Signals and Threads

Chloe Casale on automating chat trading with LLMs and comparing hiring at Jane Street and Palantir

1h 34m Listen to the original Episode page
Algorithmic TradingLLMs in FinanceEngineering CultureHiring PracticesJane StreetPalantirChat TradingRisk Management

Summary

Chloe Casale discusses her transition from mechanical engineering to computer science and her roles at Palantir and Jane Street. She details how Jane Street uses chat and LLMs to automate trade execution, balancing speed with legal and operational risks. Casale compares Jane Street's generalist hiring philosophy with Palantir's specialist-focused approach. The episode covers internal tools like Bonsai and Webs, emphasizing varying levels of scrutiny for AI-generated code based on risk profiles.

Episode sections

  1. 0:03 – 5:27

    Mongol Derby and Injury

    Chloe Casale describes participating in the Mongol Derby, a 1,000km endurance horse race in Mongolia. She explains the race format, her lack of prior competitive riding experience, and how she suffered a stress fracture on her tibia on the final day due to a stirrup issue and stuffed socks, yet managed to finish the race.

  2. 5:27 – 9:41

    Transition to Computer Science

    Casale recounts her background in mechanical engineering and auto tech, and how she pivoted to computer science after taking an intro CS class on a $20 bet with friends. She describes falling in love with programming and switching her major to applied computational mathematics sciences at the University of Washington to graduate on time.

  3. 9:41 – 12:26

    Palantir and Forward Deployed Engineering

    Casale details her four-and-a-half-year tenure as a Forward Deployed Engineer (FDE) at Palantir, a role she describes as solving specific problems for single clients rather than building general-purpose products. She notes the role involved significant travel, working in locations like Brazil, Indonesia, Mexico, London, and Seattle, and feeding customer insights back to the product team.

  4. 12:26 – 16:13

    Joining Jane Street and Interview Process

    Casale explains her decision to join Jane Street despite having no background in trading or OCaml. She highlights Jane Street's rigorous interview process, specifically the use of two interviewers per session, as a signal of high-quality hiring practices that convinced her to accept the offer.

  5. 16:13 – 22:11

    Jane Street Trading Operations

    The speakers define the roles of sales traders (client-facing, managing relationships and executing bilateral trades) and prop traders (pricing and risk management). They explain that Jane Street acts as a liquidity provider, offering better prices for large orders by leveraging knowledge of client intent and hedging through underlying assets, thereby avoiding the adverse selection risks present in anonymous exchange trading.

  6. 22:11 – 25:51

    Chat vs RFQ Platforms

    Ron and Chloe discuss why clients prefer chat over electronic Request for Quote (RFQ) platforms. Chloe notes that RFQ platforms require significant integration work and software costs, which is inefficient for clients executing few trades. Chat via Bloomberg terminals offers a flexible, low-friction alternative for ad-hoc trading.

  7. 25:51 – 27:45

    Automating Chat with Regex and LLMs

    Chloe explains the evolution of automating chat parsing. Initially, regex parsers handled standardized client requests, reducing manual entry errors. With LLMs, the firm can now parse less structured, typo-ridden messages using prompts rather than rigid code, though this introduces non-determinism and latency challenges.

  8. 27:45 – 35:46

    Speed and Human-in-the-Loop Risks

    The speakers analyze the trade-off between speed and accuracy. While automation reduces 'fat finger' errors, LLMs introduce 'alert fatigue' risks where humans fail to catch the 5% of errors. Chloe emphasizes the need for system-level checks rather than relying on human verification.

  9. 35:46 – 41:15

    Contract Formation and Error Handling

    Chloe details the legal and operational risks of LLM misparses, such as quoting the wrong asset (e.g., Tesla instead of Apple). She describes the concept of 'meeting of the minds' in chat-based contracts and the importance of using explicit, unambiguous confirmation strings to mitigate disputes.

  10. 41:15 – 45:34

    Internal UI Frameworks: Bonsai and Webs

    Chloe introduces Jane Street's internal tools: Bonsai, an OCaml-based React-like framework for robust UIs, and Webs, a distributed computational graph tool. Webs is described as an industrialized, collaborative spreadsheet that allows non-technical users to build complex data views and computations.

  11. 45:34 – 47:30

    Webs Framework and AI Integration

    Chloe Casale describes Webs, an internal spreadsheet-like framework at Jane Street that allows users to share data and build views. The system integrates with AI, enabling recruiters to use a code assistant to generate views via chat, empowering non-developers to analyze data without needing dedicated dev support.

  12. 47:30 – 49:00

    Development Workflow: From Vibe Coding to Production

    Jane Street developers use a spectrum of tools ranging from simple HTML/JS files for quick prototypes to the Webs framework for shared data graphs, and finally to Bonsai UIs for critical, custom applications. The choice depends on the value and scrutiny required for the tool.

  13. 49:00 – 51:41

    Safety and Scrutiny in AI-Generated Code

    The speakers discuss the risks of 'vibe coding' (using LLMs to generate code) by non-technical users. They emphasize varying scrutiny levels based on risk: read-only dashboards for sense-checking are low-risk, while tools affecting trading decisions or regulatory reporting require high scrutiny and human review.

  14. 51:41 – 1:34:01

    Sandboxing and Verification Challenges

    Ron highlights that web pages have built-in sandboxing assumptions, but LLM-generated apps lack this. He proposes using formal methods or separate LLM sessions to verify that generated code adheres to a safety spec, ensuring that even if the generating model is unreliable, the output is constrained.

Assets discussed

AssetKindStanceContext
AppleAAPLequityneutralfor you know buying a hundred Apple or something.
TeslaTSLAequityneutralMaybe I asked you for. a quote of Apple and instead I'm quoting Tesla, whatever.

“Stance” is how the speakers talked about the asset, not a recommendation.

Claims worth verifying

Chloe Casale participated in the Mongol Derby, a 1,000km endurance horse race in Mongolia where participants swap horses every 35 to 40 kilometers.

Chloe Casale ·0:27

How to verify: Verify the Mongol Derby rules and format on the official Mongol Derby website or reputable news articles covering the event. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Chloe Casale suffered a stress fracture on her tibia during the Mongol Derby because she stuffed socks down her half chaps to protect against a bruise, which made it harder to bend her ankle.

Chloe Casale ·3:03

How to verify: This is a personal anecdote; verification would require medical records or a personal statement from Casale, which are not public. It is a verifiable claim in the sense that it is a specific factual assertion about her body and actions. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Only 23 people finished the Mongol Derby out of 43 who started camp.

Unidentified speaker (UNKNOWN) ·4:38

How to verify: Check the official Mongol Derby results archives for the specific year Casale participated, available on the Mongol Derby website. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Chloe Casale skipped chemistry in high school to take an auto tech class.

Chloe Casale ·6:41

How to verify: This is a personal biographical detail. Verification would require personal confirmation or biographical sources about Casale. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Chloe Casale took an intro computer science class on a $20 bet with friends.

Chloe Casale ·8:00

How to verify: Personal biographical detail; verifiable via personal confirmation or interviews. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Chloe Casale worked as a Forward Deployed Engineer at Palantir for four and a half years.

Unidentified speaker (UNKNOWN) ·9:54

How to verify: Verify via Chloe Casale's LinkedIn profile or Palantir's public employee directory/archives. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Chloe Casale worked at Palantir in Brazil, Indonesia, Mexico, London, and Seattle.

Ron ·15:04

How to verify: Verify via Chloe Casale's LinkedIn profile or Palantir's public employee directory/archives. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Jane Street uses two interviewers for every interview except the phone screen.

Ron ·13:47

How to verify: Verify via Jane Street's official careers page, hiring guides, or public statements from Jane Street recruiters regarding their interview process. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Jane Street acts as a liquidity provider, trading bilaterally with clients like pension funds to avoid adverse selection on exchanges.

Unidentified speaker (UNKNOWN) ·17:18

How to verify: Verify via Jane Street's public website describing their business model, or financial literature on market making and bilateral trading. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Chloe Casale had no background in OCaml or finance when joining Jane Street.

Chloe Casale ·12:46

How to verify: Verify via Chloe Casale's LinkedIn profile or resume, which should list her technical skills and prior work experience. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Chloe Casale previously worked as a Forward Deployed Engineer at Palantir.

Unidentified speaker (UNKNOWN) ·22:33

How to verify: Chloe Casale's LinkedIn profile or Palantir press releases/employee announcements from 2026-2026. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Jane Street has a dedicated team responsible for integrating with RFQ platforms.

Unidentified speaker (UNKNOWN) ·23:48

How to verify: Jane Street careers page job descriptions for 'Integration Engineer' or 'RFQ' roles; industry reports on Jane Street's technology infrastructure. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Some counterparties require quotes to be returned in less than 5 seconds.

Chloe Casale ·29:05

How to verify: Jane Street internal documentation or trading industry SLA standards for high-frequency or electronic trading services. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Bonsai is Jane Street's internal version of React built in OCaml.

Unidentified speaker (UNKNOWN) ·41:36

How to verify: Jane Street engineering blog posts or OCaml community publications detailing internal tooling. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Webs at Jane Street originated from a large spreadsheet used for pricing securities.

Ron ·44:19

How to verify: Jane Street engineering blog posts or interviews with Jane Street developers regarding the history of Webs. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Goldman Sachs uses a programming language called Slang and a data store called SecDB.

Ron ·43:22

How to verify: Goldman Sachs technical publications, patents, or public documentation regarding their Slang language and SecDB database. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

LLMs are non-deterministic, which creates risks in trading automation.

Chloe Casale ·31:44

How to verify: Technical documentation on Large Language Model behavior; academic papers on LLM non-determinism. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Jane Street clients use chat on Bloomberg terminals for trading.

Unidentified speaker (UNKNOWN) ·24:15

How to verify: Bloomberg Terminal feature documentation for chat-based trading; industry surveys on trading communication channels. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Regex parsers were used to automate standardized chat requests before LLMs.

Unidentified speaker (speaker_2) ·26:04

How to verify: Jane Street engineering blog posts or technical talks on pre-LLM automation in trading. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Webs allows users to iframe websites into the interface.

Chloe Casale ·42:59

How to verify: Jane Street internal user guides or demos of the Webs interface. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Chloe Casale was a Forward Deployed Engineer at Palantir before joining Jane Street as a desk dev.

Chloe Casale ·56:38

How to verify: LinkedIn profile of Chloe Casale; Jane Street career pages or press releases regarding her hiring. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Webs is an internal framework at Jane Street that allows sharing data and analysis, created by a team including DH2.

Chloe Casale ·47:29

How to verify: Jane Street internal documentation or engineering blogs; job postings for Jane Street developers mentioning Webs. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Jane Street hires generalists and uses multiple similar interviews to build confidence, whereas Palantir hired specialists with hyper-focused questions.

Unidentified speaker (UNKNOWN) ·1:07:41

How to verify: Jane Street and Palantir interview guides published online; articles by former employees of both firms describing their hiring processes. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Palantir's Forward Deployed Engineers were expected to feed insights back to core product teams to improve the main product.

Chloe Casale ·59:12

How to verify: Palantir job descriptions for FDE roles; blog posts by Palantir on their product development methodology. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Jane Street's desk devs often build tools that are later found to be generalizable across multiple desks, such as chat-to-RFQ parsers.

Unidentified speaker (UNKNOWN) ·1:02:10

How to verify: Jane Street engineering blog posts or internal case studies on tool adoption across trading desks. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Chloe Casale previously worked at Palantir as a Forward Deployed Engineer.

Chloe Casale ·1:33:19

How to verify: Check Chloe Casale's LinkedIn profile or Jane Street's official press releases/website for her employment history. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.

