dbt Labs Says AI Agents Could Finally Make the Semantic Layer Stick

Why the first wave stalled

Mandell described the first wave of the semantic layer, roughly five years ago, as an effort to define metrics as code. The idea was to centralize metric logic in one repository and serve it to every downstream BI tool, so nobody calculated revenue two different ways.

According to Mandell, that wave mostly did not reach production, for three reasons:

• The value delivered was the status quo: the same dashboards, arrived at more carefully.
• Integrating a shared semantic layer into existing BI tools was genuinely hard.
• Semantic modeling, which means thinking in entities, primary keys and join paths, was unfamiliar to analytics engineers whose tools had abstracted all of that away.

Agents change the value, and raise a trust problem

What is different now, Mandell said, is the value on the table. Agents can put "a tireless analyst at the fingertips of anyone in the business," and he said usage of the dbt MCP server is doubling quarter over quarter, driven largely by conversational analytics use cases.

That shift also creates a problem of trust. Mandell used a probability exercise: Steph Curry makes free throws at a 92.3 percent career rate, better than nine of ten, yet his odds of making ten in a row come to only about 45 percent, worse than a coin flip. The same math applies to an agent that answers a business question through a chain of steps, such as finding the right table, the right column and the right aggregation. Even a high per-step success rate degrades quickly across a multi-step chain, and Mandell argued that better models alone will not fix it.

Four conditions for a second wave

Mandell named four things he believes must be true for the second wave to succeed.

• Closing the integration gap. Instead of serving BI tools only through custom connectors, dbt is working on what he called "semantic SQL," a native SQL interface that lets a BI tool query the dbt Semantic Layer as if it were querying a database directly. It would still allow arbitrary SQL joins against ungoverned, ad hoc tables when needed.
• Pushing semantics upstream. Bringing semantics into the transformation layer could help with pre-aggregation and with using metrics as dimensions in customer-360-style models. Mandell said some companies he has spoken with spend 30 to 40 percent of their warehouse bill on materializations to hit latency targets, a cost he believes dbt can help reduce.
• A deterministic spine for a context layer. The semantic layer would sit inside a broader "context layer" that also covers unstructured data such as tickets, docs and calls. Mandell argued agents reason more reliably when core entities and relationships are already defined rather than inferred from raw tables.
• Configurability and access control. Agents should see only the slice of context a given workflow needs, with observability into what data an agent actually touched. Mandell said that combination makes higher trust possible.

Standards, saved queries and open questions

In the Q&A, Mandell said dbt remains an active member of the Open Semantic Interchange (OSI) initiative and can export MetricFlow definitions as an OSI document for use elsewhere. He acknowledged that the ecosystem has not resolved how much fidelity survives when converting between different semantic layer specs.

Asked whether a native SQL interface might replace dbt's saved queries feature, he said no. Saved queries are more about materializing aggregations, which he described as a different problem. He also referenced a public Anthropic post arguing that a semantic layer or knowledge graph should be human-curated rather than AI-bootstrapped, a point he called relevant to how organizations should collaborate on semantics.

What it means for data teams

For teams that tried a headless semantic layer and gave up a few years ago, the pitch is that the payoff is now different: governed definitions that agents can rely on, rather than the same dashboards. Mandell closed by inviting attendees to a same-day roundtable with the dbt product team on two open questions, how a semantic layer should be authored and how it should be served. The discussion also covered what holds teams back from adopting the dbt Semantic Layer or other headless semantic layers.

The session is part of dbt Summit 2026, and dbt says every keynote and breakout is available on demand.