If agents are the new primary consumer of your data, is your infrastructure built for the wrong audience?

This week's announcements from dbt Summit 2026 share a single premise: the primary consumer of enterprise data is changing, and the infrastructure built to serve human analysts is increasingly inadequate for AI agentsAI agentsAgentic AI refers to AI systems that pursue goals autonomously by planning, taking actions through tools, and adapting based on results, with minimal step-by-step human direction.View full definition → that query, reason, and act at machine speed. Whether that premise becomes a durable architectural reality or remains a marketing frame depends on decisions CDOs are making right now about metadata, transformation logic, and what "ready" actually means.

What shipped in dbt Core v2 and dbt State?

According to dbt Labs (a vendor with a direct commercial interest in the adoption of these tools), dbt Core v2 and dbt State both reached general availability at dbt Summit 2026. dbt State is the more practically significant of the two for most teams today. It addresses a straightforward and expensive problem: in standard dbt runs, every model rebuilds from scratch even when the underlying data has not changed. dbt State tracks what has changed between runs and rebuilds only what needs to be rebuilt.

dbt Labs claims that Fanatics, the sports merchandise company, used dbt State to cut warehouse compute materially by eliminating unnecessary rebuilds. The company presents this as a compute cost and development speed story. It is worth treating that case study as a directional signal rather than a benchmark: vendor-published case studies select for positive outcomes.

For CDOs, the practical question is not whether dbt State works in principle (incremental computation is well-established) but whether their current dbt estate is structured in a way that makes state tracking reliable. Teams with poorly documented dependencies or inconsistent model tagging will not capture the full benefit without remediation work first.

The dbt v2 release also formalises the direction of travel forindustrialised SQLSQLSales Qualified Lead: a prospect the sales team has validated as ready for direct outreach and a proposal, having passed clear qualification criteria.View full definition → transformation at enterprise scale: the tool is explicitly positioningpositioningThe mental space you want your brand to occupy in your target customer's mind relative to alternatives.View full definition → itself as the definition layer for what data means, separate from where it is stored or computed.

What does Fivetran's Context Layer mean by agent-ready data?

The headline announcement, from Fivetran and dbt Labs jointly (both vendors, both with commercial stakes in this framing), is the Fivetran Context Layer. The claim is that it makes enterprise data "agent-ready" by providing AI agents with structured context about data assets: what a table represents, how it was built, what business rules govern it, and how it relates to other assets.

This matters because agents querying a raw warehouse without that context will produce unreliable outputs. A model that sees a column named "revenue" without knowing whether it is recognised, billed, or collected revenue will hallucinate a definition. The Context Layer is presented as the mechanism that closes that gap.

Take the commercial framing seriously here. Fivetran and dbt Labs have a shared incentive to position their joint stack as the agent-readiness standard. The architectural concept they are describing, giving agents access to rich metadata and semantic context rather than just raw tables, is sound and increasingly discussed across the industry. But the specific implementation is one vendor's answer to the problem, not a settled industry standard.

The more durable insight is aboutactive metadata and what a data catalogdata catalogA centralized inventory of an organization's data assets, enriched with metadata, that helps people find, understand, and trust the data they need.View full definition → must do when agents, not analysts, are the audience. Agents need metadata that is machine-readable, always current, and attached to the data at the point of query, not buried in a separate catalog that humans consult occasionally. CDOs who have not yet addressed that gap should not wait for a single vendor to solve it for them.

dbt Charts and the open lakehouselakehouseA hybrid architecture combining the flexibility of a data lake with the analytical capabilities of a data warehouse, on a single storage layer.View full definition → vision are worth watching, not adopting

dbt Labs also announced dbt Charts, a lightweight visualisation capability, alongside what it describes as an "open lakehouse vision." Both are early-stage claims from a single vendor. dbt Charts addresses a real friction point: teams that define metrics in dbt today still need a separate tool to surface them, and that handoff creates inconsistency. Whether a visualisation layer built inside a transformation tool displaces existing BIBITechnologies and processes that turn raw data into actionable insights via reporting, dashboards and analysis, so teams can decide based on facts rather than intuition.View full definition → investments is a question that will take quarters to answer.

The open lakehouse framing is more strategic than concrete. dbt Labs is signalling that it intends to compete at the storage and architecture layer, not just transformation. CDOs with existing Databricks or Snowflake commitments should note this as a positioning move and watch how it develops, particularly the retirement of the dbt Snowflake Native App announced for November 2026, which is a concrete change requiring attention for any team running that configuration.

Modeling for agents is a tokentokenA token is the basic unit of text that language models process, often a word fragment, whole word, or punctuation mark rather than a single character.View full definition → economy problem

The development most people will underrate this week is a short piece from dbt Labs describing how one team cut token costs by twenty times by modeling Gong call transcripts in the warehouse before feeding them to an AI, rather than pushing raw transcripts to the APIAPIApplication Programming Interface: a standardised interface that lets applications communicate and exchange data without knowing each other's internal workings.View full definition →. The reduction came from restructuring the data so the agent received only what it needed, in a format it could process efficiently.

This is a different design discipline than traditional data modeling. Analysts tolerate wide tables, redundant columns, and verbose schemas because they are good at filtering. Agents are not. They consume tokenstokensA token is the basic unit of text that language models process, often a word fragment, whole word, or punctuation mark rather than a single character.View full definition → and their cost scales with the verbosity of what they receive. The implication for CDOs is that data products built for agent consumption need explicit modeling decisions around token economy, not just semantic clarity. Teams that treat agent-readiness as a metadata labeling exercise will hit this problem when they move from prototype to production workloads.

The O'Reilly Radar framing of "intelligent data orchestration" as the successor to dashboards points in the same direction: the question shifts from "can a human read this?" to "can a machine reason from this reliably and cheaply?"

The practical priority for most CDOs this quarter is not adopting any specific tool from this week's announcements. It is auditing whether their current transformation and metadata layers would give an agent enough context to produce a trustworthy answer, and being honest about the gap between where they are and what that requires.