Do your own research

  • Review Jane Street's public hiring materials to understand their generalist vs. specialist evaluation criteria.
  • Examine legal precedents for 'meeting of the minds' in electronic chat contracts to assess enforceability risks.
  • Analyze latency benchmarks for LLM inference in production trading environments versus traditional regex parsers.
  • Verify the security and sandboxing capabilities of internal tools like Webs and Bonsai against external AI code generation.
  • Compare Palantir's FDE role structure and incentive alignment with Jane Street's desk dev model.
  • Assess the regulatory implications of using LLMs for trade confirmation and error handling in financial communications.
  • Investigate the integration costs and adoption rates of RFQ platforms versus chat-based trading for institutional clients.
  • Check for any public disclosures or case studies regarding 'fat finger' error rates in manual vs. automated trading workflows.
Full transcriptexpand
0:03
Ron: So I I'm pleased to introduce Chloe Casale, uh who's an engineer here who works on building tools for our sales trading team. So first of all, I just want to thank you for joining me and also how's your leg?
0:14
Chloe Casale: Thanks for having me. Um it's still in a moon boot. Um, so I got to get wheeled through the airport to come and get here today. Um tell
0:23
Ron: tell us more how you broke your leg.
0:26
Chloe Casale: Yes.
0:27
Chloe Casale: So I was doing this 10-day-long endurance horse race in Mongolia called the Mongol Derby, which was a ton of fun. It was the hardest thing I've ever done in my life. But basically, it's the premise is you fly out to Mongolia and then you have a thousand kilometers to cover over the course of this race. You swap
0:47
Chloe Casale: horses, random draw horses every 35 to 40 kilometers. So by the end of it, you're riding like 29 horses, I want to say, over the course of the race.
0:56
Ron: Through the Mongolian step.
0:57
Unidentified speaker (UNKNOWN): Through the
0:57
Chloe Casale: Mongolian step. And you're just given a series of navigation points on a GPS. So it's not like there's a trail. So you just you open your GPS and you made your best guesses the route. You get on a horse that's been trained by people who live in Mongolia, which are very different riders than I am. And then you ride through the step in order to try to get to your next point
1:17
Chloe Casale: in a reasonable amount of time because you have to finish on day 10, or else it doesn't count as finishing the race.
1:23
Ron: So what bet did you lose?
1:26
Chloe Casale: I don't even know what possessed me to do this thing. I think it's like this piece of lore that I had heard was like, ooh, there's this thing called the Mongol Derby, and like it's really, really hard. It's the longest, hardest horse race on the w planet. And I was like, ah, that sounds awesome. I do not have the resume for this thing. And so one year they opened up the ability to
1:46
Chloe Casale: go and like test in basically to this race via the thing that they called the Academy. And in my mind, I was like, Yeah, I'll just do the Academy. It'll be like a fun five days, whatever. I'll come home, never think about this again. Unfortunately, I finished the Mac Academy and I got into the race. And then I'm I was
2:00
Unidentified speaker (UNKNOWN): like, so
2:01
Ron: that's the bet you lost.
2:02
Chloe Casale: Yeah.
2:03
Unidentified speaker (UNKNOWN): And I
2:03
Chloe Casale: was like, oh, oh, okay. I think I have to do this now. So
2:07
Ron: do you have like a long history of horse racing before this? Nope.
2:10
Chloe Casale: Um, I like grew up around horses like as a kid. When I was in university, I uh worked on a dude ranch in like during my summer. So it's like I'd done horse things before, but it wasn't like I'm not a competitive writer. rider. So this is why I was like, oh, I don't I don't think I'm gonna get in. But I love backpacking and in remote places. And so I think really the thing that got me about the derby was the
2:30
Chloe Casale: navigation bit where it's like you have to navigate between points and you don't know where you're going and you might have to camp out on the step with a horse. And I just thought that sounded so cool. But it was, it was, it was very dangerous. Um I'm not sure I would recommend anybody do it who's not prepared to risk life and limb.
2:45
Ron: And and and speaking of risking limb, limbs, like you broke your leg.
2:48
Unidentified speaker (UNKNOWN): Yeah. How did that
2:49
Ron: happen?
2:49
Chloe Casale: I was it was so strange. silly. It was a stress fracture because when you're when you're racing, you're like your feet are in stirrups and I it doesn't matter.
2:59
Unidentified speaker (UNKNOWN): As I would expect, although I'm not
3:01
Ron: sure what I would expect in
3:02
Unidentified speaker (UNKNOWN): the
3:02
Ron: Mongolian
3:02
Unidentified speaker (UNKNOWN): step.
3:03
Chloe Casale: But and so I was my stirrup for some reason was banging against my shin for the entire race. So I had a massive bruise. Um and so one morning on day nine, I wake up and I'm like, oh I'm so smart. I'm gonna like shove socks up my pants to like protect myself from the bruise I was getting. Um, turns out that made it like four times harder to bend my ankle, which is a very important joint to be able to bend in order to
3:23
Chloe Casale: ride a horse. And so I got a stress fracture on my tibia from riding an entire day with literally just socks stuffed down my half taps. Um, and so I wake up on day 10, which is the last day of the race. I only have two stations left. So like normal on a normal day, you're going 90 kilometers. On this day, I only had 60 to go through and I couldn't
3:43
Chloe Casale: weight my leg.
3:44
Ron: You couldn't put any weight on
3:46
Unidentified speaker (UNKNOWN): your leg.
3:46
Chloe Casale: I could not put weight on my leg. So I was literally like, I I didn't even know what I was gonna do, but I knew that I if I could get on the horse, I was pretty sure I could finish the race.
3:57
Unidentified speaker (UNKNOWN): And
3:57
Chloe Casale: so I just hopped to the horse line from my tent or from the yurt on the last day.
4:02
Unidentified speaker (UNKNOWN): And
4:02
Chloe Casale: then they won't let you ride unless you can get on the horse by yourself. And so then I just like threw myself onto the horse. I got very lucky with the random draws in the last day. Um, very, very lovely horses. And then just managed to ride in the last 60 kilometers. I think I like lost so many places in the race just because it took me so long because I had to go so slow because I was in so much
4:22
Chloe Casale: pain. But I was gonna finish that race. It's like I only got 60 kilometers left. I've already had 940 kilometers like I'm finishing this race. Yeah. But I finished. Only 23 people finished the race out of the 43 people who went to start camp. So
4:38
Unidentified speaker (UNKNOWN): it was a huge
4:39
Chloe Casale: dropout.
4:40
Ron: Why are people dropping out mostly? Other than it's insane.
4:42
Unidentified speaker (UNKNOWN): I
4:42
Chloe Casale: mean it's insane.
4:43
Unidentified speaker (UNKNOWN): I'd say
4:44
Chloe Casale: that most people dropped out from heat exhaustion. This was the hottest year anyone had ever the race had ever run. And so people would heat you can't ride a horse. with heat exhaustion. And then after that was getting thrown from a horse and so getting like breaking an arm or something. And
4:56
Unidentified speaker (UNKNOWN): then you
4:57
Ron: managed to break a leg in a
4:58
Unidentified speaker (UNKNOWN): I broke a leg, didn't even get thrown. Yeah. I have
5:00
Chloe Casale: no cool story. And then I think the last way was people getting kicked on the ground. But yeah. It was it was yeah, it was I mean I was very proud of finishing, but it was definitely one of those things that you only ever need to do once, I think.
5:12
Ron: That sounds great. Just just once. Just
5:15
Chloe Casale: once. It was over, never again. So
5:18
Ron: it's anticlimactic to ask it, but I always like to ask about how people got into computer science. And now I feel like after learning about how you got into Mongolian horse racing. Yeah.
5:27
Chloe Casale: It's
5:27
Ron: it's hard it's har it's hard to make that trans transition, but I'm gonna do it anyway.
5:30
Chloe Casale: Like
5:31
Ron: how how how do you end up like,
5:34
Unidentified speaker (UNKNOWN): yeah,
5:35
Ron: I don't even know how to ask this question, but
5:36
Unidentified speaker (UNKNOWN): like
5:38
Chloe Casale: why is a Mongolian horse racer such
5:40
Unidentified speaker (UNKNOWN): as
5:40
Chloe Casale: a
5:40
Unidentified speaker (UNKNOWN): little bit of a little bit of a little bit now I feel like we're starting from the point of a Mongolian horse racer and then
5:44
Ron: the ordering is all different from that in
5:46
Unidentified speaker (UNKNOWN): real
5:46
Chloe Casale: life.
5:47
Unidentified speaker (UNKNOWN): But
5:48
Ron: yeah, like why are you a computer person? Like how'd
5:50
Unidentified speaker (UNKNOWN): that
5:50
Ron: happen?
5:51
Chloe Casale: I think these are just like very entirely parallel hobbies of mine.
5:58
Chloe Casale: I think on the one hand I love doing these extreme sports, and on the other hand, I think I just really enjoy knowing how things work. Like I just love understanding kind of like how does the world around us actually function? And so when I was first, you know, in high school, university, figuring out what I wanted to do, I was convinced I was gonna be uh an o probably a
6:18
Chloe Casale: mechanical engineer. I wanted to work on cars.
6:21
Ron: Cars are cool.
6:21
Chloe Casale: Cars are so cool. They get us all over the place, and I had no idea how they worked. And I think at some point I was like, I just want to know how these how these work and figure out, you know, how this like thing that I feel like kind of makes our world run actually like works under the hood. And so at first I thought I was gonna be an auto mechanic.
6:38
Ron: Did you take did you take like mechanic courses in high school?
6:41
Chloe Casale: Yeah. So I literally uh skipped chemistry my in high school in order to go and take an auto tech class instead. I had to drive to a different high school, just like a neighboring high school in my district. in order to go take this class because they had the auto tech course. Got to go learn about how cars work, how to how do engines work, how do you change your brakes, how do you change your oil. If you ever need anything done
7:01
Chloe Casale: on your car, I got you.
7:03
Ron: I do, but sadly it's about like the dumb computer stuff in the car, which is
7:08
Chloe Casale: honestly surprisingly easy to fix. If you ever have the opportunity to go and like watch someone work on a car, it's like you'll notice they
7:14
Unidentified speaker (UNKNOWN): have
7:14
Chloe Casale: like a computer. handheld computer that they have that they just like go and plug in, like under your dashboard. There's like a slot. And then it just like runs all the diagnostics for you and tells you exactly what to do on a lot of these modern cars. I'm convinced if I could just get my hands on one of these computers,
7:29
Chloe Casale: I'd never have to go to the mechanic ever again.
7:32
Ron: Okay, but but here you are, not a mechanical engineer, although Gsuet Get also employs mechanical engineers, but that's not what you
7:37
Unidentified speaker (UNKNOWN): do, Karen. here.
7:38
Chloe Casale: No, that's not what
7:38
Unidentified speaker (UNKNOWN): I do. here. How did
7:39
Ron: you end up in computer science?
7:40
Chloe Casale: I think my sophomore year of university, um, I, you know, was deep into my physics courses and I was hanging out with some friends who were all taking this introduction to computer science class. And I just remember they were complaining about how hard this class was. And I said or thought, but probably said out loud, something along the lines of, it can't possibly be that
8:00
Chloe Casale: hard. And my friends were like it is that hard. And they bet me $20 that I couldn't pass the class. And I was a broke college student and I needed that 20 bucks. Um, so I went and took this computer science course. Was
8:11
Ron: it was it really the 20 bucks or
8:13
Chloe Casale: it was the 20 bucks and the pride.
8:15
Unidentified speaker (UNKNOWN): Yeah.
8:17
Chloe Casale: I feel
8:17
Ron: like maybe it was more the pride.
8:19
Chloe Casale: It was a little bit of both.
8:21
Unidentified speaker (UNKNOWN): And
8:21
Chloe Casale: I went and took this computer science class, and I remember just all of those things that I got from like learning. how cars worked. I feel like I got in that class learning how to Java program like the simplest programs in the world. But being able to just type some things onto a computer and like hit run and then like have a game appear or something. I just remember thinking, like, oh, this is magic. Like
8:41
Chloe Casale: this is this is how computers work. This is really cool. And then completely I d just fell in love with it. I knew that's what I wanted to do. This was too late into my career. I'd at university really to change majors. Um,
8:53
Unidentified speaker (UNKNOWN): all
8:53
Chloe Casale: the way to computer science, because especially at the University University of Washington, they've got like a pretty long CS course. Yeah,
9:00
Unidentified speaker (UNKNOWN): they've got
9:00
Ron: a really good CS.
9:01
Chloe Casale: It's very, very good department. And I was not about to stay in university for five years in order to finish into that course. You
9:07
Unidentified speaker (UNKNOWN): had you had
9:08
Ron: you were like Mongolian hosts. You had to
9:09
Unidentified speaker (UNKNOWN): do that. Yeah, I had places to be
9:11
Chloe Casale: so and so I just pivoted my class to my major to applied computational mathematics sciences is what they called it at UW. Um and so it was like basically an applied math course where you got to spend some time programming. And so I had a lot of the prereqs for that already sorted out. And so I could graduate still in four years and get to do some computer science things.
9:33
Ron: I actually did a little bit of CS on the side, but more th more theoretical math than and applied math, but still even less relevant.
9:39
Ron: Okay. So then what did you do from there?
9:41
Chloe Casale: So then I think when I was applying for jobs out of university, I kind of just cast a really wide net. I ended up getting a job as a forward-deployed engineer, actually at Palantir afterwards, moved to New York to go and do that. And I was probably, I was at Palantir
9:54
Unidentified speaker (UNKNOWN): for
9:54
Chloe Casale: I think four and a half, almost five years, and really loved doing a forward-deployed forward deployed engineer. I think that for me it was like another kind of one of those moments where it wasn't entirely pure software. engineering. I got to spend a bunch of my time thinking kind of like applied math. It's like, how do I apply this computer science knowledge that I've
10:08
Unidentified speaker (UNKNOWN): learned?
10:09
Chloe Casale: Aaron
10:09
Unidentified speaker (UNKNOWN): Powell
10:09
Ron: So what is a forward deployed engineer? This is like, I think, like Palantir jargon.
10:13
Unidentified speaker (UNKNOWN): It's
10:13
Chloe Casale: Palantir jargon, it's become weirdly popular in the industry these days, which I find interesting. But at the time, a forward deployed For engineer at Palantir was basically there were two parallel engineering tracks, like one forward for deployed engineer, one product engineer.
10:27
Chloe Casale: At Palantir, you call this a Delta and a dev. So I was a Delta. Um, I don't remember why we called it that. But basically, the idea was our job was to take whatever our product team had built and go on site to customers and like actually make it work for them in solving their specific problems. So it was like a very specialized role in a way. Like I think
10:47
Chloe Casale: sometimes when you think of software engineering, you think about trying to solve like a problem. for many, many customers. And a forward-deployed engineer's whole job was to solve a problem for one single customer. And so you got to like go land on the ground, understand everyone's problems very, very deeply, and then try to figure out do I have to build something entirely new? What products do I have that are kind of supported by core Palantir
11:07
Chloe Casale: that I can use in order to go and solve their problems? And then how am I actually going to make this like valuable for the people who are there? And then you also had a responsibility of kind of learning things that the customer wanted and feeding that back to product. So like maybe when I'm working with a specific customer, I realize that, like, oh, this specific way of doing data analysis seems super powerful. And that's
11:27
Chloe Casale: probably not just them who wants this. It's like I get to go and feed this back to product and say, hey, I think this thing is generalizable. I think this is more interesting than like maybe just this customer. And then that's how a lot of those roadmaps would get formed. Aaron
11:37
Unidentified speaker (UNKNOWN): Powell
11:37
Chloe Casale: And
11:38
Unidentified speaker (UNKNOWN): do
11:38
Ron: you feel like this kind of forward-deployed role is something that just like match your interest and personality more? I feel like I know lots of engineers who are like, you know what I'd like to do? Anything that's not that. Like
11:47
Unidentified speaker (UNKNOWN): I would
11:47
Ron: like to do things that are building like very general purpose, you know, as kind of you know founded in like systems and theory questions as possible, and like, oh, users. I don't want to think about them. Aaron
11:59
Unidentified speaker (UNKNOWN): Powell
12:00
Chloe Casale: I think that's the only thing I want to do.
12:01
Unidentified speaker (speaker_2): And so I think I'm the opposite dev to that dev,
12:04
Chloe Casale: where it's like, I just love talking to users. Like if someone I think there have been v m various times as both a forward deployed engineer at Palantir, but also a software engineer at Jane Street where the easiest way to solve a problem was to send an email. And I'm more than happy to solve that problem that way. Like I don't necessarily need to reach for coding every single time. And I think it just does, I don't know, mash my personality and the things I enjoy doing.
12:23
Ron: So how did you end up coming here from Palantir?
12:26
Unidentified speaker (UNKNOWN): So I
12:26
Chloe Casale: think uh my time at Palantir was kind of coming to the end and I realized, like, okay, I just want to try something new. I've been here for a while and I didn't feel like I was getting a lot of growth. And I cast like a just really wide net when I was interviewing. I was totally planning on probably ending up at another startup, but Jane Stream kind of hit my radar and I was like, okay, might as well, might as well apply, might as well interview. And I I knew
12:46
Chloe Casale: nothing about trading. I knew nothing about functional programming or O'Camel.
12:49
Unidentified speaker (UNKNOWN): mean,
12:50
Chloe Casale: I I I knew all these. things were things that
