Wes McKinney

We have a show called Show Us Your Agent Skills where people come on and show us how they use agents.

In the first episode, we had Wes McKinney, Jeremiah Long from Prefect and Randy Olson.

And I had Wes present first and both the other guests were like, why'd you do that to us, dude?

Well, that's really good to know.

That's a lot more than I thought.

Oh my God, Wes McKinney replied in this thread.

So Wes McKinney-

Yeah

... is, like, one of the creators.

Can you talk to that a little

bit? So this is a comment that my friend Leonard Lin made to me last summer was that all of the frontier models seem to have ended up with broadly humanist values.

You can tweak this by changing the system prompt.

I think Grok most famously got talked into saying horrible things, but that was in the system prompt.

I grew up in an era when you could write tests and they either passed or they failed.

And now what we have is, do the tests pass at least 80% of the time? Can we get it up to 90? Can we get it up to 95? And to do that, you have to run a whole bunch of times.

My friend and colleague, Simon Willison, who works with us one day a week, spent probably about a month thinking about evals for us and built out the framework these SM evals, small model evals that is designed as a relatively human friendly way to build evals for agentic tasks that doesn't require you to run a frontier lab or have a giant fleet of, you know, a giant fleet of computing hardware in a closet somewhere.

It's on the Prime Radiant GitHub.

point out one person, and that's

Wes McKinney, actually.

He was, I think at the time, Ursa Labs, the company was our first.

First commercial customer.

So I'm Michael Chow.

I'm the host and a software engineer at Posit and joined by co-host Wes McKinney, who's a principal architect at Posit and founder of Ken.io. and then Hadley Wickham, chief scientist at Posit.

And so we have a very Posit-stacked crew and honestly so excited to be able to talk more, Tarif, about, you know, your...

I think a lot about your background and kind of how you approach things.

And, uh, so it's, it's, those would be frustrating nights.

Cause sometimes I'm like, all right, I need to get this deck out to the board and I've just hit some kind of weird edge case.

And so sometimes I'd be like sending Slack messages to some of our engineers on the West coast that were still awake at that time.

I'd be like, listen, I'm trapped.

I started to be able to learn these patterns.

Wes McKinney was the developer of a data manipulation package in Python called Pandas, right? Which is very well known as basically a way within Python to manipulate time series data.

He built it because he was at a QR at a hedge fund and he saw people just crunching time series and spreadsheets and was like,

this is

Really appreciate it.

Yeah, and then just by way of introduction, I'm the host, Michael Chow, and I'm joined by Hadley Wickham, who's chief scientist at Posit, and Wes McKinney, who is a principal architect at Posit and founder of Ken.io AI computing tools company.

So happy to have everyone together.

I'm curious, have you all met before?

I feel like we've likely crossed paths maybe at like a sci-fi conference or something, but I'm having a hard time placing it.

Met you at a sci-fi, Wes.

I obviously remember it because it was a chance for me to meet one of my heroes.

You obviously will not remember it because it was a long time ago and I hadn't done much by that point.