Posit Assistant: A coding agent built for data scientists
Posit's new AI coding agent is built specifically for data scientists, and it works very differently from general-purpose tools like Claude Code or ChatGPT. I sat down with Sara Altman and Simon Couch of Posit's AI core team to talk about Posit Assistant: a coding agent that lives in RStudio and Positron, has direct access to your R session, and includes a new data cleaning mode that pauses to ask you the awkward judgment calls instead of pretending data quality is a solved problem. Sara walked through a live demo using American Community Survey language data, and we got into the design choices, the privacy story (zero data retention with Anthropic), and what happens to DataBot and Positron Assistant going forward. Chapters 0:00 - Intro 2:00 - Sara and Simon's roles on the AI core team 3:19 - What is Posit Assistant 6:00 - How agents differ from ChatGPT and claude.ai 8:00 - Demo: Posit Assistant in RStudio 14:00 - Plotting and visual interpretation 15:14 - Slash commands and Quarto reports 17:00 - Thinking mode explained 19:49 - Auditability and reproducibility 21:12 - Permission system and plan mode 23:23 - Data cleaning mode demo 32:07 - Why Posit Assistant over Claude Code 36:04 - Security and zero data retention 39:07 - What this means for DataBot and Positron Assistant 41:07 - Closing thoughts 42:19 - Outro About the guests Sara Altman is a senior developer advocate on the AI core team at Posit, where she focuses on how AI can be effectively and responsibly used for data science. She authors the biweekly Posit AI newsletter, and previously helped build Posit Academy and taught data science and R at Stanford. Find her at [Sara's website or social]. Simon Couch is an engineer on the AI core team at Posit. His work spans coding agents, model evaluations, inference engineering, and next edit suggestion systems. Before shifting his focus to LLMs, he spent several years authoring and maintaining core packages in the tidymodels framework, including stacks, broom, and infer. He blogs at https://simonpcouch.com. Rescources mentioned in this episode Posit Assistant documentation: https://posit-dev.github.io/assistant/ Comparing Claude Code and Posit Assistant (video): • Comparing Posit Assistant and Claude Code Trust and privacy with LLM tools (Posit blog post): https://posit.co/blog/trust-llm-tools Posit AI newsletter: https://opensource.posit.co/tags/ai-n... stacks package: https://stacks.tidymodels.org/ broom package: https://broom.tidymodels.org/ infer package: https://infer.netlify.app/ American Community Survey language use data: https://www.census.gov/topics/populat... Connect with Sara Altman Github: https://github.com/skaltman LinkedIn: / sarakaltman Bluesky: https://bsky.app/profile/sara-altman.... Connect with Simon Couch Github: https://github.com/simonpcouch LinkedIn: / simonpcouch Bluesky: https://bsky.app/profile/simonpcouch.com

Using AI in R without breaking reproducibility, version control, or trust

Make high-quality tables in R with Tom Mock

Top 10 Ways I Use Local AI to Run My Software Company (Real Workflows)

How Hackers Trick AI Models (Prompt Injection Explained)

Using AI to collect data

Using Large Language Models | Build Your Own LLM Workshop #1

How AI agents & Claude skills work (Clearly Explained)

Full Walkthrough: Workflow for AI Coding — Matt Pocock

They're laughing at the SpaceX bubble

Stop Prompting Claude. Use Karpathy's Method Instead.

Data dictionaries, parquet, & Claude | Hadley Wickham | Data Science Lab

How To Think SO CLEARLY People Assume You're A Genius

Harness Engineering: What Separates Top Agentic Engineers Right Now

Andrej Karpathy: From Vibe Coding to Agentic Engineering w/ Stephanie Zhan

How Model Context Protocol (MCP) actually works

Inside Anthropic, the $965 Billion AI Juggernaut | The Circuit

Cédric Scherer and Georgios Karamanis Discuss Their Data Viz on Drought in the US

Pros and Cons of Positron

Google Just Dropped a Masterclass on Agentic Engineering (It's SO Good)

