Responsible by Design: Building AI That Works With You
Artificial intelligence is poised to transform data science, but this power comes with profound responsibility. At Posit, we’ve been busy experimenting with large language models to understand the ways they can – and can’t – support data science.
Through countless hours of building, experimenting, and testing, we’ve learned that when it comes to data science, today’s large language models are remarkably capable, yet they are not infallible. When applied to data science, they can make subtle but critical errors—errors that, if left unchecked, could have catastrophic consequences for scientific research, financial modeling, and public policy.
At the same time, our experiments building with AI have led us to the conclusion that the promise of AI for data science is too great to ignore, so we must venture into this new territory with our eyes wide open. This reality requires a thoughtful approach.
At Posit, our guiding principle has always been to ensure that analyses created with our tools are correct, transparent, and reproducible. To uphold this principle in the age of AI, we believe today’s large language models demand a specific design philosophy when applied to data science: a human expert, who can understand and vet the code, must always be in the loop. Our goal isn’t to replace the data scientist but to build power tools that augment their expertise.
As part of this mission, we are excited to introduce agentic AI capabilities in Positron, our new, free code editor for R and Python, that are designed from the ground up to follow these principles. Positron’s AI capabilities automate the tedious parts of the data science workflow, but always keep you, the expert, in control.
Positron Assistant
Most AI coding assistants are designed for general software engineering. They’re great at understanding code on disk but often miss the unique, interactive context of a data science session.
Positron Assistant is different. It’s a coding assistant that has access to the live state of your R or Python session, including in-memory data, plots, and variables. This awareness allows it to provide much more relevant and useful suggestions for writing and debugging your code.
For data scientists, this means you get highly relevant help debugging tricky data issues, which dramatically cuts down on roadblocks. And Positron Assistant provides context-aware code suggestions and edits directly within the IDE, so a developer can get help while maintaining their flow, without switching tabs.
By design, Positron Assistant requires your expert review to turn those suggestions into action, ensuring you maintain control and can catch potential issues before they are executed. It helps you write better code faster, without ever taking your hands off the wheel.
Databot
Exploratory data analysis (EDA) is the foundational, and often the most time-consuming, part of any data project. It’s an iterative process of questioning, visualizing, and summarizing a new dataset to uncover its structure, quality issues, and interesting relationships.
Databot is a new, experimental agent designed to dramatically speed up this process for experienced R and Python users. It acts like an interactive partner for your analysis. You give it a high-level instruction, like “Generate visualizations to summarize the last 12 months of data”, and Databot enters a loop:
- Write code: It writes the R or Python code needed to answer your request.
- Execute: After approval, the code runs automatically in your session, and you see the output, plots, and tables.
- Analyze: Databot interprets the output and makes observations.
- Regroup: It then proposes a few intelligent next steps for the analysis, keeping you in full control of the direction.
This cycle reduces the effort required to probe and visualize data, letting you move from one insight to the next as quickly as you can think. Databot handles the routine coding, while you provide the critical thinking to recognize what matters and decide where to go next.
When you’re done, you can ask Databot to export your most important findings to a notebook or Quarto document for closer inspection and sharing with others.
Keeping You, the Expert, in Control
A core principle behind both agents is that you, the expert, are in control. We know that today’s AI models cannot be trusted to perform open-ended data science without making mistakes. That’s why Positron’s AI experience is intentionally code-forward. The generated code is always visible to you, and a final analysis isn’t complete until you have reviewed it for accuracy.
Positron’s AI is built to augment your expertise, not to create a black box that operates without your oversight. These are our guiding principles in action.
No Hidden Costs, Just Your API Usage
While AI coding assistants have the potential to dramatically enhance your productivity, we know the true cost to use them is often uncertain.
Positron’s AI capabilities only require your existing LLM provider credentials. Today, you can use Positron Assistant and Databot with an Anthropic API key, with support for additional models and providers coming soon. There’s no cost to use Positron’s AI capabilities beyond your API usage. Positron Assistant and Databot are both available in the free, desktop version of Positron.
Download Positron today to explore how these new AI agents can responsibly fit into your data science workflow.