12:51
Unidentified speaker (UNKNOWN): are.
12:51
Ron: We're trying
12:52
Unidentified speaker (UNKNOWN): to be very
12:52
Ron: clear. Yeah. We're trying to be very clear when talking to people that you can apply without knowing anything about trading or functional programming. So
12:57
Unidentified speaker (speaker_2): I was
12:57
Unidentified speaker (UNKNOWN): that
12:58
Chloe Casale: person.
12:59
Chloe Casale: And I think that when I was going through the interviews, I just remember being so impressed with the interview process and being like, Okay, people have clearly put a lot of thought into this. And by the time I was leaving Palantir, I was deeply involved in their interview process as well. So I was very much
13:12
Unidentified speaker (UNKNOWN): um
13:12
Chloe Casale: maybe like an interview process snob at this point. Where I was very much like, I had very strong opinions about the right way to hire. And I think a lot of those opinions were probably wrong. But I think coming to Jane Street, I could tell that people had put a lot of thought into the way that you were interviewing here. And to me, that was like a signal of a company that had it together and a company that cared deeply about its people. And that's really what convinced me to join.
13:32
Ron: Yeah, I don't know how good of a job we do, but we definitely put a lot of time and energy into it.
13:36
Unidentified speaker (speaker_2): There's
13:36
Unidentified speaker (UNKNOWN): a lot
13:36
Chloe Casale: of energy going into it. And that was very interesting. apparent, even as a candidate. Even just things like having two interviewers in the room and every single one of my interviews, other than my P1. So other than my phone screen.
13:47
Ron: Yeah. I was actually surprised to learn how rare that is.
13:50
Chloe Casale: Super rare.
13:51
Ron: Like it's almost always you're being interviewed by one person. And we for a very long time have interviewed with two people. Part of the idea is that you walk out of the interview room and it just feels like it's very useful to have another person who is. was there who you can like say, Wait, did you see what I saw? And maybe I was doing a bad job of driving the interview. And and we and that just it just feels a very important way of
14:11
Ron: like both learning how the specific interview went, but also for people to over time improve how they run interviews, getting that feedback seems really valuable.
14:19
Chloe Casale: Yeah. It's super expensive though, right?
14:21
Ron: Super expensive.
14:23
Unidentified speaker (UNKNOWN): I
14:23
Chloe Casale: think that's the thing. It's like I I just I remember just being so amazed. of like, wow, we're spending so much dev hours on this second interviewer. But it is, I mean, it's a thing I don't it's the thing I think about every single season. It's like, uh are we getting the value from this thing? And I think every single time I think about it, I'm like, I think we still are. Like it still does feel so valuable to be able to have that second pair of eyes, that second opinion.
14:43
Ron: Yeah, it's clearly better, but that's obviously the easy question. The question is not whether it's better, but whether it's worth it.
14:48
Chloe Casale: Yeah.
14:49
Ron: So one other thing. So you so you worked at Palantir in New York. Yeah.
14:53
Chloe Casale: But
14:53
Ron: actually at Genstry, you work in London.
14:54
Unidentified speaker (UNKNOWN): Yes. How did that
14:55
Ron: happen?
14:56
Chloe Casale: So I technically worked at Palantir in New York, but because I was a forward deployed engineer, I think I maybe spent four months in New York the entire time I was there. Oh
15:04
Ron: wow.
15:04
Chloe Casale: So I worked in Brazil, Indonesia, Mexico, London, and Seattle when
15:09
Unidentified speaker (UNKNOWN): I was at
15:09
Chloe Casale: Palantir.
15:10
Unidentified speaker (UNKNOWN): This is another
15:11
Ron: benefit of being a forward deployed to play.
15:12
Unidentified speaker (UNKNOWN): engineer.
15:13
Chloe Casale: Yeah, you don't have to have an apartment.
15:15
Chloe Casale: Your entire life is out of one single suitcase, which was really fun when I was like 20. I think I wouldn't enjoy it as much now.
15:21
Ron: Yeah.
15:22
Chloe Casale: So I think at some point in my Palantir rotation, I ended up in London and I just absolutely fell in love with London. And it was like just I think it was the place for me. I think I was I always like to say I'm too soft for New York. It's like New York is a very, it's a very busy place. People, there's a lot of energy here. It was really hard for me to go out and see. trees.
15:40
Unidentified speaker (UNKNOWN): Yeah,
15:40
Ron: I don't I don't understand people who don't like New York. It's like it's loud. It's it's sometimes kind of smelly. Yeah.
15:47
Unidentified speaker (UNKNOWN): It's
15:47
Ron: incredibly crowded. What's not to love?
15:51
Ron: I really do love New York, but I I do understand why people it's not everyone's jammed.
15:55
Chloe Casale: Yeah. And I think London was the perfect compromise for me. It was like still a big city, but it wasn't like quite as intense as New York. And so
16:01
Unidentified speaker (UNKNOWN): when
16:01
Chloe Casale: I was interviewing, I was like. no, I'm going back to London. So I tried to find a job that would land me in London.
16:07
Ron: Nice. Okay, let's talk a little bit about what you do here. So you work on the software to help sales traders.
16:13
Chloe Casale: Yeah.
16:13
Ron: At a very basic level. Let's start. What's a sales trader?
16:16
Chloe Casale: So a sales trader is one of our traders. The way I think about it is they're one of our traders who's working with someone who they know who's taking the other side of the trade, maybe. So like instead of trading anonymously on exchange, they're trading with like their clients, which maybe is like a pension fund or maybe like a big wealth management firm or something along those lines. And so a sales trader's job is
16:36
Chloe Casale: kind of twofold. It's like on the one hand, like they're responsible for having this client relationship, like understanding like, oh, what are the kinds of things this client is going to want to trade with me? How do I facilitate that trade and making it kind of come through James Jane Street? Um, and also they're responsible for like helping actually execute those trades once they do get here. So thinking about not just like capturing flow, but
16:56
Chloe Casale: also like getting the flow through our systems. So maybe that's thinking about things like a client might reach out to us over a chat message directly to one of our sales traders and ask to trade something with us or request a quote. And so our sales trader would be responsible for getting that into our systems, figuring, getting a price back from our prop traders and then sending it back to the client.
17:15
Ron: Right. So maybe a bunch of things to unpack here. So, first of all, like,
17:18
Unidentified speaker (UNKNOWN): what
17:18
Ron: does Jane Street do anyway, right? Writ large, like we're mostly liquidity providers. Yes. Other people have desires to want to trade this or that. And mostly our role in the market is to like sit in the middle and buy from one and sell to the other in lots of complicated ways To that don't just mean like posting a bid and an offer on an electronic exchange. And one of the ways we do this is through.
17:39
Ron: these bilateral relationships with various other institutions that want to trade through us rather than them just going directly to the markets.
17:48
Chloe Casale: Yeah.
17:48
Ron: Right. And maybe and and you said this thing about prop traders. What's a prop trader as opposed to a sales trader?
17:54
Unidentified speaker (speaker_2): So the way
17:54
Chloe Casale: I think about it is our prop traders are responsible for like understanding different asset classes. So like um a future or an equity or an ET. ETF and understanding like deeply how to price that thing. And so when I think about it from like my point of view as a desk dev sitting with sales, it's like our sales traders are the people who are trying to like kind of get in this bilateral flow,
18:14
Chloe Casale: understand this flow, and then our prop traders are the people who are responsible for thinking about like pricing this flow. Do we even want to take this flow on? Things like that.
18:22
Unidentified speaker (UNKNOWN): Right.
18:22
Ron: And the prop traders, in some sense, that's like the older role at Jane Street,
18:26
Unidentified speaker (UNKNOWN): where
18:26
Ron: you don't have this kind of client-oriented work. And your primary job is
18:32
Unidentified speaker (UNKNOWN): under
18:32
Ron: building trading strategies, understanding the opportunities out there, thinking about, you know, actually, are we making money doing this or not? You know, those kinds of issues. And that's like the kind of trad and sometimes thinking directly in a kind of very detailed moment-by-moment Moment by moment way about individual trades, sometimes building software that executes trading strategies, lots of different pieces. of that.
18:51
Chloe Casale: Yeah.
18:51
Ron: And the sales trader, on the other hand, is much more about understanding the relationship with a client.
18:56
Chloe Casale: Yeah.
18:57
Ron: And why does a client want to have a relationship with us? Why don't they just want to trade on exchange? What like what's the value add that you get from like hitting us up on some chat platform and trading with Jane Street that
19:08
Unidentified speaker (UNKNOWN): way?
19:09
Chloe Casale: I think maybe the easiest way to think about it is um a lot of these clients, they want to do very, very large trades. Like trades that maybe just like don't make sense to actually execute on the exchange because you're not going to be able to get the best price. The simplest example is you can imagine that maybe I, you know, I'm rebalancing my pension fund, uh, and I want to be able to put through like a bunch of like very large orders for very
19:29
Chloe Casale: specific ETFs. Maybe right now the open book on the exchange doesn't have enough um like sell orders for them to actually go through and like basically execute this whole trade. That's
19:41
Unidentified speaker (UNKNOWN): right.
19:41
Ron: Like there's this all this like big continuous electronic system
19:44
Unidentified speaker (UNKNOWN): where
19:44
Ron: people are constantly posting orders, bids, and offers, the willingness to buy or sell a given security for a certain size or less for a c a specific price or better.
19:54
Unidentified speaker (UNKNOWN): Yes.
19:55
Ron: And there's actually just a limited amount of liquidity that's out there at any one time.
19:59
Chloe Casale: Yeah.
19:59
Ron: And there's this like fundamental adverse selection problem, right? Which is
20:02
Unidentified speaker (UNKNOWN): like
20:03
Ron: if you're a prop trader and you have a system that posts orders all the time,
20:07
Unidentified speaker (UNKNOWN): a
20:07
Ron: lot of the time someone will come to trade with you who's like just someone who wants to trade in or out of something. Sometimes someone will come to trade with you specifically because they realize that the pricing in the market is off
20:18
Unidentified speaker (UNKNOWN): and
20:18
Ron: they are trading with you conditional on the fact that they think your prices are wrong. And so when you trade with those people, you lose a bunch of money every time. Yeah.
20:25
Chloe Casale: And
20:25
Ron: then you make a little bit of money.
20:27
Ron: just kind of collecting like the bid-ask spread, or half the bid ask spread every time you do a trade for everyone else. And so it means that people post less size than they otherwise would. Right. And they post wider spreads. They give out worse prices than they would, because they kind of have to make the money back from all the kind of ordinary back and forth facilitation trades to make up for the fact that
20:47
Ron: occasionally someone comes who knows a lot more than the market maker does.
20:51
Chloe Casale: Yeah.
20:51
Ron: Right. And then the dynamic is very different in a client.
20:54
Chloe Casale: Yeah. Because in
20:54
Unidentified speaker (UNKNOWN): a
20:54
Chloe Casale: client, in a way, it's like, I I know I have information about why you're coming to trade with me. Like I think sometimes when you're trading anonymously on exchange, it's like to your point, you don't necessarily know what the person you're trading with knows. But when you're trading with a client, it's very, very frequently I know exactly why you're coming to trade. Like I'm not super worried about you having a bunch of information and coming to pick me off because I know why it is you want to do.
21:15
Chloe Casale: this trade. And so for that reason, we can give better prices than going straight to the exchange and trying to trade there.
21:21
Unidentified speaker (UNKNOWN): Right.
21:21
Ron: And a lot of this also leverages the fact that we are like really good traders of a bunch of different kinds of derivatives. So there might be like an ETF, an exchange traded fund, where like you have a single security that is a proxy for a basket of things underlying underlings that are like m themselves maybe much more liquidly traded, but like not very that many people trade the particular ETF.
21:41
Ron: So it's not that thickly quoted. There's, you know, so it's hard to get like a big order off there. But if you like go to us and say, Oh, I have a really big order to do, we can do a good job of pricing the ETF off of the underlings. And then we can unwind the trade on our side. We can like, you know, we buy to you or we sell We to told you. We can like do the other half of that of getting rid of that risk
22:00
Unidentified speaker (UNKNOWN): in
22:01
Ron: the underlyings if there's not enough liquidity in. the ETF. Yeah.
22:04
Chloe Casale: And
22:04
Ron: like that's another part of the dynamic.
22:05
Chloe Casale: And that's a yeah, huge whole process that a lot of these companies they're not going to want to try to figure out how to build an in-house.
22:11
Ron: That's right. Right. Part of what
22:12
Unidentified speaker (UNKNOWN): you're doing is
22:13
Ron: there's there's a kind of simplicity of you can just go to someone and be like, I want to do this big trade. Can you figure it out? And we'll get like some extra fee complicit in the prices for it. But
22:22
Unidentified speaker (UNKNOWN): there's
22:23
Ron: a huge amount of expertise they don't have to develop.
22:25
Chloe Casale: Yes. Yeah.
22:26
Ron: And then I guess another thing is like why chat? Like it sounded like
22:32
Chloe Casale: that really surprised me when
22:33
Unidentified speaker (UNKNOWN): it was talking to Jane
22:34
Chloe Casale: Street. I think when someone was like, oh yeah, people are still sending me chat messages or sometimes even picking up the phone, I was like, whoa. Um but I think maybe the simplest version of why chat is uh you can imagine maybe the alternative is some sort of like electronic system where you could go in and type in maybe a request for quote. is a good example, an RFQ. So someone is requesting a quote, maybe
22:54
Chloe Casale: for you know buying a hundred Apple or something. And so
22:57
Unidentified speaker (UNKNOWN): And when
22:57
Ron: you say requesting a quote, what does it mean to request a quote? Like that's kind of also jargon.
23:01
Chloe Casale: Yeah. So I guess like uh if I was to come to you and I basically you're saying if I wanted to buy a hundred shares of Apple, how much would you sell that to me for? And that's the quote. So you're gonna going to give me back a price, probably per share, and then I'm gonna going to know whether or not I basically want to take that offer that you just gave me. And so that is a request for quote.
23:20
Unidentified speaker (UNKNOWN): Right.
23:20
Ron: And someone who's coming to trade might send out an RFQ to many different liquidity
23:25
Unidentified speaker (UNKNOWN): providers at the
23:25
Ron: same time and then pick the best price from those.
23:28
Chloe Casale: Yeah. Yeah. And so you could imagine like doing this process over what we call RFQ platforms. Um so those are, you know, companies, systems where you can go and just enter it in into like a roughly like a form and then hit enter and then it gets sent electronically. And so it's a very standardized protocol. It's lovely to work
23:47
Unidentified speaker (speaker_2): with,
23:48
Unidentified speaker (UNKNOWN): um,
23:48
Chloe Casale: especially when I start thinking about chats. But these like platforms, it's kind of hard to onboard to them. So it's like you have to build an entire system internally that can like have communication with these platforms. At James Street, we have a whole team who's responsible for thinking about integrating with RFQ platforms. And there's like a cost associated with onboarding to them. And so in
24:08
Chloe Casale: some ways, especially when we have these clients who are maybe doing very, very few trades a year, it doesn't make sense for them to build out
24:15
Unidentified speaker (UNKNOWN): all
24:15
Chloe Casale: of this integration with these RFQ platforms and doesn't make sense to pay for the cost of this thing when you can just send a chat across a Bloomberg terminal.
24:23
Ron: Although I would have thought like we have to do a lot of integration work because we want to hook a bunch of automated systems in. But I would have thought that like an end client would have to, you know, I don't know, install a thing, like open up their web browser and go to a URL and pull up the RFQ platform that way. Like, why is it more complicated than that?
24:41
Chloe Casale: I think sometimes it is more complicated than that because like you actually do want to integrate so that you can get the standardized protocol. I think sometimes it's not more complicated than that. You do just need to install the software, but you do have to pay for the software.
24:52
Chloe Casale: And that's kind of paying for something on top of maybe just like a Bloomberg contract.
24:56
Ron: And like a thing that maybe you're not going to use very often.
24:59
Unidentified speaker (UNKNOWN): So
24:59
Ron: it's not kind of not
25:00
Unidentified speaker (UNKNOWN): for itself.
25:01
Chloe Casale: And I think also sometimes these platforms don't have the exact kind of flexibility that clients want. I think that like a simple thing to imagine is sometimes a client wants to send off a request for quote, but not include the size. Maybe this is because they don't know the size. yet. Maybe this is because they don't want us to know the size yet. But some platforms don't allow you to send an RFQ
25:21
Chloe Casale: without a size field. And so they would have to either find a platform that allows them to handle this kind of custom flow is the way I call it.
25:29
Chloe Casale: Or they can just send a message over chat.
25:31
Ron: And also I guess chat is just older. It's older,
25:35
Unidentified speaker (UNKNOWN): It hasn't been
25:35
Ron: competed out, right?
25:37
Chloe Casale: Yes. Maybe one day.
25:39
Ron: So how do we deal with chat? Like that's obviously like kind of messy. Humans are typing in messages and then agreeing to trades, which are maybe like quite large. As you pointed out, some of the reasons
25:51
