R for the Rest of Us Podcast Episode 29: Mike Stackhouse

In this episode, Mike Stackhouse, Chief Innovation Officer at Atorus Research, shares valuable insights on effectively using AI with R programming, especially in regulated environments like pharmaceuticals and healthcare.

Think of AI as a "Very Confident Junior Programmer"

Mike's most memorable analogy frames LLMs perfectly: they're like junior developers who are extremely confident but occasionally wrong. Since LLMs fundamentally work with text in and text out, with limited context windows, you can't simply dump a 5GB dataset into Claude and expect magic. Instead, you need to break down tasks into manageable pieces and provide clear, step-by-step guidance.

Planning Mode Changes Everything

Mike's approach to AI has evolved beyond simply asking for code. He now uses Claude to:

• Build requirements documents

1. Create design specifications

2. Implement solutions step-by-step

This planning mode represents a fundamental shift—it's not just about the code AI writes, but about the thinking and structure that goes into that code.

Enterprise-Level Security for Sensitive Data

For those working with sensitive data, Mike explains how AWS Bedrock enables AI use in controlled environments. Your data never leaves your network, making it viable for pharmaceutical companies, healthcare organizations, and other regulated industries that previously couldn't consider using AI tools.

Real-World Application

During the episode, Mike demonstrates Claude Code within Positron, creating a clinical analysis report with Quarto from scratch. Starting with just a prompt and data files, Claude generates all the code, creates tables, and renders a complete report. While impressive, Mike emphasizes the critical importance of reviewing and verifying AI's work.

Reproducibility is Non-Negotiable

Mike advocates for a code-first approach: rather than having AI simply summarize data, ask it to write code that you can review, understand, and reproduce. This approach is essential in regulated fields where you must demonstrate quality control and be able to reproduce results on demand.

The key takeaway? AI can dramatically accelerate your R workflow, but success requires understanding its limitations, establishing proper workflows, and maintaining rigorous quality control.

Resources

• GitHub repository with demo code

• Atorus Research

Connect with Mike on LinkedIn.

Learn More

If you want to receive emails to help you on your R Journey, sign up for the R for the Rest of Us newsletter.

If you're ready to learn R, check out our courses.​