One of the most stubborn problems in enterprise AI is not getting models to work. It is getting them to work and stay governed at the same time. Dataiku is calling this the “build-govern gap” and says its new Cobuild agent, now generally available as of June 18, 2026, is built specifically to close it.

The announcement came June 11, and it lands in the middle of a broader industry reckoning. Enterprises have spent years building data foundations. Many have internal AI strategies in place. But the path from “we have a use case” to “this is running in production with proper oversight” has remained frustratingly long. Teams either build fast without governance or govern tightly without moving fast.

Cobuild is Dataiku’s attempt to make that tradeoff unnecessary.

What Cobuild Actually Does

The core mechanic is straightforward. A business team describes what they want to accomplish in plain language. Cobuild generates a complete AI project inside Dataiku’s visual interface, including data pipelines, predictive models, agent workflows, and deployable applications.

Critically, it does not generate opaque scripts that get handed to an IT team for review. Everything Cobuild creates lives inside Dataiku’s existing governance and permissioning framework. The output is a structured visual flow that teams can inspect step by step, validate assumptions against, and approve before anything goes live.

That distinction matters more than it might seem. The wave of AI coding tools that arrived in 2024 and 2025 created a new governance problem alongside the productivity gains: outputs that worked technically but bypassed the enterprise controls organisations depended on. Cobuild is designed from the ground up to make governance a first-class part of the build process rather than something bolted on afterwards.

Model Flexibility Without Lock-In

Cobuild connects to a wide range of underlying AI providers. Teams can use Dataiku AI Services directly or bring their own models through Dataiku LLM Mesh, with out-of-the-box support for Snowflake Cortex AI, OpenAI, Anthropic, AWS Bedrock, Google Gemini, Microsoft Foundry, Databricks AI Gateway, and others.

This matters for enterprise buyers who have already standardised on a preferred LLM provider or who want the flexibility to switch without rebuilding their AI infrastructure. The orchestration layer stays constant even as the underlying models change.

What This Means for Business

The “build-govern gap” is not just a Dataiku framing. It shows up in the numbers across the industry. Research from mid-2026 puts enterprise agentic AI production adoption at 72%, but only 21% of organisations report having a mature governance model for their deployed AI. More than half cite data quality and lack of oversight as the biggest blockers to scaling.

Cobuild is a direct response to that dynamic. Its practical implications for business leaders are a few:

Broader team participation in AI development. When building a production-ready AI project requires no code, the work stops being exclusively owned by a centralised data science team. Business analysts, operations managers, and domain experts can contribute to AI projects that go live, not just projects that get stuck in a handoff queue.

Faster path from use case to deployment. The traditional pipeline from business problem to working AI system involves multiple handoffs across analytics, engineering, and governance. Cobuild compresses those steps by creating governed, traceable outputs from the first moment of development.

Reduced shadow AI risk. One reason AI governance fails is that business teams route around slow central processes and build things on their own. When building inside the governed environment is as fast as building outside it, the incentive to bypass controls goes away.

Better audit readiness. Every step of a Cobuild-generated project is visible and traceable. That matters for regulated industries and increasingly for enterprises responding to customer, investor, and procurement AI scrutiny.

Why It Matters for Data Teams

For the data and analytics professionals in EDNA’s community, Cobuild represents a shift in what “building AI” looks like at the enterprise layer. The question is no longer just which model is best or which framework to use. It is how to make AI development something the whole organisation can participate in, not just the technical specialists.

The timing coincides with the Databricks Data + AI Summit in San Francisco this week, where the prevailing theme is getting agents to work reliably in production. Cobuild is answering the same question from a slightly different angle: not just reliability, but the human and governance layer that makes reliability possible at scale.

Dataiku Cobuild is available to all Dataiku customers from June 18.