Introducing Dataiku Cobuild on Snowflake

May 20, 2026

3 min read / Patrick Phrayme

From Shadow AI to Shadow Agents
The Hidden Cost of Autonomy
Why Traditional Governance Breaks Down
Visibility Is the Missing Primitive
Human-in-the-Loop as a Design Choice
The Real Shift: From Limiting AI to Seeing It

Companies are still searching for their “golden goose” data scientist.

Someone with a Ph.D. in math who can work with petabyte-scale data, train Kaggle Grandmaster-level machine learning (ML) models, build CI/CD pipelines to automate them, and roll their own open source development stack.

Our golden goose must be able to understand the business like a 30-year veteran, build compelling slides, and communicate highly technical solutions to non-technical executives. And, of course, our golden goose must usher in a company-wide AI transformation. We suddenly need 500 geese — and everything rests on their shoulders.

With Dataiku Cobuild on Snowflake, we're here to turn everyone into a golden goose.

If you’ve been following Dataiku, you know that this has been our goal from day one. Just last week (day 4,700ish), our E2A (expert-to-agent) launch put visual, inspectable AI in the hands of business experts. Now with Cobuild, we are extending that to anyone who can describe what they want to build.

Why Snowflake as our first launch partner? They share our goals to help enterprises build successful data, analytics, and AI solutions — and our view that success comes through building tools that break down technical barriers, empower others, and encourage collaboration.

What it does

Cobuild is an agent that helps build AI projects across analytics, data cleaning, ML, custom agent development, and more. A user describes what they want in plain language — "build a churn model on the last four quarters of account activity, weighted by contract size, and flag the top decile for outreach" — and Cobuild assembles the whole project as a visual, inspectable Dataiku workflow using our E2A engine.

Projects can include data preparation, joins, data quality checks, custom trained models, and charts to interpret model bias. Cobuild can help build agentic workflows with custom routing logic, or polished webapps that tap into the data, models, and agents it created previously.

The output isn't code (though it can be, if you ask it to). It's a visual Dataiku flow. Every node is inspectable. The analyst who asked the question can click through and see exactly which Snowflake dataset Cobuild pulled, how it was filtered, which features the model trained on, and what the performance looks like across different segments. They can edit any step. They can approve it. Nothing reaches production without sign-off.

When building enterprise data and AI applications, it’s crucial to follow data, transformations, models, and agents at each step. That's the distinction between enterprise building with Cobuild on Cortex, and vibe-coding with traditional agentic coding tools. Other AI coding tools produce a wall of code that only a developer can audit. Cobuild produces a flow the person closest to the business problem can actually evaluate.

Built on Snowflake Cortex

When Cobuild runs on Snowflake, REST API calls to models (Claude, GPT) go through Cortex inside the customer's Snowflake account. For organizations that standardize on Snowflake, this means that data, conversations, and workload computation all stay in Snowflake. IT can rest easy knowing that their existing, Snowflake-centric governance model applies to every workload that Dataiku Cobuild generates.

Snowflake customers have invested in a foundation that's secure, governed, and ready for AI. Cortex brings the intelligence of leading foundation models into that same trusted environment. Now, Cobuild gives teams a conversational way to build with that intelligence, across more teams and more use cases, without ever leaving the Snowflake architecture.

What changes for teams already running both

More than 350 organizations run Dataiku on Snowflake today. For those teams, Cobuild expands who can contribute to AI development on top of their existing data foundation.

Work that's been kicked down the road three straight quarters moves forward. Work that needs our Ph.D. data scientist still goes to the data science team, with a shorter queue in front of it. And employees who understand the business better than anyone else now have a way to build custom data and AI applications inside a governed environment, where their work can be reviewed, reused, and operationalized — instead of spinning up separate, ungoverned tools.

For teams already running production workflows on Dataiku and Snowflake, development moves faster. Describe what needs to change in plain language — Cobuild does the rework against the same governed data, the same Cortex models, the same Dataiku project flow that's already running.

See it at Snowflake Summit

We'll be demoing Cobuild on Snowflake at the Dataiku booth (#2414) at Snowflake Summit, with sessions throughout the week. If your backlog of good ideas is longer than the team you have to build them, come find us.