Unidentified speaker (UNKNOWN): people come to us is they're
25:52
Ron: very large trades. That sounds a little error-prone, like and hard to automate. So how do we how do we deal with that? Aaron
25:57
Unidentified speaker (UNKNOWN): Powell
25:58
Chloe Casale: It is hard to automate. Like I think that especially kind of going back to pre-LLM time. times, um,
26:04
Unidentified speaker (speaker_2): when we were
26:04
Chloe Casale: thinking about kind of automating chats, it's like some of our chats were like relatively easy to automate because you could tell that someone was like kind of like putting in a standardized format every single time they were sending us one of these requests for quote. So it was relatively easy for us to build a system that could read the regex, like could use a regex parser, parse out this request for quote, and then like send it through
26:24
Chloe Casale: our RFQ systems. And this felt relatively low risk. It was like we had very consistent messages coming from clients. We still had a sales trader looking at the chat room every single time, even though the parsing was happening automatically, a sales trader would still go back into the chat room in order to maybe go and send the price that we ended up quoting. And so we kind of
26:44
Chloe Casale: thought a lot about kind of just the automation of like getting this into our systems and like understanding, okay, what's like good reasonable regex. parsing here? I think a lot of doors feel like they've opened since like uh all of the access to LLMs that we have. I think now instead of you could imagine a world where instead of having to write some
27:04
Chloe Casale: specialized regex parser for every single counterparty we're trading with, you could just maybe write a specialized prompt, maybe just one prompt for one asset class. Like there's so many things. you can start thinking about here in terms of how can we start parsing these requests from clients out of chat into a more standardized way for our traders to look at.
27:24
Unidentified speaker (UNKNOWN): So
27:24
Ron: even in like the old regular expression world, it's maybe worth going through like what's the actual value add? Like if at the end of the day, there's a sales trader who's looking at the chat and thinking about it and like typing in the final price. Actually, why do we care about capturing the data? I mean, like the trader is there. They can see it with their own eyes. What's what's the benefit that we get by this kind
27:44
Ron: of automatic extraction?
27:45
Chloe Casale: I think some of it just is speed. I think that some clients do care about speed. Like we are racing in a way with some other brokers and we want to win on price, but sometimes, you know, people just really want to get their orders through very, very quickly. I think there is also a thing here where when you have very standardized parsing, it's like you can be pretty confident that you're not gonna have like a big fat finger mistake. Um, and
28:05
Chloe Casale: so even though we have a human kind of like double checking to say, like, oh, okay, these two things are the same, it's like a human kind of like quickly looking at you know, your parsed RFQ and your chat message is faster. And I might even argue less error prone than a human trying to figure out how to exactly Actually, type that thing over into our internal system. Aaron
28:23
Unidentified speaker (UNKNOWN): Powell
28:24
Chloe Casale: Sure.
28:24
Ron: In fact, I would expect it to not be close, right? If you I expect you'd you'd be much more likely to make mistakes in the hand entering of things. Like it's really easy
28:31
Unidentified speaker (UNKNOWN): to
28:31
Ron: like
28:32
Unidentified speaker (UNKNOWN): hit one digit
28:33
Ron: in the wrong place, put the decimal point in the wrong spot.
28:35
Chloe Casale: Yeah.
28:36
Ron: Can you like give even a little more detail? I'm just trying to understand like why would the speed matter to the end user there? Like what are they doing? At what time scale does speed matter? Aaron
28:45
Unidentified speaker (UNKNOWN): Ross Powell
28:45
Chloe Casale: Oh, uh this matters, I'd say like different counterparties have different amounts of speed. And it's also interesting because it's actually quite different every region, I'd say, how much people are like sensitive to speed. And so you have some counterparties who we're aiming really to try to get quotes, at least are trying to get quotes into our system
29:05
Chloe Casale: within less than 20 seconds. Some counterparties were trying to get it in less than five, um, so that we can basically have the time for it to go get a price. be able to write that price back into a chat. So this is like still like human level
29:18
Ron: for sure
29:19
Chloe Casale: processing, but it is if I think if we started taking, I don't know, 45 seconds on something where someone is expecting us to be back in five, like we might lose that trade. That being said, there are some clients who are sending us trades that they're expecting a price back within an hour. And so it's like very different people are kind of like executing different strategies on their side is the way I think about it.
29:39
Chloe Casale: I think some counterparties might be thinking about trades that are maybe a little bit more time sensitive, while others are thinking about things that are maybe closer to trades where you know you're like striking the price on the close or something.
29:50
Ron: Yeah, is it I wonder how much is it the actual time sensitivity of the trade they want to do and how much of it is like, well, I want to do this one.
29:57
Unidentified speaker (UNKNOWN): Well,
29:57
Ron: it's really annoying to have to wait 45 seconds for the next one. So I'm gonna like wait a little while and see what it comes back, and I'll choose among those. and just like a workflow issue.
30:05
Unidentified speaker (UNKNOWN): Yeah.
30:05
Chloe Casale: I mean, there is a human on the other side. And it's really hard, I think, in some ways on our side to guess necessarily as to why different people. Uh you can take some guesses as to why, oh, maybe they me might be more time sensitive versus this other counterparty. But at the same time, you're right. Like it might just be the human on the other side kind of got bored waiting. Might be the human on the other side wants to kick off their next trade. They don't want to have to
30:25
Chloe Casale: have be thinking about like five, 10, ten live trades all at the same time. Like there's so many reasons why someone might want to say, I'm ready to execute this thing.
30:33
Ron: I'm sure you spend a lot of time with our sales traders trying to understand their workflows and what's important to them.
30:37
Chloe Casale: Yeah.
30:38
Ron: How much time do you end up spending thinking about what the actual clients want and how that works and what we can learn from that?
30:46
Chloe Casale: I spend a lot of time figuring out how I can facilitate our traders. thinking about what their clients want. So I think that like they have the expertise of understanding, they're talking to these clients very frequently, they're seeing the flow go by, like they have hypotheses as to what's going on. And so when I'm kind of thinking about like the extent to I start thinking about our clients
31:06
Chloe Casale: is thinking about like, oh, what kind of questions do my sales traders want to be asking in order to figure out how to support like their clients? and then building systems that can kind of facilitate that question and answering?
31:18
Ron: Okay. So now we're in this new like large language model world. Like how does that change the parsing story?
31:24
Unidentified speaker (UNKNOWN): It's
31:24
Chloe Casale: kind of open the doors to being able to parse things that are maybe coming through in a less structured way. So it's like counterparties who are actually like typing things in by hand every single time and maybe are doing different formats. Maybe they have like very verypos. cons typos. You can imagine we can start handling. those things. At the same time, LLMs, they're non deterministic. I think that I get
31:44
Chloe Casale: nervous about what things that they're going to get right or wrong. And I think it's kind of hard to tell like how comfortable we should be even inserting them into a process. And a thing that I think about very frequently is like we look very frequently talk about like, okay, well, we can put a human in a loop, and then like a human can give us some sort of like fact checking or some
32:04
Chloe Casale: sort of saying, like, yes, this chat message looks like this parse. But when you're kind of thinking about when, when does that work? Like if your parser is right,
32:13
Unidentified speaker (UNKNOWN): I
32:14
Chloe Casale: don't know, 95% of a time, a human is kind of unlikely to notice that five percent of a time in which it was wrong because they're just so used to it being correct, they're just constantly kind of hitting the yes button. This
32:23
Unidentified speaker (UNKNOWN): is
32:24
Chloe Casale: just like human nature.
32:25
Ron: Alert fatigue.
32:26
Chloe Casale: Yeah. If the thing is right 25% of the time, the trader's just never going to look at it because it's always wrong. So kind of trying to have a thinking really hard about like, okay, these LLMs, I'm not going to make them artificially incorrect. Like
32:41
Unidentified speaker (speaker_2): that feels
32:42
Unidentified speaker (speaker_2): bad.
32:43
Chloe Casale: But also, okay, so let's say they're probably going to be right 95% of the time. When they're wrong, 5% five percent of the time, let's say, like what kind of risk. checks can we put in place? What kind of things can we do with our system in order to make us feel protected about like what that five percent might look like? Aaron
32:58
Unidentified speaker (UNKNOWN): Powell
32:58
Ron: I mean you could literally insert errors, not like make the LLM make a mistake, but like insert errors into the confirms at some rate and see whether people catch it. And when they don't, you're like, actually that was wrong.
33:09
Unidentified speaker (UNKNOWN): Yeah,
33:09
Chloe Casale: so this has crossed my mind.
33:13
Ron: I'm
33:13
Unidentified speaker (UNKNOWN): not sure
33:14
Chloe Casale: how
33:14
Unidentified speaker (UNKNOWN): much
33:14
Ron: you're your how much the sales traders would enjoy this process.
33:16
Unidentified speaker (UNKNOWN): Yeah,
33:17
Chloe Casale: I pitched it and I think the sales traders were a little less unhappy.
33:21
Unidentified speaker (UNKNOWN): And
33:22
Chloe Casale: I think at the same time it's like it's hard to figure out like how much is that actually gonna cause people to actually I don't know, bring up that extra scrutiny level. I think
33:31
Ron: And you could tell w how good they were at it.
33:32
Chloe Casale: Yeah.
33:33
Ron: It's tempting.
33:33
Unidentified speaker (UNKNOWN): It's
33:35
Chloe Casale: It's tempting, but I think that instead I think spending some time thinking about like how
33:38
Unidentified speaker (UNKNOWN): you're gonna have to be a good
33:39
Chloe Casale: do I cache those five percent? How do I get this closer to maybe ninety-nine percent? How do I make sure that when the five percent goes wrong, it's like at least easy for someone to notice in a way. It's like, can we have two different prompts parsing the same chat message? Do these things agree? Do they disagree? Like there's all sort
33:55
Unidentified speaker (UNKNOWN): of
33:55
Chloe Casale: ways that we can get creative to get ourselves a little bit more confident, at least caching those five percent that isn't necessarily relying on a person. Right.
34:03
Ron: And obviously you're in a world where the underlying technology is changing really fast. Yeah. It's also maybe not always changing in a way that's good for what we want, in the sense that one thing that happens now with models is they spend a lot more time chewing on and thinking about things. Yeah. And like you actually probably want them to be able to turn around and respond pretty quickly.
34:22
Unidentified speaker (UNKNOWN): Very fast.
34:22
Chloe Casale: Yeah. So we've been even thinking about, you know, some of our counterparties, like they just can't move on to LLM parsing because they just want their responses too quickly. And it's very hard to guarantee how fast that response is going to come from, versus other counterparties, like maybe it's okay to have the LLM like deep reasoning in order to get the exact symbol that this counterparty has asked for.
34:41
Unidentified speaker (UNKNOWN): Right.
34:41
Ron: And there's a question of like, do you give the LLM just like the one chat message or like the history of chat messages, where it might in fact be better able to build priors if it sees more stuff. but it's of course going to take longer to do its inference
34:53
Unidentified speaker (UNKNOWN): in
34:53
Ron: that world. Have you thought about like there are all these like fancy fast chips for doing much faster LLM inference that can get like, you know,
35:01
Unidentified speaker (UNKNOWN): a
35:01
Ron: factor of 10 or something faster inference at much higher cost?
35:04
Chloe Casale: I think we've kind of bounced around the idea. I think that like right now, I'd say that like our LLM, like even our LLM parsing just feels like it's very, very like first stage. Like I don't think we're at the point where we're trying to figure out like how to like optimize to get these things through very, very fast. I think we're still at the part of like trying to like fine-tune our prompts so that even when they're thinking for a very long time, like we're getting
35:24
Chloe Casale: the right thing out. I think that maybe at some point we'll get there to this like level of optimization. But I do think one of the nice things I think about working in client-facing Line facing tech tag is we are always kind of working on these human scales. And so I think that like we do kind of have some escape hashes.
35:39
Ron: And then when the LLM is wrong, what happens? What are the implications of mistakes in this context?
35:46
Chloe Casale: It's really complicated.
35:48
Unidentified speaker (speaker_2): It's like you
35:49
Chloe Casale: can imagine that the LLM, let's imagine like a worst case scenario where the LLM is wrong and no one notices that like our RFQ maybe got parsed completely incorrectly. Um we get a price back. Maybe I asked you for. a quote of Apple and instead I'm quoting Tesla, whatever. We get kind of a price back. And then
36:08
Chloe Casale: maybe like someone just copy and pastes the price in no context back into the chat room. I don't think that's going to happen, but like hypothetical world. It's like then there's kind of this idea of like there's been a meeting of the minds over chat, um, where it's like we've kind of decided together uh that this is this is the request for quote they asked for. This is the price that we gave them.
36:27
Unidentified speaker (UNKNOWN): Uh
36:27
Chloe Casale: someone on their side says done. And then someone on our side says done. And then we need to go and execute that trade. And we are working with clients. So they're not necessarily adversarial. Like you can imagine a world where we send them some crazy price back and someone's gonna respond and just be like, really? Um,
36:43
Unidentified speaker (speaker_2): like, and
36:43
Chloe Casale: so it's like there are kind of moments where like someone might notice that something looks a little bit wrong, but it is possible that like we send back the wrong price and like we've agreed. to the wrong price, and then we make a trade
36:53
Unidentified speaker (UNKNOWN): on
36:53
Chloe Casale: a price that we didn't necessarily, you know, want to trade at. And there's, you know, relationship cost for going to a client and saying, Hey, I know I agree to this trade trade with you, but we actually can't do it. And that's like kind of a thing that a sales trader has to be thinking about when we like maybe run into a mistake like this is like, what is the cost of going back to this counterparty and saying, we can't do this? Is it too heavy? Should we just like kind of
37:13
Chloe Casale: like take the loss on this trade? How are we thinking about this?
37:16
Unidentified speaker (UNKNOWN): Right.
37:16
Ron: One thing that's worth noting is that even when both sides say done, which
37:19
Unidentified speaker (UNKNOWN): is
37:19
Ron: like you know, a weird bit of trading jargon for like now the trade is settled, it doesn't actually mean the trade is done.
37:24
Unidentified speaker (UNKNOWN): Which
37:25
Ron: is to say if it's like just obviously a screw up, then like people will go back and be like, oh no, that's not what we meant.
37:31
Chloe Casale: Yes.
37:31
Unidentified speaker (UNKNOWN): And
37:32
Ron: fix it.
37:32
Chloe Casale: Maybe.
37:33
Ron: Maybe.
37:33
Chloe Casale: Maybe. I think it's it just kind of depends on like what the cost of doing that is, right? Because it's like
37:38
Ron: you
37:38
Chloe Casale: could imagine that like I go back, maybe I noticed the mistake, I don't know, 30 minutes later, and now the price has moved. It's like there's like a whole thing here of like we have made a mistake in this scenario and like we have to ask ourselves like what is our obligation to our counterparties in that moment.
37:52
Ron: That's right. And there are relationship questions there, and you don't want to like impose like weird things on your clients and
37:57
Unidentified speaker (UNKNOWN): things
37:57
Ron: like that. On the other hand, if it's like, you know, the price is off by a factor of a hundred from what you're
38:02
Unidentified speaker (UNKNOWN): gonna take. taking.
38:03
Ron: Which whether a hundred sounds like a crazy amount for it to be off, but it's actually a totally ordinary kind of mistake. It's like, was that meant to be quoted in pence or pounds?
38:09
Unidentified speaker (UNKNOWN): Yes. And
38:10
Ron: like you could just be off by a factor of a hundred and everyone could be very confused. But then in that world, everyone understands it's erroneous.
38:16
Chloe Casale: Yeah.
38:17
Ron: And and your use of the word meeting of the mind is like that's like not accidental terminology, right? Part of what's going on is you're like forming a contract here.
38:24
Unidentified speaker (UNKNOWN): Yes. We're forming a contract.
38:26
Ron: Whether or not there is a meeting. of the minds is kind of like part of the legal standard of whether like that's like the real thing or not.
38:31
Unidentified speaker (UNKNOWN): Yeah.
38:31
Chloe Casale: It's like whether or not we've kind of made this trade.
38:33
Ron: So are there things that we can do on the technical side to like reduce the chance of people being confused, even when we have a misparse?
38:41
Chloe Casale: For sure. So I think one of the things that we think about is uh giving our traders specific strings that they can copy and paste back into that chat message that are non-ambiguous. Um, so like let's say someone gets a price back. And instead of just sending the price, it's like we can say exactly what it is we're going to do. Like, hey, this is our price for buying a hundred
39:01
