Genie Code is an addition to Genie, which allows any knowledge worker to talk to their data and receive reliable answers based on the context and semantics captured in Unity Catalog. In a similar way, Genie Code takes over the complex engineering needed by data professionals to bring new ideas to production using all available business data.
In addition to the launch of Genie Code, Databricks today also announces the acquisition of Quotient AI, an innovator in evaluation and reinforcement learning for AI agents, which will integrate continuous evaluation directly into Genie and Genie Code.
Read also: How to bring AI structure to your team
The rise of data work by agents
Modern data tools primarily use AI as an assistant that writes code, runs local tests, and iteratively improves upon them. The heavy lifting such as planning, orchestrating, managing, validating, and maintaining remains with the data teams. Genie Code flips this approach. The agent reasons to solve problems, plans multi-step processes, writes and validates production-level code, and maintains the solution, while humans retain control over the decisions that truly matter.
'Software development has shifted in the past six months from code assistance to full-fledged agentic engineering,' says Ali Ghodsi, co-founder and CEO of Databricks. 'Genie Code brings that same revolution to data teams. We are moving from a world where data professionals receive help from AI to a world where AI agents do the work – driven by humans. We call this approach agentic data work and it will fundamentally change how organizations make decisions.'
Read also: Equinix launches Distributed AI Hub to simplify and secure companies' AI infrastructure
How Genie Code works
Existing agentic coding tools struggle with data tasks because they lack access to crucial context, such as lineage, usage patterns, and business semantics. Genie Code helps teams bridge this context gap and ensure the high accuracy and governance needed in production environments. Genie Code:
- Acts as an experienced machine learning engineer: Genie Code supports complete ML workflows from start to finish. The agent reasons through complex problems to plan, build, and deploy models, while experiments are logged in MLflow and serving endpoints are fine-tuned for maximum performance.
- Captures deep data engineering expertise in the existing tool: where a novice engineer writes a script that only works on test data, Genie Code designs like a senior architect. The agent considers the differences between staging and production environments, builds workflows for change data capture, and applies data quality rules.
- Proactively maintains and optimizes: in the background, Genie Code monitors Lakeflow pipelines and AI models to assess failures and investigate anomalies. The agent independently analyzes agent traces to recover from hallucinations and adjusts resource deployment before a human needs to intervene.
- Understands the context of the enterprise: Genie Code is integrated with Unity Catalog to enforce existing governance guidelines and access controls. The agent understands business semantics and audit requirements and can retrieve business data from various sources, including external platforms.
- Gets better over time: Genie Code becomes smarter as more teams use the agent. Through persistent memory features, the agent automatically updates internal instructions based on previous interactions and preferred coding styles. In practical tests in data science, Genie Code achieved a success rate that was more than twice that of other code agents – from 32.1% to 77.1%.
'At SiriusXM, Genie Code supports everything from writing notebooks and complex SQL to understanding table relationships and debugging pipelines,' says Bernie Graham, VP Data Engineering at SiriusXM. 'The agent acts as a development partner that collaborates with us and helps our data teams deliver high-quality work in less time.'
'Genie Code changes how our data teams work,' says Emilio Martín Gallardo, Principal Data Scientist, Data Management & Analytics at energy company Repsol. 'Instead of manually stitching together notebooks, pipelines, and models, we can hand over complex workflows to an AI partner that understands our data, governance, business context, and internal libraries like Repsol Artificial Intelligence Products. This accelerates everything from time series forecasting to deployment into production, without compromising on robustness or control.'
Read also: 6G: what does the future of networking look like?
Acquisition of Quotient AI strengthens continuous evaluation
To fully ensure quality in production, Databricks has acquired Quotient AI. Quotient automatically monitors the performance of agents: it measures the quality of answers, detects regressions early, and locates failure points. These insights then feed into a reinforcement learning cycle that ensures agents continuously improve. The founders of Quotient bring deep expertise in evaluating AI coding systems; they previously led the quality improvement of GitHub Copilot. By embedding these capabilities into Genie Code, Databricks ensures that data and AI systems not only run in production but also continuously evolve.
