The fourth annual State of Analytics Engineering Report from dbt Labs landed this week with a number that should reframe the entire agentic AI conversation: 83% of data professionals now say increasing trust in data is their top organizational priority. Last year, it was 66%.
Trust didn’t just climb the priority list. It overtook speed.
That shift happened while AI-assisted coding became the dominant workflow — 72% of respondents prioritize it. But AI-assisted pipeline management, the testing and observability layer that catches errors before they reach production, sits at 24%. Three times as many organizations are accelerating their data work as are governing it.
The result is predictable. 71% of data professionals cite incorrect or hallucinated outputs reaching stakeholders as a top concern. Not a theoretical risk. A thing that is happening now, at scale, as autonomous agents operate on top of organizational data.
The Governance Design Problem — Again
This is the same structural pattern the enforcement stack has been exposing for months. Organizations deploy first, then discover the design layer is missing.
In the security stack, it manifests as 94% concerned about agent sprawl and 12% with centralized control. In the data stack, it manifests as 72% accelerating and 24% governing. The ratios are different. The architecture of the problem is identical: the tooling assumes decisions that haven’t been made.
The 2026 OutSystems report found only 36% of organizations place AI governance under a centralized IT or security team. The remaining majority rely on project-level rules, team-by-team conventions, or nothing at all. That’s not a coverage gap. That’s an architectural absence: the default posture for most organizations is no governance design at all, just ad-hoc constraints that don’t connect.
Trust as Infrastructure
dbt Labs frames it precisely: “organizations that succeed in the next phase will treat trust as infrastructure.” Not as an audit. Not as a compliance checklist. As something that is built into the system from the design layer up.
That framing matters because it exposes why the enforcement-first approach keeps falling short. You can monitor outputs. You can flag anomalies. You can build dashboards that show what went wrong. But if the governance rules — what agents are allowed to do, what data they can access, what decisions they can make — were never designed, enforcement is just a faster way to detect failures you haven’t prevented.
The data trust gap and the agent governance gap are the same gap, viewed from different altitudes. Data professionals see it in hallucinated outputs reaching stakeholders. Security teams see it in over-privileged agents executing transactions. Compliance teams see it in audit failures they can’t remediate within 90 days.
The Cost of the Missing Layer
Grant Thornton’s 2026 AI Impact Survey found that organizations with fully integrated AI are nearly 4x more likely to report revenue growth — 58% versus 15% of those still piloting. Not because governance slows things down — because it creates the confidence to scale. The 78% who can’t confidently pass an AI governance audit within 90 days aren’t just exposed to regulatory risk. They’re leaving growth on the table.
The enforcement stack now has 25+ named entities building products. Identity. Authorization. Runtime. Compliance. Every one assumes the governance rules already exist. dbt Labs just confirmed what the security data has been showing for months: those rules, for most organizations, don’t.
The acceleration isn’t the problem. The missing design layer is.
Sources: dbt Labs 2026 State of Analytics Engineering Report (14 April 2026, 363 data practitioners). OutSystems 2026 State of AI Development Report (1,879 IT leaders). Grant Thornton 2026 AI Impact Survey (950 leaders). Wyrework enforcement stack tracker.