Chloe Casale: shares of Apple at this settlement date. And then like send that through exactly. And so this means that even though maybe our counterparty Party didn't read it, like there are still, you know, obviously human mistakes that can be made, it gives us all like a better opportunity for everyone to kind of catch the mistake as it's being made and our external counterparties to be able to say, oh, okay, like I see this thing.
39:21
Ron: So if you've gotten it wrong, we're at least explicit.
39:23
Chloe Casale: We're at least explicit in what we did wrong.
39:24
Unidentified speaker (UNKNOWN): In some sense, we put
39:25
Ron: in the same information into the chat that we use in like calculating the price that we gave them. So even if we calculated the price for the wrong thing, we've at least told them that we've done
39:34
Unidentified speaker (UNKNOWN): the
39:34
Ron: wrong thing. And that part's not some text generated by an LLM. That's like some hard system.
39:39
Ron: that understands what's being done and like has an explicit transformation to present that in a way that is unambiguous.
39:45
Unidentified speaker (UNKNOWN): Yeah.
39:45
Chloe Casale: And we work with our traders on like defining what is unambiguous in this context. Because I think as a dev, like I don't have necessarily the context of knowing what an unambiguous trade is. But when we go and have conversations with our traders, I can say like what exactly needs to be in the string so that everyone knows what it is we're agreeing to. And
40:00
Unidentified speaker (UNKNOWN): that's right.
40:01
Ron: You might think, oh, it's the the name of the security. and the price and the size. But you mentioned settlement date.
40:05
Unidentified speaker (UNKNOWN): That's
40:06
Ron: like one example of a weird thing of like when do the actual shares l really trade change hands, and that can affect what's going on. Like what other weird things do you want to sometimes need to stick in there?
40:16
Chloe Casale: Oh, I don't even know.
40:19
Unidentified speaker (speaker_2): I think that
40:20
Chloe Casale: um I think options are like particularly interesting, especially when you start trading complex options. Um it's like you have to
40:26
Unidentified speaker (UNKNOWN): where it's
40:27
Ron: like a single trade in a basket of multiple options and maybe multiple options and the underlying.
40:31
Chloe Casale: Yeah, exactly. And so like those symbols can look crazy. Um and so being able to understand exactly what kind of symbol needs to go through, like where exactly the plus and minuses signs go.
40:40
Ron: Oh, right. There's horrible quoting conventions of like the price could be like positive or negative depending on how you try to decide to quit quote the basket.
40:48
Chloe Casale: Yes.
40:49
Unidentified speaker (UNKNOWN): And
40:49
Chloe Casale: then you also have to think about
40:50
Unidentified speaker (UNKNOWN): all
40:51
Chloe Casale: sorts of things I barely understand. Um, but it's like quoting things, you know, maybe like using delta risk versus some other kind of risk. It's all these things you have to be thinking about when quoting options. And so like
41:00
Unidentified speaker (UNKNOWN): the
41:00
Chloe Casale: fact that I have an expert that I can go and talk to makes my job much, much easier.
41:04
Ron: Awesome. Okay. So chat parsing is obviously an important thing, but you're also building tools for people. And building
41:12
Unidentified speaker (UNKNOWN): tools
41:12
Ron: for people involves building user interfaces.
41:14
Chloe Casale: Yes.
41:15
Ron: So say a little bit more of like
41:17
Unidentified speaker (UNKNOWN): as
41:17
Ron: an engineer at Jane Street, what are the options in front of you of how you might build a UI for a sales trader?
41:22
Chloe Casale: Yeah. So I think that there's a couple of things. I think one way that you might build a UI for a sales trader is through our do we call bonsai a framework?
41:33
Ron: I guess. Yeah. It's like a web framework.
41:36
Unidentified speaker (UNKNOWN): Whatever you
41:36
Ron: call React.
41:37
Chloe Casale: Yeah.
41:37
Unidentified speaker (UNKNOWN): So
41:38
Chloe Casale: our internal reaction. version of React, I guess, which we call bonsai, that feels a little bit mean to bonsai, honestly.
41:43
Unidentified speaker (speaker_2): I think that
41:44
Chloe Casale: bonsai is like a super powerful way of building front ends in OCaml. Also, I'd say is an expert way of building front ends. I think that in order to like build a front end in bonsai, like you really have to you have to know bonsai and you have to know OCaml and you have to be like willing to put in the time to build a proper UI. Sometimes when you're working with sales trader on the desk, you don't necessarily need an entire proper
42:04
Chloe Casale: UI that's built out. Maybe you just need something that like gives someone like a quick view of their data or um maybe just like a quick dashboard that they're only going to need for four weeks and then like we can tear it down. And so another way that you can build a UI for our sales traders or a lot of people at Jane Street is something that we call internally webs. Webs is
42:24
Chloe Casale: exceptionally difficult to describe.
42:27
Chloe Casale: Because I think you can think about it kind of as like a distributed computational graph, maybe.
42:33
Unidentified speaker (UNKNOWN): I think
42:33
Ron: that's fair.
42:34
Unidentified speaker (UNKNOWN): Okay.
42:34
Ron: That's one of the things it is.
42:35
Chloe Casale: You could also think about it like everything Excel wanted to be.
42:39
Unidentified speaker (UNKNOWN): Like
42:40
Chloe Casale: it's like this amazingly powerful tool where you can build what I would call called UIs, um, where you can display data, usually tabular data. but so much more than that. Like I've seen people iframe in websites into webs before. Um and so it's really possible to create like these very
42:59
Chloe Casale: powerful, very powerful UIs and webs.
43:02
Ron: And it's maybe worth saying like this is a weird kind of technology that shows up in some form or another in lots of different trading firms. Yeah. Right. So if you like Goldman Sachs famously has a system called slang. Slang is their programming language. And there's like some underlying data store called SecDB, securities database, where they build all sorts of applications on top of
43:22
Ron: it. And a thing that you mentioned like graph, like why is there a graph there? Like a thing that has shown up for a long time in trading systems is you have streaming data that's coming in. And then you want to build some, you want to compute something from that streaming data. But
43:38
Unidentified speaker (UNKNOWN): you know,
43:38
Ron: the streaming data doesn't, everything doesn't change all at once. Little parts of it change, and then you don't want to have to do all the computational work of recomputing all of it all at the same time. And so you have some kind of graph that gives you like some structure of the computation. So only the parts that need to be recomputed actually are. It's like a form of incremental computation. And also spreadsheets like totally
43:58
Ron: do this, right?
43:59
Chloe Casale: Yes.
43:59
Ron: You know, spreadsheets of this like fun functional programming language. Every cell is a little purely functional expression. Yep. It expresses dependencies on other cells. That induces a computation graph. Yep. And webs literally developed from what was just like an overgrown spreadsheet, right? We had big spreadsheets on one of the desks that were used for like pricing a bunch of their securities
44:19
Ron: and webs kind of in fact that spreadsheet was called webs eventually. And then we've built a system in the image of that spreadsheet. Um
44:27
Ron: which takes that same basic structure and kind of industrializes it and makes it bigger and more scalable. And maybe one of the things that's different from what you would normally expect from a traditional spreadsheet is it's much more of a collaboration tool. Like the big computational graph is shared among a whole bunch of
44:47
Ron: people. And so it's both a way of building some new computation.
44:51
Ron: and it's also like a playground where like all the other computations that other people have done are just sitting there available for you. And then it has a kind of built-in little GUI builder in it, like
45:00
Unidentified speaker (UNKNOWN): an
45:00
Ron: easy way of building views. Like traditional spreadsheets have this like very flat grid view that you can embed all sorts of things in. But no, this is like, even though there's lots of tabular data, you have like a little programming language for constructing views as part of webs. Yeah.
45:15
Chloe Casale: And I say those programming languages, like very approachable, which I think is like one of the powerful things about webs, is like it's very easy to basically maybe take someone who's a little bit less technical of a user and like give them the basics of how to build a new view in webs and have them go up and be able to create a view that is exactly what they want. And I think kind of what to one of your earlier points, like webs is this like it's
45:34
Chloe Casale: allows you to share so much analysis, but it also allows you to share data. So once you have one piece of data in webs, anybody else who's using that webs instance can go and basically access that piece of data in order to like run whatever analysis that it is that they want to do. And so when we're spinning up a new webs instance to actually think about using and recruiting, I think one of the conversations we had with a lot of
45:54
Chloe Casale: our recruiters is like webs is more powerful the more brave you are. It's like the more frequently you're willing to go and try to build a new view, other people can build on top of it. that view that you've built. And like these things, like they really compound in terms of like, I added something and then you added something, and then some this other person added something, and all of a sudden you have this UI that's like way more powerful than it would have been if everyone was too afraid to
46:14
Chloe Casale: touch webs.
46:15
Ron: Oh, that's really cool. So is is is recruiting on board or are you guys starting to actually do this?
46:19
Chloe Casale: Yeah, we are. Yeah. So recruiting's on board. We've got a couple of views that are in webs now, and we are starting to do this. And I think it's been really cool to watch. especially webs has and its UI has a really great integration with AI. And so it's like you can basically kind of like hit a hotkey and then like you'll have a little code assistant basically appear. And then you can have a chat with your AI and say,
46:39
Chloe Casale: hey, this is the thing that it is that I want to build. You'll get a webs view from that. And this has been really great for onboarding a lot of our recruiters to being able to write their own webs views of the data analysis that they want to be able to run. and so that they can dig into the data that they've had access to without necessarily needing a whole dev to get involved in order to pull that data for them. Great.
46:58
Ron: So you get to empower the people more directly,
46:59
Unidentified speaker (UNKNOWN): just kind
47:00
Ron: of build the things that they want.
47:01
Unidentified speaker (UNKNOWN): Yeah.
47:01
Chloe Casale: And
47:01
Ron: I guess another nice thing about webs is it's not just like a language and a kind of runtime for this, but also like a deployment story. Like you go, just like with Excel, you just go and like put stuff in Excel and w when does that start working? Immediately.
47:13
Chloe Casale: Now. Yeah. That's
47:13
Unidentified speaker (UNKNOWN): right.
47:14
Ron: And it's sort of the same thing. that you add a view and webs, it just like locks in and hooks into the production system right there, and you can immediately extend it.
47:20
Chloe Casale: Yeah.
47:20
Ron: Okay. So we can we can do like full-on web applications in our own super weird alternate world
47:26
Chloe Casale: Bill
47:27
Ron: Camel-based web programming thing.
47:29
Chloe Casale: Yep.
47:30
Ron: You can do more or less the same thing in this like widely shared like graph of data and computation thing called webs. with its own also super weird custom programming language that DH2 created.
47:42
Chloe Casale: Yep.
47:43
Ron: What other options are there?
47:45
Unidentified speaker (UNKNOWN): Aaron Powell
47:45
Chloe Casale: I think the biggest other thing that at this point I see people doing is just uh going and vibe coding up just a UI? Aaron
47:52
Unidentified speaker (UNKNOWN): Powell
47:52
Ron: Just like an index.html within
47:53
Unidentified speaker (UNKNOWN): the website. Index.html and some
47:55
Chloe Casale: JavaScript that just, you know, they just you know push up to a location where everyone on the desk can see, and people just can start using. it basically immediately. And I'd say that like I kind of said these things in that order because I think that like the this is kind of the way that I think about like maybe productionizing something. It's like
48:12
Unidentified speaker (UNKNOWN): although in the opposite direction.
48:13
Chloe Casale: Opposite direction, yeah. Where it's like you'll have a trader maybe have an idea and then like very quickly vibe code something up just to like maybe see what it is. And then we can like very quickly kind of like iterate on like, okay, like how valuable is this thing? How many people on the desk are using this thing. Does this thing need to be productionized, or is it kind of just like a read-only thing that doesn't need like a bunch of dev scrutiny? Um, and then
48:33
Chloe Casale: from there, it's like we can kind of start having that conversation of like, oh, maybe this thing could live in webs and like continue to make webs more powerful, to my kind of a point earlier. Or maybe, no, we think this thing is going to be, you know, very critical for a bunch of things in trading. We want to have like but also we need like a very custom UI in order to make it very easy for people. to interact with. And that's kind of the point in which we would start moving
48:53
Chloe Casale: over into a bonsai UI.
48:55
Ron: So obviously AI is like inflecting all of the different choices here.
48:59
Chloe Casale: Yeah. Right.
49:00
Ron: And you know, you can, in fact, you can use AI tools. We sort of mentioned it for the latter two options. But for option number one, you could also like have an AI write a bonsai app for you. And like the skills and tools around that are getting getting better all the time. How do you think about the kind of safety trade-offs of how? having, especially less technical users who maybe don't understand the system that well and maybe
49:19
Ron: aren't really coders at all, use this stuff and have LLMs generating things for them that they actually don't understand and that nobody understands and no human has read?
49:29
Chloe Casale: Yeah.
49:32
Unidentified speaker (speaker_2): I
49:32
Chloe Casale: think, especially on the sales desk, a thing that we've been thinking about a lot is how do we think about varying our scrutiny? And I think this is already has always been like a skill that we expect depths to have. It's like some things need very, very, very high scrutiny and you really need to stare at it in order to believe it's working exactly the right way. And maybe some things you can like maybe have a little bit less scrutiny on, because maybe this thing is like not super very important, or maybe this
49:52
Chloe Casale: thing is only gonna be used once for something that maybe doesn't super matter. And I think when we've been thinking a lot about like vibe coding our apps on sales, a thing that we've been thinking about is like, what are things that feel very safe to do with a system that we don't fully understand? And what kinds of things feel not safe to do with a system that we don't fully understand? And so maybe like an example of one is like
50:12
Chloe Casale: we want to pull some data to give ourselves a sense check about whether or not any clients that previously were trading with us have stopped trading with us. So this is like a read-only dashboard. It's not going to cause any trading decisions to change. We're not going to go and publish this data. to any regulators. All sorts of things about this make this feel like a very easy thing to vibe code up. But if
50:32
Chloe Casale: any of those things that I just listed were not true, it's like now all of a sudden it's like we really do want someone to start thinking about the code and start like having some more scrutiny on like, okay, do we understand exactly what this thing is doing? Is this pulling data in the correct way?
50:45
Ron: But then how do you actually deal with knowing what it can do if no one has read the code? Aaron
50:51
Unidentified speaker (UNKNOWN): Powell
50:51
Chloe Casale: I've never really considered that someone has like built up a dashboard that was then doing something, like very completely unexpectedly, like trying to like send a request off to like a database that like it wasn't prompted to do. I guess it could do this thing. Aaron
51:07
Unidentified speaker (UNKNOWN): Powell I
51:07
Ron: mean, if Hugging Face is any
51:09
Unidentified speaker (UNKNOWN): example, like there's lots of
51:11
Ron: I actually do think there are real things to worry about though. there. In part, even if even if you don't if you sort of ignore the like you know, model with its cyber capabilities, guards turned off, running wild and hacking into things. There's just the model's trying to be helpful
51:28
Unidentified speaker (UNKNOWN): and
51:28
Ron: not having a lot of context and knowing what's actually reasonable to do and what's not reasonable to do. And so and like the model doesn't necessarily know what's reasonable to do and not reasonable to do. And the person driving the model doesn't
51:40
Unidentified speaker (UNKNOWN): maybe
51:40
Ron: doesn't know either.
51:41
Chloe Casale: Yeah.
51:42
Ron: So one thing that's nice about the index.html thing is it's actually the whole world has already been built up under the assumption that you can't trust web pages. So there's like sandboxing built in and limitations there. And then when it does le you know reach out over the network, we can actually can and put in various kinds kind of controls over like what
52:03
Ron: calls do you make available to like any comer under the assumption that like ye random web page is actually not fully trustworthy. Yeah. Even if it is like authenticated as the person who's doing it. Maybe the person who's logged into that web page isn't even the person who wrote it. And you don't want to like misuse their credentials.
52:20
Chloe Casale: Yeah.
52:21
Ron: And so we like, we have like tools in that place.
52:25
Unidentified speaker (UNKNOWN): And
52:26
Ron: then like but then as you think about like, but what you throw into a web instance, well, like anything that web instance can do, you're maybe worried about that. And if you have someone vibe coding like just like a full-on program
52:36
Chloe Casale: for
52:36
Ron: something, you have more problems of that shape.
52:39
Chloe Casale: It's problems that I at least I feel like I don't have good answers to at this point. Like I think that it it very much feels like uh, you know, you're kind of like trying to keep up with like all of these advances in AI. And I think especially as like a dev, it's like I don't want to slow down all my users. Like it's like I
