The Deal That Defines the Agentic Data Stack
When Fivetran and dbt Labs announced their all-stock merger in October 2025, the thesis was straightforward: the two most widely adopted tools for moving data into warehouses and transforming it for use were more valuable together than apart, particularly as AI agents began to consume data infrastructure in ways that the original designs of both products were not built to handle. The merger completed on June 1, 2026, with George Fraser continuing as CEO and Tristan Handy, dbt Labs' founder, serving as President. The combined company now supports more than 100,000 data teams globally, including customers such as OpenAI, Zendesk, Coupa, and HubSpot.
The timing of the completion is deliberate. The data engineering market is at an inflection point driven by agentic AI. Databricks reported this week that over 80 percent of new databases launched on its platform are now created by AI agents rather than human engineers — a statistic that illustrates how rapidly the consumption model for data infrastructure is changing. When agents are the primary consumers of data pipelines and transformation logic, the requirements for reliability, governance, and context clarity change fundamentally. Data that works for a human analyst querying a dashboard does not necessarily work for an agent that needs to reason across it and take autonomous actions.
Agents Schema: An Open Standard for Agent Context
The most significant product announcement accompanying the merger is Agents Schema, an open-source standard that Fivetran and dbt Labs are releasing as a foundation for the industry. The concept is architecturally elegant in its simplicity: designate a single schema within a data warehouse as the shared context layer for all AI agents operating within an organisation. Every agent — whether it is a customer service bot, a financial reconciliation agent, or a code review assistant — reads from and writes to this governed schema rather than each maintaining its own data context.
The implications for data governance are significant. One of the most persistent problems in early enterprise AI deployments has been context fragmentation: different agents operating on different versions of the same data, leading to inconsistent outputs and, in production environments, conflicting actions. Agents Schema provides a structural answer to this problem by giving organisations a customer-owned, warehouse-native context layer that is not locked to any specific agent framework or AI provider. The Apache 2.0 licensing means any data team can adopt the standard without commercial dependency on Fivetran or dbt Labs.
dbt Core v2.0 and the Open-Source Commitment
Alongside Agents Schema, the merger marks the release of dbt Core v2.0, which open-sources the dbt Fusion engine runtime under Apache 2.0. The Fusion engine, which powers dbt's transformation compilation and execution, was previously a proprietary component of dbt Labs' commercial offering. Making it open source is a significant strategic commitment — one that expands the community of developers who can build on dbt's transformation layer and signals that the combined company sees its commercial differentiation in the managed platform and enterprise governance features rather than in the core transformation runtime.
The open-source move is also a competitive positioning statement against cloud-native competitors including Databricks' own transformation tooling and Snowflake's expanding analytics engineering capabilities. By deepening the open-source foundation, Fivetran and dbt Labs are betting that the ecosystem effects of a widely adopted standard will outweigh the short-term revenue that proprietary core tooling might generate. It is the same bet that drove the success of Apache Kafka, Apache Spark, and other foundational open-source data technologies — and the evidence from those precedents suggests it is the right call for a category-defining platform.
The Combined Platform and What It Solves
Prior to the merger, organisations using both Fivetran and dbt were managing two separate vendor relationships, two separate data lineage models, and two separate points of failure in their pipeline-to-transformation workflow. The combined platform addresses this by providing end-to-end visibility from data source through pipeline through transformation to agent consumption. Data lineage that previously required custom tooling or third-party metadata platforms to trace across the Fivetran-dbt boundary is now native to the combined product.
For data engineering teams managing AI agent deployments, the governance implications are practical and immediate. When an agent produces an unexpected output, the question of which data it consumed, at what freshness, through which transformation logic, is now answerable within a single platform. That auditability is not a nice-to-have in regulated industries — it is a compliance requirement that many data teams have been unable to satisfy with the tooling available before this merger.
Market Context: The Consolidation Continues
The Fivetran-dbt merger is one of several significant consolidations reshaping the data infrastructure market in 2026. Google Cloud's announcement of BigQuery Graph, now in preview, extends its analytics platform into graph analytics for large-scale relationship modeling — a capability that has significant implications for recommendation systems, fraud detection, and knowledge graph applications. The Managed Service for Apache Airflow on Google Cloud has shipped Airflow 3.1 with AI-powered agentic troubleshooting, positioning the orchestration layer as an AI-aware system rather than a static scheduler.
The pattern across these moves is consistent: the major platforms are building agentic capabilities into the data infrastructure layer rather than leaving that work to the application layer. The data stack is becoming the AI stack. Organisations that invest in understanding and managing this convergence — through choices about platforms, governance frameworks, and the data quality disciplines that make agentic AI reliable — are the ones that will extract value from the 2026 generation of AI tooling. Those that treat data infrastructure as a legacy concern, separate from AI strategy, are building on a foundation that will limit them.
What Data and Engineering Leaders Should Do
For data engineering leaders, the Fivetran-dbt merger and the launch of Agents Schema create a decision point. The Agents Schema standard is an open-source option that can be adopted independently of the combined platform — any organisation running dbt transformations against a cloud data warehouse can begin designating a governed agent context layer immediately. That is worth evaluating now, regardless of whether the Fivetran-dbt commercial platform is the right long-term choice.
The broader strategic implication is that the data engineering roadmap needs to be developed in close coordination with the AI strategy. The questions of which agents will operate in your environment, what data they need, at what freshness and quality thresholds, and with what audit trail, are data engineering questions as much as they are AI product questions. The organisations that build that coordination into their operating model — rather than managing it as an interface between separate teams — will be the ones that can actually deliver on the agentic AI potential that 97 percent of enterprises say they are exploring.
