whether the business meaning underneath it remains governed, reusable and stable enough for both people and agents to depend on.