52:54
Ron: like
52:54
Chloe Casale: they have access to these really powerful tools and they can get things done that they weren't able to get done maybe even two months ago. But also, I some of the things I do find kind of scary, and it's hard, it's I don't know. I think it's one of the things that I've been thinking about a lot. Even I mean, even as simple as Is like, we have a lot more code review to do as devs these days.
53:13
Chloe Casale: And so it's like there's a thing where it's like more code is just being generated by AI. We have more code review to do. I don't know about you, but I I sometimes I get a little tired. And I think there's a thing where how do I keep my scrutiny up on the code review, especially when I'm code reviewing code that maybe I'm the first dev to have looked at? I think it's all really hard.
53:30
Ron: Yeah. I mean, there's a lot of things that we're thinking about here and doing. There's a lot of work on sandboxing of agents to help deal with this kind of issue. An idea that I've been thinking about just in the last couple of weeks is to give safe pathways for vibe coding like whole applications where you don't have good sandboxing at the application layer. Yeah.
53:51
Ron: By doing stuff where you have like a separate document that specifies, in some sense, what the guards are supposed to be.
53:57
Unidentified speaker (UNKNOWN): And then
53:57
Ron: have maybe you could have like hard semant semanti semantic checks. Like you could like say it's like, oh, this document is there. So we have to build this in a way that like there'll be sandboxing built in that will enforce the rules that are stated there. Maybe simpler and easier to implement, although having its own limitations of having mandatory LLM review so that when you make a
54:17
Ron: change, you have like another round. Like the LLM that
54:19
Unidentified speaker (UNKNOWN): you're
54:20
Ron: trying to work with that is trying to solve your problems that might do creative and problematic things. But then you have like another separate session. whose only goal is to like look over the diff, look over the spec, and to try and make sure that the spec isn't being violated.
54:33
Chloe Casale: Yeah.
54:34
Ron: So that like even if a human isn't checking that this very limited form of verification that you need the LLM to do, maybe it can do a good enough job of that that you can prevent people from making mistakes. And then that little file that specifies the bounds, that you would actually have to have a person review.
54:49
Chloe Casale: Yeah. We kind of have a similar workflow. that we spun up on uh sales where when a person goes and vibe codes out their app, they then go and build something, basically a document that explains what it is that they think their app is doing. And then we have a skill that runs on both the app and the document that lets them know how much scrutiny we think should be applied to this thing. And
55:09
Chloe Casale: it will
55:09
Unidentified speaker (UNKNOWN): like say, oh
55:10
Chloe Casale: hey, I think you should pull in a dev, or I think this thing is fine for it just gets spun up under. vibe coded. And so we have seen like some adoption of that, which I think has been actually quite useful because it forces someone to like kind of take their app, go and think about what are the actual requirements of the thing I'm building, what is it actually doing, and then has an LLM go and say,
55:30
Chloe Casale: Hey,
55:31
Unidentified speaker (UNKNOWN): I
55:31
Chloe Casale: think you should ask a dev, or I think this is super safe and I wouldn't worry about it. But I think having kind of some LLM L LM jump in. has been like nice there, especially because it's hard for us to read every single app that's getting vibe coded out. But having a skill that has been a little bit trained on how to think about these things has been nice.
55:50
Unidentified speaker (UNKNOWN): It
55:50
Chloe Casale: doesn't get it right all the time. It might be overly conservative, but I think I prefer it that way.
55:54
Unidentified speaker (UNKNOWN): Yeah.
55:54
Ron: And I I think the technological choices here are really hard to work through. Like the other thing that we're like actually just starting to do is we've started hiring a team of people to work on formal methods. so that you could actually like prove properties of your program in a way that, you know, you gets you a much higher level of confidence that that the conclusions are right. And I think that will also be useful in some places. But actually I think the
56:14
Ron: trade-offs here are actually really hard to figure out. And like we're trying to c figure out things that we can build that will work across a range of possible future model capabilities. Right.
56:24
Unidentified speaker (UNKNOWN): Because
56:24
Ron: like the future is very hard to handicap. You know, models could vary quite a lot in how strong they are in a year or two or five.
56:30
Chloe Casale: Yeah. Even just watching the advances in the past one year, it it's it's crazy.
56:35
Ron: It's really weird. Yeah. And our world is changing really fast.
56:38
Chloe Casale: Yeah.
56:38
Ron: You mentioned before that you're a desk dev, right? Uh and at Palantir, you were an FDE, a forward Or deployment deployment deployed engineer.
56:46
Chloe Casale: Yeah.
56:46
Ron: Those sound kind of similar. Like desk dev is the term that we use for basically a team that is in embedded with a trading team of one kind or another.
56:53
Chloe Casale: Yeah.
56:54
Ron: I'm curious, like what you know, you've now seen similar ish-shaped roles here and somewhere else. And like
57:01
Ron: how do they differ? I feel like what, you know, what is it, what is it like in different places doing this kind of deeply embedded role.
57:07
Chloe Casale: Yeah. I think the biggest difference between doing the FDE thing at Palantir
57:12
Unidentified speaker (UNKNOWN): and
57:12
Chloe Casale: I think being a desk desktop dev here is like
57:14
Unidentified speaker (UNKNOWN): you're
57:14
Chloe Casale: your incentives are just very aligned with your customers in a way. Um, like here at Jane Street, my customers are the traders that it is that I work with.
57:23
Ron: Mm-hmm.
57:23
Unidentified speaker (UNKNOWN): The
57:24
Chloe Casale: thing that they're trying to optimize for is, you know, Jane Street. And the thing that I'm trying to optimize for is Jane Street. And so there's this thing where our incentives are very aligned. Having all sorts of conversations with traders, I find so much easier than I did with customers.
57:39
Chloe Casale: Because you can have these very open and honest and frank conversations with like a trader at Jane Street about things that are like easy, things that are hard, things that I think are going to take a lot of time versus a little bit of time. And like you can prioritize things together in a way that when you're working with like an external customer, part of you know, my job at Palantir is to sell Palantir.
57:59
Unidentified speaker (UNKNOWN): Right. The
57:59
Chloe Casale: job of the external customer is not to buy Palantir.
58:03
Unidentified speaker (speaker_2): It's to like do their job the best they can do.
58:06
Chloe Casale: And so I would have to think a lot harder about okay, well, what kinds of things
58:09
Unidentified speaker (UNKNOWN): am
58:10
Chloe Casale: I bringing to a customer? Like what kinds of things do I want to present as easy? Are those things actually easy? Wasn't necessarily always true. And kind of having those incentives
58:20
Unidentified speaker (UNKNOWN): a
58:21
Chloe Casale: little bit misaligned, I think just made that job like a little bit harder. I think that being said, it's like when I was working
58:26
Unidentified speaker (UNKNOWN): on the case.
58:26
Chloe Casale: at Palantir, I cared very, very deeply about solving my customers' problems. Um, and so I think like that piece, I think feels very similar. It's like going and understanding someone's workflow very, very deeply to the point where I feel like I can actually go and make it better, I think feels very similar. I think here all the conversations about trade-offs and like
58:46
Chloe Casale: prioritization and how exactly we're gonna going to accomplish this thing just feel a lot a lot easier, a lot more upfront.
58:53
Ron: Are there any things about how Palantir structured the forward deployed story that seem like different? Like there's obviously like you're pointing at ways in which it's just structurally different. The incentive structures are different. But like putting that part aside for a second, are there other ways in which it was handled in a materially different way than what we do? And is there anything for us to learn from that?
59:12
Chloe Casale: One of the pieces of forward Ford deployed and the way that it worked was as a Ford deployed engineer, you were expected to be able to give information back to product so that they could go and build a better product. And what that meant was like we had one core product that was like constantly improving in ways that kind of like helped every single customer.
59:29
Unidentified speaker (UNKNOWN): I'd
59:29
Chloe Casale: say that there are like a couple of places in Jane Street where like you can see maybe a similar model. I think like Webbs actually has a similar model where they have a mantle, but I think that for the most part, as a desk dev, I spend a lot of my time thinking about my desk. I spend some time thinking about what things maybe could be contributed back to like core products that
59:49
Chloe Casale: like can help all of Jane Street. But I'm not sure that necessarily feels like the crux of my job.
59:55
Unidentified speaker (UNKNOWN): And
59:56
Chloe Casale: I think that maybe that was the one thing that felt different structurally between the two in terms of like my core responsibility. responsibilities on the day-to-day. That I think that there are pros and cons to it as all
1:00:07
Unidentified speaker (UNKNOWN): things. So
1:00:08
Ron: just to go back, so you you mentioned webs and and sort of as a a goofy bit of internal terminology,
1:00:13
Unidentified speaker (UNKNOWN): there's
1:00:13
Ron: like a core webs team.
1:00:15
Unidentified speaker (UNKNOWN): Yes. And then
1:00:15
Ron: there's like more of a kind of client services subteam called Webb's Mantle is a joke about core, which I'm not sure I I necessarily approve of, but
1:00:24
Unidentified speaker (UNKNOWN): people
1:00:25
Ron: don't run naming decisions by me. Trevor
1:00:26
Unidentified speaker (UNKNOWN): Burrus,
1:00:26
Chloe Casale: Whimsy is
1:00:27
Ron: important.
1:00:30
Ron: I'm a firm opponent of naming Whimsy.
1:00:36
Ron: Yeah.
1:00:37
Unidentified speaker (UNKNOWN): I
1:00:37
Ron: feel like it's totally good to have a mildly whimsical, like identity-filled name for your system.
1:00:42
Unidentified speaker (UNKNOWN): Yeah.
1:00:43
Ron: But if you have like a whimsical name for every little subsystem in your system, you're just going too far, it's confusing people. You just need
1:00:49
Unidentified speaker (speaker_2): to like
1:00:50
Unidentified speaker (UNKNOWN): it's not going to be as funny.
1:00:51
Ron: in six months as it is right now. Like I feel like people maximize the wrong thing sometimes in their application of whimsy, but but I digress.
1:01:00
Unidentified speaker (UNKNOWN): But
1:01:01
Ron: anyway, yeah. So it's like a team that's like trying to reach out to understand like the people who are using webs and
1:01:06
Unidentified speaker (UNKNOWN): like
1:01:06
Ron: again, in part enabling stuff and in part figuring out how to kind of turn that around into, oh, actually, what do we need to do to improve webs for everyone? So I mean, the interesting thing about your team is it's, I mean, we don't have this kind of like striation of like there are these kinds of devs and that kind of devs.
1:01:25
Unidentified speaker (UNKNOWN): Yeah.
1:01:25
Ron: Right. And you guys both work with the people who you're building software for. Yeah. And you also build real deal systems for supporting those people. So when you say like giving information upstream to like the people who are providing the underlying product, it's like there's just it's just actually a much messier. There's like
1:01:45
Ron: a network of different people providing different pieces of software to whom you would supply information. So it's like, if nothing else, it's more complicated.
1:01:53
Chloe Casale: Yeah. Maybe like an easy example is like we had this thing that parsed parse chats and turns it into RFQs and it worked for sales for a really long time. It turns out not only sales gets chats and turns them into RFQs. Um like there are some places in which like our prop desks are doing the same thing.
1:02:10
Unidentified speaker (UNKNOWN): Right.
1:02:10
Ron: Because it turns out having these relationships with customers doesn't strictly fall on the official sales trading teams. Sometimes there are traders on the prop desk who do essentially the same kind of
1:02:19
Chloe Casale: exactly
1:02:20
Chloe Casale: and I think that at this point we figured that out and we built made our product more generalizable so that it can go and be used for other desks who are not just sales. And I think that's great. But I think it actually took us a really long time to kind of go and poke around the firm and realize, like, oh wait, I think I've solved a problem that a lot of people have, and I should share the system that I've built so they don't have to build
1:02:40
Chloe Casale: the same one.
1:02:41
Unidentified speaker (UNKNOWN): Right.
1:02:41
Ron: So in some sense, part of the time you're like a forward-deployed engineer and part of the time you're just like a regular software engineer building a generalizable service
1:02:49
Unidentified speaker (UNKNOWN): Yeah.
1:02:50
Ron: and figuring out like when you should be operating in one key and when you should be operating in the other. It's
1:02:54
Unidentified speaker (UNKNOWN): not
1:02:54
Ron: like always clear.
1:02:55
Chloe Casale: Yeah, it's definitely not.
1:02:56
Ron: Are there things that you feel like, oh, like there's useful ways in which Palantir approached it that like we could import? Like what are what are the good ideas there?
1:03:02
Chloe Casale: I think this is really I think it really was Palantir's secret sauce. And I really do think that this is like the thing of what you're
1:03:08
Unidentified speaker (UNKNOWN): doing. Yeah, it's like a thing they're famously good at. That's
1:03:12
Chloe Casale: I think that there are some devs who are great at this user thing and they're great at this product thing. I'm not one of those devs. I think that I'm great at this user thing. And I love being able to work on a team of people who are great at building this product thing. And I think that I think that at Jane Street, I think that you're expected to be able to wear a
1:03:32
Chloe Casale: lot of hats. And I think that this means that we have a lot of really, really I mean, exceptionally talented generalist devs. But I think that this also means that sometimes it's harder when like maybe you are much, much stronger in one place to figure out how I can relay race this idea to someone who's going to be able to take it and like work it even stronger than I would be able to do it. And I think that on
1:03:52
Chloe Casale: my team, like we've started restructuring ourselves a little bit into devs who are really good at this product thing and devs that are really good at this desk dev thing and starting to try to figure out like, how do we make it so everyone is enjoying their jobs and learning how to do the whole gamut, but also is very aligned to their strengths. And I think sometimes at Jane Street, I think sometimes I just wonder it's like if our desk devs had like an easier way to call
1:04:12
Chloe Casale: back and say, hey, here's these ideas that I think maybe apply to more than one person. And we had some sort of way to say, Oh, yeah, no, I'm seeing that too. It's like maybe we could turn and get some like product that was working for the whole firm, maybe a little bit faster. And I do say faster because I do think it does happen at James Street quite frequently, where it's like you'll see a bunch of the same idea pop up on a bunch of
1:04:32
Chloe Casale: desks at various different times or maybe the exact same time. And then all of a sudden a bunch of people turn around and they're like, oh wait, I built that. I built that, I built that. And then you get to have an opportunity of Yeah,
1:04:41
Ron: like the Spider-Man
1:04:42
Unidentified speaker (UNKNOWN): thing.
1:04:44
Chloe Casale: And then it's like you get the opportunity to turn that into a product. And I think that does happen quite frequently. But I think it just happens a little bit slower than I saw it happening at Palantine. Palantir because it does kind of require like a bunch of different people to all have built the same thing and then someone to notice, and then someone to organize, turning it into kind of like one product that the firm can use.
1:05:02
Ron: I
1:05:02
Unidentified speaker (UNKNOWN): mean,
1:05:02
Ron: web's a little bit, I mean, it's both an example and a counterexample of what you're saying. Web was a thing that was built
1:05:08
Unidentified speaker (UNKNOWN): by
1:05:09
Ron: the desk dev team on one particular desk. And then over time, they're like, actually, this is very general, and we should apply it to lots of places. It's not like a bunch of teams built the same thing. Yeah.
1:05:19
Unidentified speaker (UNKNOWN): Like
1:05:19
Ron: really one team built it and then figured out it was more general and figured out to generalize it. I think the thing that slowed that down was mostly just opportunity cost. It's like very valuable where it was. And at first it made sense to work on it mostly in that area. But I think we did eventually work around to it, maybe more slowly than we might have otherwise. So that's kind of in the direction of what you're saying.
1:05:37
Unidentified speaker (UNKNOWN): I feel
1:05:38
Ron: like the other thing I'm hearing from what you're saying is maybe that we should specialize more at a human level. Like one thing that's really different at Palantir is they really hire different kinds of people for different roles. They shouldn't
1:05:48
Unidentified speaker (UNKNOWN): go,
1:05:48
Ron: oh, this person would be a good FDE, and this person would be a good core engineer, and they like point them in different directions. Yeah. And we tend to hire people who we think are strong generalists, and
1:05:58
Unidentified speaker (UNKNOWN): then we
1:05:58
Ron: figure out where they go.
1:05:59
Chloe Casale: Yeah. It's a very different way of thinking about people. I think it was like a huge update for me when I joined Jane Street, and especially when I started hiring at Jane Street, to start thinking about people as like, okay, no, I'm hiring generalists, which was very different than the specialists I was hiring at Palantir.
1:06:14
Unidentified speaker (UNKNOWN): Aaron Ross Powell
1:06:14
Ron: So you said before that you're like kind of a recruiting snob.
1:06:19
Unidentified speaker (UNKNOWN): And
1:06:19
