OpenAI Podcast

Joyce Ruffell & Chase Holden on why racing's data wars reward the organizer

Joyce Ruffell and Chase Holden· Researcher at OpenAI; Co-founder of RaceTek Systems at OpenAI & RaceTek Systems

2026-07-16·~41 min·English·OpenAI

AgentsAI CompanyBusiness Strategy

TL;DR

An OpenAI researcher and a racing-software founder describe a sport drowning in data, where AI earns its keep by making the softer, unstructured material usable alongside the telemetry — and where the edge still lands on the human who knows what to ask for.

01Core Mental Model

The Data Wars Are Fought Over Simplifying

Racing has gotten data-dense enough that Chase calls it a war, and the work he points at is simplifying the streams until the whole crew can move on the same picture.

What counts as data now, and where it jams

I believe we are in what's called the "data wars" right now in racing, maybe on the IndyCar side and the NASCAR side.
— Chase Holden, OpenAI Podcast

Key Insight

The word wars is doing real work here — it frames data as contested ground rather than shared plumbing. And notice where Chase puts the decisive part of that contest. Capturing every scrap is what makes a team dangerous in the first place, but he lands on simplifying the growing streams until everybody is on the same page. Capture creates the potential; shared understanding is what converts it into a call someone can actually make.

02The Technical Unlock

The Softer Stuff Was Always Data

The numbers coming off the car have long had scripts and plotting tools, while the driver describing how the car felt and the engineer scribbling notes stayed hard to organize, retrieve, and line up against the telemetry.

Where traditional tooling worked easily, and where it struggled

A driver describes how the car felt. Or how an engineer jots down some notes over the course of a session. That's data too, but that data is harder to organize traditionally.
— Joyce Ruffell, OpenAI Podcast

Key Insight

The asymmetry explains a strange shape in the sport: motorsport built sophisticated tooling around the numbers while the experiential knowledge sat right next to it, harder to organize and harder to retrieve. Which means a tool that can read a sentence lands differently than one more tool for reading numbers — the numbers already had scripts and plots pointed at them.

03The Format Gap

Motorsports Lives in Excel

The sport's knowledge sits in spreadsheets that scroll on seemingly forever, and Joyce's pitch to teams is that paying the cost of converting them is worth the results you can pull out afterward.

The conversion investment, and what it buys

a lot of the world lives in Excel. The motorsports world definitely lives in Excel. And, you know, our models can work with Excel, but it's not their favorite format.
— Joyce Ruffell, OpenAI Podcast

Key Insight

The conversion cost is a moving target, which is why Joyce says her team re-tests every three months to see whether the models can now read a little further upstream. That leaves teams working against two clocks at once: they need the downstream results now, while each model release may cut into how much legwork it takes to unlock them.

04The Surprising Finding

The Most Impactful Delivery Was Education

Asked what OpenAI gave the race team that mattered most, Joyce's answer was education — showing the engineers how the models work under the hood.

Demystify the machine, and prompting follows

One of the most impactful things that we've delivered to the team is, frankly, education. It's education about how our models work.
— Joyce Ruffell, OpenAI Podcast

Key Insight

There is a reason this landed with this audience in particular. Pulling back the veil only helps if the person has a mental model of machines waiting to receive it — and these are people who take engines apart for a living. The same demystification given to a team without that instinct would remove the magic and leave nothing standing in its place.

05The Philosophy

Man and the Machine, Evolving Together

Racing's oldest theme is getting the human and the machine to work as one system, and Joyce puts AI on the machine side of that pairing — which means the models have to move toward people, not only the reverse.

Both sides move; the gap is the work

It's not just that we need to make AI work with how people work today. It's about having them evolve together and having people learn how to work with and communicate with our models in the most seamless way possible.
— Joyce Ruffell, OpenAI Podcast

Key Insight

Evolving together is a bigger commitment than it sounds, because it makes the gap a moving boundary that both sides push on rather than a fixed limitation of the team's tooling. And Joyce puts the harder half on her own side of the table: it is the people building AI who have to figure out where the gap is so they can close it.

06The Leveling

Closer to the Juggernauts Without the Headcount

The big teams put multiple engineers on the box, each watching their own dataset, and Chase argues the tooling lets a smaller team cover some of that same ground without the headcount it could never afford.

Closer to the juggernauts, without the headcount

I don't think they'll ever fill like an entire team of engineers, but it's something that can get smaller teams and medium-sized teams closer to what some of these juggernauts on the track are doing.
— Chase Holden, OpenAI Podcast

Key Insight

Joyce's number reframes what juggernaut even means here: Chip Ganassi is under 200 people and everyone wears lots of hats. At that size, freeing up one person's time is not a rounding error — it is the difference between chasing a question and letting it stay deprioritized, and she points out you never really know which of those questions was the next edge.

07Forward-Looking

When Everyone Has It, Taste Is the Edge

Asked where the advantage goes once every team can hand the problem to an AI, Chase says it lands back on instinct, taste, and who can say what needs to happen fastest.

What survives when the tool advantage spreads

Man, that's when it all comes down to who has been doing it the longest by the time that arrives. Who has the best instinct? Who has the best taste?
— Chase Holden, OpenAI Podcast

Key Insight

Joyce sharpens this into something uncomfortable for the sport. Motorsport has gone through cycles where the edge was simply investment: put in the resources and the engineering, and you would find it. If the cost of proving out an idea collapses, the scarce input stops being capital and becomes the idea itself — which inverts racing's oldest assumption, that the edge is something you can buy.

08The Human Element

These People Don't Ask AI to Lead

The most goal-oriented people Joyce has worked with do not ask the model for direction — they show up with a precise deliverable and a quality bar it has to clear.

Exploring versus arriving with a specification

I actually think the application of AI we're seeing in motorsports is extremely human-centric, because these are some of the most goal-oriented people I've ever met and worked with.
— Joyce Ruffell, OpenAI Podcast

Key Insight

This is the part that should interest anyone building models rather than using them. A user who arrives with a specification and a stopwatch is a harsher evaluator than any benchmark, and Joyce says the pressure runs upward: working with a team this demanding is what taught her how high the models' performance actually needed to be. The domain is not just consuming AI — it is grading it.

Through Line

Data is everything in racing now, and that abundance is exactly what makes the data wars a war: a storied team has thirty-five years of it, every telemetry point and handwritten note is worth catching, and none of that helps until the streams are simple enough for the whole crew to move on one picture. The softer material — the feel, the session notes — is the part that stayed hardest to organize, the spreadsheets it lives in are the part the models had to be taught to read, and the delivery that landed first was not a model at all but education about how the machines work. From there Joyce's framing takes over: man and machine evolving toward each other, with the builders obliged to find the gap and close it. What that levels is real, letting a smaller team cover ground it could never staff, though never a whole bench of engineers — and the leveling quietly erases the advantage it hands out. What is left to race on is instinct, taste, and a crew that arrives already holding the specification and the quality bar.