What’s new in Genie Code at Data + AI Summit 2026

What’s new in Genie Code at Data + AI Summit 2026 | Databricks Blog Skip to main content

Summary

Genie Code is the specialized agent for data and ML work on Databricks, helping teams build, debug, and improve production systems.

A new full-page command center gives data teams one place to manage complex, multi-threaded work, with thread status, review points, and quick access to instructions, skills, and connectors.

Scheduled tasks will soon let Genie Code run work autonomously, while Genie ZeroOps extends that automation into data and ML operations.

Genie Code helps data and ML teams build and improve systems faster on Databricks. Over the past year, Databricks’ Genie products have grown over 10x and are used by 90% of Databricks customers. Teams are using it to build models and pipelines, debug failures, create dashboards, analyze data in notebooks, and improve production systems. At Data + AI Summit 2026, we’re expanding Genie Code for more complex, agentic data and ML work. We’re introducing a new full-page command center, upgrades for production data and ML engineering, and scheduled tasks. These updates are part of a broader shift across Databricks toward AI-native data and ML workflows. Genie Code helps data teams build, debug, and improve data and ML systems, and we  introduced Genie ZeroOps to extend agentic automation to  operations. Together, these products help teams move faster across the full lifecycle, from building systems to operating and improving them over time. Here’s what’s new: Manage complex work from a full-page command center Data and ML development rarely happens in a single prompt. A user may need to inspect existing logic, update multiple assets, run code, review outputs, and refine the next step based on results. That work can span notebooks, SQL, Lakeflow pipelines, dashboards, jobs, models, serving endpoints, and Unity Catalog assets. We've redesigned the Genie Code experience to give teams a dedicated  command center for this kind of complex data and ML work . Instead of managing longer tasks in a smaller side panel, users can use a full-page experience to describe a task, track progress, review outputs, and continue iterating. Teams can manage multiple Genie Code threads, see when a thread is executing or waiting for input, and return to each one when new results are ready. They can rename threads, search previous conversations, and stay oriented as projects evolve. Genie Code has completely changed how I work. I run upwards of 15 parallel threads scoped to different notebooks and assets every day, and managing all of that across tabs is one of the biggest sources of friction in my workflow. Full page Genie Code with concurrent sessions would give me a true workspace for running everything in parallel without constantly losing context. — Moritz Schiek, Solution Consultant, Bosch The command center also makes customization easier to discover. Instructions, skills, and connectors are more visible, so teams can guide Genie Code with the right standards, tools, and workspace knowledge. To access the full-page command center, open the Genie Code side panel and click the maximize button in the top left. Bringing agentic development to ML workflows In an ML project, the model is the small part. Most of the time goes to the engineering around it: turning raw data into features, running experiments, getting a candidate to production, and keeping it healthy once real traffic arrives. That work is slow and expensive, and it is why most teams run far fewer models than they have use cases for. Genie Code for machine learning goes after that engineering. It is a set of capability and intelligence upgrades built into Genie Code, so you do not adopt a new tool. The same agent you already use becomes a specialist for production ML engineering across your existing Databricks ML stack. Genie Code's expertise comes from two places. The first is Databricks. We have run production ML with customers for more than a decade, and we have seen where models break, where teams lose time, and what separates a model that works from one that looks right and fails without warning. Genie Code applies those lessons as it works, handling the details a seasoned practitioner would, like correcting for class imbalance and checking feature quality. The second is your team. A generic coding agent has not seen your past experiments, your business metrics, your evaluation sets, or how you weigh one objective against another, so it guesses.  Genie Ontology closes that gap. It learns how your team builds features, trains models, and evaluates candidates, and Genie Code follows those patterns instead of falling back on irrelevant defaults. With both kinds of knowledge, Genie Code is a stronger partner for day-to-day model development. It writes features in your team's patterns, makes coordinated edits across the many files involved, runs and debugs your code, and compares candidates with your own evaluation scripts. You stay in the loop and decide what to keep. Genie Code is now natively integrated with the entire Databricks ML stack. The latest upgrades: MLflow. Genie Code reads your experimentation and observability data: runs, artifacts, model lineage, quality metrics, and system metrics. Ask it "How do I improve GPU utilization during training?" or "What other metrics should I track for this model?" and get answers grounded in your own runs. Model Serving. Genie Code inspects endpoint health and performance, diagnoses serving issues, and finds ways to optimize a running endpoint. Compute awareness. Genie Code moves to AI Runtime when a job needs a GPU for training, and uses workspace environment features to set up the environment, so you skip the infrastructure setup.

The result is an agent that finishes real-world data science tasks far more often than a generic coding agent. With Genie Code, we moved from raw data to a governed, production-ready ML workflow in 90 minutes. Because it uniquely understands production ML workflows on Databricks, it helped us create Delta tables, explore the data, train and compare models, register them with MLflow and Unity Catalog, and deploy the champion model to a serving endpoint, with time left to optimize for the business outcome that mattered most. — Radu Dragusin, Principal Engineer, Data & AI, Danfoss Let Genie Code work autonomously with scheduled tasks Until now, Genie Code has been primarily interactive: you ask, it...