Ron: you liked a lot of what you saw of recruiting at Palantir and then you came here. Like what what has struck you now that like you are like pretty pretty heavily involved in London in the hiring process there and also like how that affects things globally. Like like what strikes you as like the good and the bad of Jane Street's approach to hiring?
1:06:39
Chloe Casale: Jane Street hires very, very differently than we hired at Palantir, which I think like at first I found really surprising. And I think that I But was I I don't know. I think at some point you start thinking that you just know things.
1:06:49
Unidentified speaker (speaker_2): And then and you then like come
1:06:51
Chloe Casale: to Jane Street and it's like I was like, wow, we hire entirely different Ely here, but the people here are very, very talented. So they must be doing something right. And it was like trying to figure out like, okay, you can use an entirely different process to still get very talented people. I think I don't know. I found that really interesting. I think I
1:07:06
Ron: feel like I've in general been surprised over time of like lots of things about how Jane James Street works. You think, well, we're just doing it in the natural, correct way. What
1:07:13
Unidentified speaker (speaker_2): got
1:07:13
Unidentified speaker (UNKNOWN): it? And then
1:07:14
Ron: you look out and like other companies do very different things and are often very successful. And you're like, maybe there isn't a natural correct way. Maybe there are maybe, maybe there are many different correct ways of doing things. And you know, you should just like keep your eyes open and learn from what other people are doing.
1:07:28
Chloe Casale: Yeah. And I think like one of those things I think with Jane Street is like, I'd say that like a Palantir At when we were hiring, we were trying really hard to get somewhat confident that someone was really good at at least one thing.
1:07:41
Unidentified speaker (UNKNOWN): I
1:07:41
Chloe Casale: think at Jane Street, we're trying to get very confident that someone is good at software development, I guess, like as kind of like a broad thing. So I think at Jane Street, I'd say that like we tend to ask very similar questions, like several times, in order to like give a candidate maybe the opportunity to like, oh, maybe they did it like a little poorly on like their first question
1:08:01
Chloe Casale: of the day, but their second and third question. of the day, they came back. And now I'm actually very confident that they're very good at this thing that I've kind of tested many, many times. Versus at Palantir, I think we gave very different questions that were very hyper focused on like one specific what we would call like facet of like a person. And what we were really looking for was like someone who was like had
1:08:21
Chloe Casale: a one thing that they were excellent at, and then were like good enough at the rest of things.
1:08:26
Unidentified speaker (UNKNOWN): And I
1:08:26
Chloe Casale: think that like
1:08:27
Ron: So multiple different dimensions in which you're looking and each read in those dimensions is lower confidence. Yeah. Because you have fewer, fewer, fewer instruments applied, fewer readings of that. Exactly.
1:08:39
Unidentified speaker (UNKNOWN): In
1:08:39
Chloe Casale: fact you only have one. So it's just you get one interview where you're trying to read that property. And maybe that some of the interviews have a little properties that overlap. But it was like very much like in a way, it was like you're putting one interviewer in the room is very It's different than Jane Street. You're putting one interviewer in the room in order to get one read. And then you're kind of like working off of that one read as like kind of saying, like, okay,
1:08:59
Chloe Casale: maybe I'm not like super confident in this thing because it was one interview, one interviewer. But like we had a lot of faith in our interviewers in those moments. Versus Jane Street, I think we think about it quite differently. It's like we want to give someone the opportunity to succeed over like a bunch of relatively similarly shaped interviews. where you have like two interviewers in each room. It's like we're getting very, very confident on like a
1:09:19
Chloe Casale: set of like very specific reads, I'd say here.
1:09:22
Ron: You are trying to, in those conversations, learn about multiple different things at the same time, right? Like when I think about things that I want to learn about in one of our interviews, are they like a good software engineer who knows how to write code? Right. Like are they you look at like a thing that'll often look for is when you look at somebody, like look at the actual like code that they write on the board and you
1:09:42
Ron: ask them to think about it, you see,
1:09:44
Unidentified speaker (UNKNOWN): you know
1:09:44
Ron: how you think about like someone you can tell if somebody writes a lot because they're like calluses on particular parts
1:09:49
Unidentified speaker (UNKNOWN): of
1:09:49
Ron: their hands.
1:09:49
Unidentified speaker (UNKNOWN): Yeah.
1:09:50
Ron: And I feel like they're similar, they're like marks on how someone talks and thinks about code that comes from have they like thought hard and worked hard about how do you build careful, reliable software, like the habits of mind that develop from that kind of work? That's one kind of thing I look for. I feel like I'm also looking for people's communication skills. Yeah. Like can they
1:10:09
Ron: explain something to you? Can they hear the thing that you're saying when you like point them in the direction of an idea? Like, do they stop and listen and understand? So you're like this kind of like back and forth how comfortable how comfortable it is to work with someone is something you're looking for. And this is a little bit of a kind of trading industry specific thing, but like
1:10:29
Ron: we care a lot about people's like understanding of their own levels of confidence.
1:10:34
Chloe Casale: Yes. Right?
1:10:35
Ron: This kind of second-order knowledge of not only like, do you know a lot of things, but are you like appropriately calibrated? Do you understand how well you know? Are you like wildly over or underconfident? Another kind of thing we look for. So I feel like we have a bunch of interviews where the questions poke at like somewhat different aspects, but we're actually looking for an ensemble of different
1:10:55
Ron: facets, facets, and at least at their best, interviewers are evaluating multiple of these at the same time.
1:11:01
Chloe Casale: For sure. I
1:11:01
Ron: do feel like early on we were much stupider about interviewing. Like I feel like early on we'd like to go into interviews and be like, I know some hard questions. I'm gonna ask some puzzles.
1:11:10
Unidentified speaker (speaker_2): And
1:11:10
Unidentified speaker (UNKNOWN): like
1:11:10
Ron: we'll hire the people who are really good at solving the puzzles. And like that wasn't completely.
1:11:15
Ron: useless, but it did succeed in hiring some people who really did not work out. Like it is not that is not like a good general recipe of like just hire the best puzzle solvers.
1:11:25
Chloe Casale: Yeah. It's hard to come up with a hiring process. Because I think, like, in a lot of ways, I think the thing that you're trying to hire at the end of the day is people who are going to be successful at Jane Street. So in order to build a hiring process that does that, you first have to figure out what makes someone successful at Jane Street's. Street.
1:11:38
Unidentified speaker (speaker_2): And
1:11:38
Unidentified speaker (UNKNOWN): I think
1:11:38
Chloe Casale: that. like some of these things are kind of obvious, I guess. Like maybe it's like you have to be a strong communicator sometimes.
1:11:46
Unidentified speaker (speaker_2): You
1:11:47
Unidentified speaker (UNKNOWN): have
1:11:47
Chloe Casale: to
1:11:48
Unidentified speaker (speaker_2): know how
1:11:48
Chloe Casale: to write code
1:11:49
Unidentified speaker (speaker_2): if you're gonna be a dev.
1:11:50
Chloe Casale: Um like there are
1:11:51
Ron: and
1:11:51
Unidentified speaker (UNKNOWN): to be clear, we were
1:11:52
Ron: basically just talking about the software engineering process. So maybe the writing code is like not a terrible thing to
1:11:57
Unidentified speaker (UNKNOWN): want at
1:11:58
Ron: least. Although, you know, there are different mixes of skills there that could be
1:12:01
Unidentified speaker (UNKNOWN): good. Yeah,
1:12:01
Chloe Casale: there are different mixes of skills. I think some people like have a very deep understanding of maybe their programming language. Like I've had conversations with people who are coding in a programming language I frankly have never seen before in an interview. And I have to ask a lot of like very basic questions. And sometimes you're talking to a dev who like has a very in depth knowledge of the language that it is that they're using. Sometimes you're talking to someone who's just like using the language as
1:12:21
Chloe Casale: a tool. I think we want both kinds of these people at Jane Street, but I think it is the kind of like different like levels of understanding of the tools. tool that it is that you're using. And so I think that I think that there is just this like thing about trying to turn what makes someone successful at a job into like a series of interviews that I think is just like an exceptionally difficult problem.
1:12:39
Unidentified speaker (UNKNOWN): So
1:12:39
Ron: I'm actually really curious about the facet thing. Like what is an example of a facet or two that they would use at it Palantir and have a whole interview focused on that facet?
1:12:49
Chloe Casale: So like for instance, all of our interns got a lot. learning facet. And so what we would do is we'd give them a programming language that they had never seen before because it was made up.
1:12:56
Ron: Okay.
1:12:57
Chloe Casale: We would teach it to them and then we would ask them to solve a bunch of questions in it. Um, and the whole goal was to just understand how did they learn? How did they approach learning a new language? How did they approach relying on their interviewer in that moment? And it's like you get other information, you know, about how they structure problem solving, how they structure code. because there are writing code. But the goal
1:13:17
Chloe Casale: of this interview was just how teachable is this person? How staerable is this person? How much do I know about their ability to basically learn something very quickly on the fly?
1:13:27
Ron: Would you tell them that was the goal of that interview? Or is it just a thing they're not supposed to know? And oh
1:13:31
Chloe Casale: no, we would tell them.
1:13:32
Ron: Okay.
1:13:33
Chloe Casale: Yeah. It's like, yeah, this is the learning interview, and this is what we're gonna going to be doing in this interview. And
1:13:37
Ron: then what's what's another facet?
1:13:38
Chloe Casale: We had a reengineering, I think is what we called it interview. So that was specifically focusing on debugging skills. So we gave them a whole program written in Java, I want to say. And then we gave them a series of like client bug reports. And then we would help them with the Java because a lot of people didn't know it. And then we would try to see like, oh, how do they navigate a code base that already exists? Can And they like understand the structure of the
1:13:58
Chloe Casale: code? How how do they tackle debugging? We had an interview just for decomping a problem. So we would just give them a very, very large problem. If you had to redesign the whole New York subway system from scratch, how would you do it? Kind of vibe. Just hearing how they would like kind of decomp it almost felt more consultant-y, but you would end up with some sort of like computer vague shape
1:14:18
Chloe Casale: of a program that maybe you would want to write at the end. And like all of these things, like you maybe you get a little bit of problem-solving overlap. You definitely get communication overlap. but they were all focused on something very, very specific we were trying to learn.
1:14:30
Ron: It's interesting because I feel like there are echoes in what you're saying of how JaneShuet's process works, in the sense that we actually do have this like a handful of questions that we use that are meant to be more design-focused. And questions that trying to get it, like people'll be able to read and debug code and various other pieces like that. Um but it sounds like the system the
1:14:50
Ron: the system. systematic approach is very different. It sounds propellant here. It's like when you're constructing the interview, you're like, okay, what facets are we going to cover? Yeah. And you go off and cover those facets. Whereas here, it's more like we have like a kind of standard interview process that we use across a wide variety of people that's again trying to get to this generalist set of skills. And then sometimes we'll be like, oh, for this particular kind of
1:15:10
Ron: person, we want to test some different things. And so we'll like pull in different questions and try and like. lean things a little bit. But it's it's more like you feel like there's a relatively like wide shape of a distribution of what you're covering. And your questions are like shift a little this way and shift a little that way. But it's almost like, you know, these are these like delta functions that you're trying to like, I'm gonna focus on this
1:15:30
Ron: facet, I'm gonna focus on that facet. I feel like this is like a closed-minded bias biased thing to say, but like I feel like our thing sounds much better.
1:15:37
Unidentified speaker (UNKNOWN): So
1:15:37
Ron: like what's can you make like what's the out what's what's good about the facet approach
1:15:40
Unidentified speaker (UNKNOWN): to
1:15:40
Ron: that?
1:15:41
Unidentified speaker (UNKNOWN): It
1:15:41
Chloe Casale: hires you very different kinds of people, right? So like you can imagine that like at Palantir, you interview someone and they crushed their decomp interview. So they're really good at decomping problems. And like maybe they just did like meh on the coding bits or meh on the learning. And like the rest of it was just kind of like, I don't know, fine. We felt like we had enough signal on this
1:16:01
Chloe Casale: one very specific facet about this one thing that this one person was excellent at. And we it the interview wasn't clouded by the things that they were maybe just less good at because the interview is very, very focused on this excellent thing that allows you to hire what I would call like a very spiky person. Um, so they're a person who's very, very good at like one particular thing. At Palantir,
1:16:21
Chloe Casale: we had a lot of spots for people who were very, very spiky. And so it's like, you could be a kind of person who wasn't a generalist and still do very, very well at Palantir. I think at James Street, we might miss this person. Like I think that like all of our interviews are covering skill sets that are like kind of like broad enough, but not quite as like hyper focused in on
1:16:41
Chloe Casale: like, can this person do this one thing really, really well? That I think that like we might miss, like, oh, actually, this person maybe would make a great desktop because they're really good at talking to users and decomping problems. We might not notice that because it might get maybe muddled by like the different signal that you're getting in the interview. Maybe someone like maybe even looks a little bit worse than they really are are, because like our interview is kind of like testing the
1:17:01
Chloe Casale: thought process
1:17:02
Unidentified speaker (UNKNOWN): that
1:17:02
Chloe Casale: isn't quite in the same framework of the way that this person thinks.
1:17:05
Unidentified speaker (UNKNOWN): I
1:17:05
Ron: feel like the two concerns I'd have about the kind of facet approach. One is, have you really done a good job of picking the facets? It seems I'll say it seems really hard of like picking facets that conditional on people doing well on that facet, like there really is a good spot for them.
1:17:19
Chloe Casale: Yeah. So
1:17:19
Ron: I think you have to get a lot of confidence in that like decomposition of people's skill set.
1:17:24
Chloe Casale: Yep.
1:17:24
Unidentified speaker (UNKNOWN): And
1:17:24
Ron: I guess the other thing is just the fact that you're taking a single measurement of it opens you up to a lot of noise. Oh
1:17:30
Chloe Casale: yeah.
1:17:30
Ron: Both false positives and false negatives, right? Certainly someone can like, you know, be very good at learning things, but do badly on the learning facet you throw at them.
1:17:37
Unidentified speaker (UNKNOWN): For
1:17:37
Chloe Casale: sure.
1:17:37
Unidentified speaker (UNKNOWN): And
1:17:37
Ron: that and then you just don't get to. measure that dimension.
1:17:40
Chloe Casale: Yeah. And that happens all the time. And so
1:17:42
Ron: I mean, this is like the big dark matter of hiring is that like processes tend to be conservative and that they tend to want to hire people who will be good fits. Although I mean, it's different by organizations. There are lots of organizations that hire more aggressively and then also fire more aggressively.
1:17:57
Chloe Casale: Yeah.
1:17:58
Ron: Which is like a kind of necessary like if you're if you need to keep a high level of people. um, then you really do have to fire a lot of people. And like we're not like huge fans of firing lots of people. So we like try and build a build part of the part of the structure of the recruiting process is it tries to build people who are like quite likely to be successful.
1:18:15
Unidentified speaker (UNKNOWN): Yeah.
1:18:15
Chloe Casale: But
1:18:15
Unidentified speaker (UNKNOWN): yeah,
1:18:16
Ron: I don't know. Do you feel like there were like how much was the noise thing a problem?
1:18:21
Unidentified speaker (UNKNOWN): I mean,
1:18:22
Ron: I feel like it's somewhat of a problem here. Like you certainly certainly like the correlation between how people did it on internet. an interview and how they did, say, in their internship, and then how they did in their internship and how they did
1:18:32
Unidentified speaker (UNKNOWN): as
1:18:32
Ron: full-timers is like not perfect. Even though the internship you have people there for like three months. Yeah. So you have a lot of time to learn. You'd think it would be perfect. But
1:18:39
Unidentified speaker (UNKNOWN): no, it's really not.
1:18:40
Chloe Casale: No. I think it was pretty noisy. I think that like we also did like
1:18:45
Unidentified speaker (UNKNOWN): I
1:18:45
Chloe Casale: think first of all, we spent a bunch of time trying to figure out like what our facets were. So it's like I think that it's like, you know, you'd spend years. because a recruiting cycle is long. So it's like when you want to change something in recruiting, you have to think about okay, let's say I change something in my recruiting process today.
1:19:00
Unidentified speaker (UNKNOWN): I'm
1:19:00
Chloe Casale: gonna make my first hire what in a month. That person's gonna start maybe in another three months. I'll find out how they're doing a year after
1:19:07
Unidentified speaker (UNKNOWN): that.
1:19:08
Unidentified speaker (speaker_2): And so
1:19:08
