The Test Set 1 Year

Recently, we crossed the 1 year mark with Posit’s data podcast The Test Set. I’m in good company as host, with cohosts Wes McKinney (creator of pandas, arrow) and Hadley Wickham (creator of the tidyverse). Frankly, hosting a data science podcast was not on my bingo card for the past year. But I’m glad we did it!

In this post, I want to reflect on why we started The Test Set, what we’ve learned from our 20 fabulous guests, and how we’ve grown over the past year—to give a sense for what’s next.

Why we started The Test Set

Like most good things, The Test Set started by accident. Essentially, Wes and Hadley were chatting one day and came to an idea: “wouldn’t it be cool if we podcasted together?” This was back before everybody was AI-pilled*. Eventually I ended up on a list of potential hosts that described me as “quirky (in a good way!)” and the rest was history.

But there’s a reason I think it stuck. Wes built the Python side of modern data tools, and Hadley built the R side. They work together at a company that renamed itself away from “RStudio” to emphasize data work more broadly. There may be no better pair to host conversations across R, Python (and SQL). At the end of the day, The Test Set aims to feed people’s curiosity about parts of the data ecosystem they haven’t encountered yet.

You can see this curiosity in who they want on the show—often people keeping the ecosystem afloat. For example, Alenka Frim maintaining Apache Arrow and Kelly Bodwin advocating to keep R weird (and supplying me with R-themed earrings).

What we’ve learned from 20 fabulous guests

I would break down what we’ve learned into 3 categories: the AI inflection point, data people are neat, and data ecosystems are broad and mysterious.

AI inflection point. In early 2025, our AI conversations largely centered on people talking to ChatGPT. For example, Mine Çetinkaya-Rundel discussed handling AI in her data science courses at Duke. Our conversation with Julia Silge hit on whether StackOverflow’s decline started before or after ChatGPT. Around November, I learned in chat with James Blair that I was the only person not using Claude Code. From there it took off with Charlie Marsh (Astral CEO; creator of Ruff and uv) and Wes discussing at length their non-stop use of the tool.

Data people are neat. On the other (more human) side of the coin, we got to hear in detail about many people’s journeys into data work. For example, Paige Bailey got her start by coding text-based games out of Byte magazine and has an AI agent send an email roasting her taste in music every morning. (Reminding me I’m leaving a lot of the table in terms of imagination). Mike Bostock told us he wasn’t sure how much he’d have to say about his journey creating the visualization library d3.js—but then revealed that he’s been cooking up visualization tools since the early days of the web. And Marco Gorelli, creator of the library narwhals and the charming Discord community that comes with it.

Data ecosystems are broad and mysterious. My favorite category is episodes that put the co-hosts in contact with people in a different slice of the data community (or who cut broadly across it). For example, Emily Riederer has engaged in the intersection of Python, R, and SQL. She can speak about R’s tidyverse, Python’s pandas and polars libraries, and how dbt folks like their SQL models. Benn Stancil thinks everything is a fad. Finally, Tristan Handy graciously broke down how R users might think about the dbt community. I really like Tristan’s episode, because it’s a good reminder that R, Python, and SQL folks have different cultures—and there’s an aspect of translating culturally across them.

How we’ve grown over the past year

While I can’t speak in depth for the co-hosts, I’ll offer some observations that really stood out to me.

Wes McKinney. Around October 2025, Wes’s public coding activity really shot up. This is chronicled in a post by Rich Iannone, whose graphic of Wes’s github activity I’ve included below. It shows his daily contributions, with specific tool starting points circled.

Note that 6 tools are marked with red circles, starting with moneyflow (a finance tracking app) and ending with msgvault (an email archiving and navigation tool). One interesting tool, Spicy Takes (spicy blurbs extracted from talks by different data and software folks), emerged in part because we were preparing to interview Benn Stancil.

Hadley Wickham. I think Hadley’s trajectory is best summarized in his May 18, 2026 “Returning to Life” article on his tidydesign substack. He appreciates AI enabling him to clear his backlog on tools that are tedious to work on but valuable to the R community, like Roxygen2. On the other hand, he’s been careful to acknowledge people’s “conflicted feelings about AI”. 

Michael Chow. In preparing for interviews, I have listened to so many podcasts by other interviewers—like Michael Kennedy, Chris Bailey, Hugo Bowne-Anderson, and the gone-too-soon Python Sad Girls Club—which have helped me get a handle on interviewing. I really like listening to a guest’s last 3 or 4 interviews, to develop a sense for questions they’re often asked, and stories they tend to repeat. There’s so much to learn about guests, and so I hope to get better at getting to their stories!

In Summary

The Test Set reached a year. We’ve had really incredible guests talk with us about AI, their journeys into data, and how their communities think of data work. In the podcast, you’ll hear how we reckoned with the “AI inflection point” around November 2025, and integrated AI into our work. This coming year, I suspect AI will stay a regular topic, but personally want to maintain focus on what gets guests up in the morning.

Check us out on spotify, apple podcasts, or youtube.

Looking forward to year 2!