R for the Rest of Us Podcast Episode 30: Sara Altman and Simon Couch

Most coding assistants treat data quality as a speed bump. You hand them a messy CSV, they squint at it, and a moment later they've shipped some plausible-looking analysis built on a quietly broken foundation. Posit's new AI agent takes a different stance: it pauses, asks you the awkward questions, and treats the messy bits as the point of the work, not an obstacle to it.

That agent is called Posit Assistant, and I sat down with two of the people building it: Sara Altman and Simon Couch. Sara is a senior developer advocate on Posit's AI core team. She authors the biweekly Posit AI newsletter and helped build Posit Academy. Simon is an engineer on the same team, and before he shifted his focus to LLMs he spent years maintaining core tidymodels packages like stacks, broom, and infer. Both of them are deep in the question of what an AI agent for data science should actually feel like.

Posit Assistant lives in the sidebar of RStudio and Positron (and you can also drive it from the command line). It's the successor to two earlier Posit experiments, DataBot and Positron Assistant, and it unifies their work into one tool. Sara walked me through a live demo using American Community Survey language data, and a handful of things stood out:

- It has direct access to your R session by default. The agent can see your objects, your error messages, and even your plots as images, so it's interpreting what you actually have rather than guessing at column names.

- It operates in two distinct modes. When it thinks you're trying to understand your data, it deliberately does small bits of work and keeps pace with what you can absorb. When it thinks you're trying to complete a task (build a Shiny app, refactor a package), it runs until the job is done.

- Auditability is a core design choice. Code is visible inline with syntax highlighting, output appears as it's generated, and the chat is built so you can actually read what the agent did rather than just accept the result.

- There's a new data cleaning mode (largely Simon's work) that's worth its own paragraph.

Cleaning mode is the bit I found most interesting. When the agent decides you're cleaning data, it enters a special mode where it runs a bunch of inspection code that you don't see, then surfaces the actual judgment calls to you as questions: how do you want to handle these state row headers, what do you want to do with these X markers, and so on. You answer; it applies your decisions; it writes a reproducible cleaning script. Simon's framing for why this matters stuck with me:

It seems like a lot of existing coding agents have a sort of superficial relationship with data quality. They really wanted to move this illusion of progress forward and make it seem like the data analysis is just like progressing smoothly. So this gives us an entry point to pause any notion of forward motion and really dig in on the quality, which in the reality of data science work is a huge portion of the work.

We also got into a question I get asked a lot: what about data privacy? Sara walked through how Posit's subscription puts you under a zero data retention agreement with Anthropic, which means your requests aren't stored or used for training. That kind of arrangement is usually only available to enterprise customers paying enterprise prices, so it's a real practical benefit for individuals and small teams who couldn't otherwise get it.

One thing I appreciated in the conversation is that Sara and Simon are not pretending this is solved. As Sara put it near the end, everyone is still figuring out what coding agents are good at, where their blind spots are, and what people actually want them to do. That honesty is rare, and it's reassuring coming from the people building the tool.

Whether you're already using coding agents in your R work or you've been waiting for something that feels designed for data science rather than retrofitted to it, this episode is worth your time. Watch it and let me know what you think.

R for the Rest of Us Podcast Episode 30: Sara Altman and Simon Couch

Let us know what you think by adding a comment below.