Chloe Casale: then in a year and a half, I'll know whether or not this decision I made was good.
1:19:12
Unidentified speaker (speaker_2): Like,
1:19:12
Chloe Casale: like that is not that is not the kind of timeline I like working with
1:19:16
Unidentified speaker (speaker_2): when I'm making decisions.
1:19:18
Chloe Casale: But I think that we spent a lot of time at Palantir kind of like going through a bunch of different facets and like testing them out and like figuring out which ones were like better than others. And I think that we also were happy with a little bit more noise. And so I think that like we were kind of conservative about the people that we would take and we wanted all of our interviews to have more false negatives and false positives. And so we thought really hard about maybe getting harder or s by some
1:19:37
Chloe Casale: proxy interview. But I think that like it was noisy for sure. And I think that we were optimizing in a way for people who were going to be noisier even when they showed up, right? Because it's
1:19:49
Unidentified speaker (speaker_2): like
1:19:49
Chloe Casale: you're optimizing for a person who maybe does have like a very specific job that it is that they can fill at the time. And so it just looked very different than I think James Street, where we try to optimize for something that looks very different.
1:19:59
Unidentified speaker (UNKNOWN): Aaron Powell
1:20:00
Ron: I mean it's interesting, I think we do also have roles that are more specialized in various ways. Like sometimes subject matter specialization.
1:20:07
Unidentified speaker (UNKNOWN): Yeah.
1:20:07
Ron: Right. Like, you know, I mentioned like we're hiring people for this formal methods team. And
1:20:11
Unidentified speaker (UNKNOWN): like
1:20:11
Ron: we have a somewhat custom interview process. It it hits a lot of the standard engineering questions because it's primarily an engineering role, but there's other things that we look at there. I think performance engineers are kind of an interesting corner because I think I feel like this is a place where I actually worry that we're like active. actively making mistakes. I mean, we do have a lot of performance-minded people and hire a lot of
1:20:30
Ron: performance-minded people. But I think our overall process, like one of the things we give people advice on doing, is we're like, you know, you're here for an interview, we want you to like write some code for this problem. Please focus on writing clear and simple and understandable code and like don't worry too much about the performance details of this code. And if you like are trying to hire someone who's going to be a performance
1:20:50
Ron: engineer,
1:20:51
Ron: it's like you can ask them to do that. And like also in their real job, it's actually important for a good performance engineer to like know when to bother to do the performance engineering work and when it kind of doesn't matter. Yeah. So it's not like that decision isn't important, but like, I don't know, they have intrusive thoughts about performance all the time. Oh
1:21:08
Unidentified speaker (UNKNOWN): yeah.
1:21:08
Ron: And it's like hard for them to not think about it. And it's also hard for them to believe that like whatever you say really what you care about is how good they are in thinking about performance. Because that is is
1:21:18
Unidentified speaker (UNKNOWN): the
1:21:18
Chloe Casale: job.
1:21:19
Ron: It's a large part of what the job is. And I worry that we sometimes get you know think that people aren't steerable or are not good fits or something because even though they are perfectly good fits, but like they're just like a different kind of person who thinks about the world differently. Yeah. And and our normal interview process might under underrate them. Aaron
1:21:36
Unidentified speaker (UNKNOWN): Powell
1:21:37
Chloe Casale: It's hard to figure out like what happens. habits are habits that like an experienced hire is like showing in an interview because this has just been the way they've been coding their entire careers and they'll be like relatively easy to change. What habits someone is showing in an interview that we actually want them to have when we get here, but we just don't want them to have in this moment in this interview. And what habits we're having in an interview where this is actually going to
1:21:57
Chloe Casale: be hard to change, which is gonna mean they're not gonna be successful at Jane Street. I think that like, especially when you're hiring experience, experienced can't. candidates, it's like you we hire experienced candidates because we want their experience, right? Yep. But I think that there's also a thing where
1:22:09
Unidentified speaker (UNKNOWN): it's
1:22:09
Chloe Casale: just harder to interview someone who has habits in the way that they do things and like trying to understand, like, oh, am I over-selecting or under-selecting or like what's going on here? I it's really hard. Yeah.
1:22:19
Ron: I mean, we hire a ton of experienced people. Like I think uh half-ish of the people
1:22:23
Unidentified speaker (UNKNOWN): we
1:22:23
Ron: hire are like out of the internship program and half of them are tend to be about experience very roughly. And it's great. I feel like we've learned an enormous amount from having experienced people start. But I I do think it is just easier to evaluate the kind of straight out of school people. Like you're just like a more uniform set of people with a more uniform set of experience. People in their careers
1:22:43
Ron: specialize.
1:22:44
Chloe Casale: Yeah. And,
1:22:45
Ron: you know, they bake in whatever the environment they that they are in. And it's just like more of a challenge to like evaluation. evaluate them properly.
1:22:52
Unidentified speaker (UNKNOWN): I
1:22:52
Chloe Casale: think it's even when you look at like hiring campus, I think it's one of the things I think about in London is like the kinds of students we see in London are different than the kinds of students we see in the US. Like I think that there are all sorts of places where like there are like some like assumptions baked into is some sort of standardization that maybe like isn't necessarily there. And I think like when you're thinking about hiring a lot, I think
1:23:12
Chloe Casale: spending a bunch of time kind of questioning those assumptions. I think, is like a huge part of your job. But also to your point, it's like kind of just like acknowledging that you can never be perfect at this thing. So trying to figure out like where is that optimization that I want to be having in this process.
1:23:26
Unidentified speaker (UNKNOWN): Yeah.
1:23:27
Ron: I think in many ways, JaneSheet is a very efficient hiring machine and is like very good at getting a bunch of very strong people to come in the door.
1:23:34
Ron: But like as the organization grows and you want to like keep on hiring more. people, like you actually always have to be doing better than you did before.
1:23:43
Unidentified speaker (UNKNOWN): And I think
1:23:44
Ron: in some ways, like happily, there's a lot of opportunity for doing better because I feel like what you're always doing in these processes is like you're like searching for your keys under the lamppost. You know, this is like an analogy of like, you know, there's a guy who like lost his keys and he's looking on you know he's looking on at night under this lamppost and someone says, oh, you need helping for your keys? He's
1:24:03
Unidentified speaker (UNKNOWN): like,
1:24:03
Ron: Yeah, yeah, I can't find them. He's like, Well, where'd you lose them? He's like, Um, yeah, like so you know, I'm not sure it's not like it's like, but why are you looking under the lamppost? He's like, Well, here there's light.
1:24:12
Chloe Casale: Right.
1:24:13
Unidentified speaker (UNKNOWN): But
1:24:13
Ron: you could look in the other places, but you wouldn't find it.
1:24:15
Chloe Casale: Right.
1:24:16
Ron: And so you're like optimizing the search for where you happen to be good at finding people.
1:24:20
Chloe Casale: Yeah.
1:24:20
Ron: And I think in recruiting, there's something similar where like you build a process that's good at getting great people. And I think if you get too arrogant about it, you think, oh, this is just the way to find good people.
1:24:30
Unidentified speaker (UNKNOWN): Right.
1:24:31
Chloe Casale: But
1:24:31
Ron: there are actually lots of different ways of finding good people. Yeah. And I think over time, if you're in an organization that's growing and wants to like increase the number of great people it can get, you have to like widen out how you search and be open to like trying different approaches and interviewing people in different ways and sourcing people in different ways and hiring people with different kinds of backgrounds for different kinds of roles.
1:24:51
Ron: if you want to keep on expanding what your capabilities are.
1:24:54
Chloe Casale: Yeah. You have to be willing to experiment, even though those experiments are really hard, because you're not going to get that signal on whether or not they worked
1:25:01
Unidentified speaker (UNKNOWN): for a really
1:25:01
Unidentified speaker (speaker_2): long time.
1:25:01
Unidentified speaker (UNKNOWN): For a really long time. But
1:25:03
Ron: you know, I've been here like twenty-four years.
1:25:05
Unidentified speaker (UNKNOWN): I feel
1:25:05
Unidentified speaker (speaker_2): like
1:25:06
Unidentified speaker (UNKNOWN): you know,
1:25:06
Ron: four years of an experiment doesn't seem so bad. Yeah.
1:25:10
Ron: I'm just very old.
1:25:12
Unidentified speaker (UNKNOWN): So
1:25:13
Ron: like taking all that into account, like what do you think we're probably doing wrong? Like what are things where it's like, ah, here's like here's the thing we could be doing better. Like what are what are what do you think are like our most likely mistakes?
1:25:24
Unidentified speaker (UNKNOWN): I
1:25:25
Chloe Casale: think that when I very first started helping out with like more time spending in Jane Street hiring a couple years ago, I think my biggest thing was it felt like we were interviewing experienced hires very similar to the way that we were interviewing campus hires. So like interns and trainers. straight to full-time people.
1:25:40
Unidentified speaker (UNKNOWN): And
1:25:40
Chloe Casale: I think that like kind of experienced hires, they just look different. And I think that in terms of that conversation of like, how do you start expanding the kinds of profiles and the kinds of people that you're capturing in your process? You have to
1:25:54
Unidentified speaker (UNKNOWN): be
1:25:54
Chloe Casale: willing to engage a little bit more critically with that question of like what makes someone successful at Chain Jane Street. And so I think that I worry that we are missing people that could be successful at Jane Street because our process isn't quite capturing like the nuance of what some differently shaped people could look like who would still be
1:26:14
Chloe Casale: successful here. I think that there's kind of a question of like what do we want Jane Street to be tomorrow? Like I think that it's like I hope that we're never the kind of firm where if you want to change teams, you have to go through an entire new interview process. But that's like
1:26:25
Unidentified speaker (UNKNOWN): a
1:26:25
Chloe Casale: thing that happens in some places. right? Where
1:26:27
Unidentified speaker (UNKNOWN): you
1:26:27
Chloe Casale: end up with these very, very specific hiring processes that are hiring exactly for what's successful on this team, and you can be like a very specific person. And so then it's really hard to change teams. But I do feel like there's some in-between for
1:26:39
Unidentified speaker (UNKNOWN): like
1:26:39
Ron: the generality is like a real value. Like one
1:26:41
Unidentified speaker (UNKNOWN): of the
1:26:41
Ron: things internally that we get to do is we get to pivot as an organization.
1:26:45
Unidentified speaker (UNKNOWN): Exactly.
1:26:46
Ron: Internally, like things are very like dynamic and entrepreneurial and like at times. we decide it's like, oh, we we, you know, this is more important than that, and we're gonna like move a bunch of people around and change what we're working on.
1:26:59
Ron: It's fairly common that we have like restructuring of like how teams are organized or who works where, or moving people around. Um, and at some point I realized I had to be very careful in And talking about this with people from some other firms where there's like there's a re-org in the and they
1:27:14
Chloe Casale: group.
1:27:14
Ron: grip their desk and be very nervous. I'm like,
1:27:17
Unidentified speaker (UNKNOWN): what
1:27:17
Ron: did I say? Like, we reorganize things all the time. It's
1:27:21
Unidentified speaker (UNKNOWN): like very
1:27:22
Ron: it's fairly chill. And like then I realized that like reorg, there's a lot of places where reorgs are like a bad thing. It's like the way in which you get the job that you liked taken away from you and you end up with a new job that you don't like anymore, and like all sorts of terrible things. But like I actually think being in an environment where like small scale reorganizations reorganization have is happening kind of continuously and all the time is like kind of great. My
1:27:42
Ron: job's been really different over many years. I worked in lots of different groups doing lots of different things. I think that's true for a bunch of people. And just like the shape of the trading desks and the tech groups and all of that has just like morphed a lot over time.
1:27:55
Chloe Casale: Yeah.
1:27:56
Ron: And and we've done like even at the pure level of like the technology we choose, we have optimized in the direction. of people's flexibility, right? Like the fact that we have tried to keep the number of programming languages down is like a real thing, right? The fact, you know, we we like I'm super glad we added Python, even though it's like, you
1:28:15
Unidentified speaker (UNKNOWN): know,
1:28:15
Unidentified speaker (speaker_2): I don't know, they're
1:28:16
Ron: not it's not my favorite language in all in all the ways. But like the ecosystem for ML and stuff is like incredibly valuable. It's been super useful to have. It's been a lot of work to make like Python well supported here. And like it does even even just having two languages can like reduce people's flexibility, some of like, oh, they spent all their time here
1:28:36
Ron: only working in one language and like, you know, maybe they don't know as well, or maybe there's like not psyched about working in the other language. And even there, like I we're s you know trying to move into a world where like people more broadly know both
1:28:47
Unidentified speaker (UNKNOWN): so
1:28:47
Ron: that they can like do whatever is important and not like pick the technology because they happen to know it, but pick the technology because it's the right one for solving the particular problem. And
1:28:57
Unidentified speaker (UNKNOWN): and
1:28:57
Ron: I think that flexibility has been incredibly valuable. And at a human level, having people who are themselves strong generalists has been really good.
1:29:07
Unidentified speaker (UNKNOWN): And
1:29:07
Ron: and also like we're big enough that some amount of specialization is like totally called for.
1:29:12
Chloe Casale: Yeah.
1:29:12
Unidentified speaker (UNKNOWN): Like
1:29:13
Ron: we have hardware engineers, right?
1:29:15
Chloe Casale: Yeah. Our
1:29:15
Ron: hardware engineers are actually like unusually good at software.
1:29:18
Chloe Casale: It's
1:29:19
Unidentified speaker (UNKNOWN): actually
1:29:19
Ron: a weird thing that like most hardware engineers in the world don't think like software engineers at all, even though you'd think it they're very similar things, but like they're really very different in a lot of ways. Yeah. Like, and and I think, you know, in many ways I feel like hardware engineering just seems harder and more intellectually
1:29:36
Unidentified speaker (UNKNOWN): challenging than
1:29:36
Ron: software engineering. Um but it is surprising. like
1:29:40
Unidentified speaker (UNKNOWN): a
1:29:40
Ron: lot of hardware engineers are like not comfortable writing recursive functions. Where I feel like writing a recursive function, that's like the easiest thing in the world. Yeah. Um but but that's super not how they feel about it. And they do stuff that seems super hard to me. Um and in the end, we end up hiring hardware engineers who are much more like software engineers. Um and that's
1:30:00
Ron: but but they're still like in practice. it's like a mostly a waste of their talent to have them work on pure software problems, like the right thing for them to be working on is hardware. So they're in that sense, they are less fungible even if they have like a lot of the skills that would let them do the other things.
1:30:14
Chloe Casale: Yeah. I think we you see the same thing in our front end engineering pipeline at Jane Street as well. It's like a lot of our front end engineers actually totally could go and be writing back Back-end end code if they wanted to, but it's like they do have this specialization
1:30:25
Unidentified speaker (UNKNOWN): that is
1:30:26
Chloe Casale: super useful for Jane Street and very, very valuable.
1:30:28
Ron: And it there's a weird thing in which the outside world has somehow decided that front-end coding is kind of second class. And like, you know, you get lots of kids out of school saying, So I
1:30:37
Unidentified speaker (UNKNOWN): don't
1:30:37
Ron: want to do front-end stuff. I wouldn't
1:30:39
Unidentified speaker (UNKNOWN): want
1:30:39
Ron: to be pigeonholed into doing front-end things. That's those are the bad things. And like it's so cool. Front-end
1:30:43
Unidentified speaker (UNKNOWN): stuff
1:30:44
Ron: is really interesting. It's actually really challenging to figure out how to design like performant and flexible UIs and building good abstractions in that context is actually really quite interesting. But like it's somehow weirdly low status for reasons I don't quite understand.
1:30:57
Chloe Casale: I don't get it. I also think I think, especially as a person who's very like user motivated, I think that I just sometimes you go and talk to someone and they're like, all I want is for this button to be blue. And I'm like, I can make your day.
1:31:09
Chloe Casale: Like that is like just so fun because it's like there's this thing about front-end engineering, which is like there is these iterations. with users that you can just have so quickly.
1:31:17
Unidentified speaker (UNKNOWN): But
1:31:18
Ron: like what in some ways, it's easy and true to say that like there are lots of great people that we miss. Like that's obviously true.
1:31:26
Unidentified speaker (UNKNOWN): But
1:31:26
Ron: what would you do about it? Like, you know, you have the keys to the kingdom, you can change our recruiting process however you want. Like, what would you do?
1:31:34
Chloe Casale: If I had the keys to the kingdom and I could change our the recruiting process however we wanted.
1:31:38
Ron: As opposed to the complicated process of consensus building that is actually there where no one has the keys to the kingdom.
1:31:45
Ron: We talk about it
1:31:45
Unidentified speaker (UNKNOWN): a lot and eventually come
1:31:46
Ron: to occasion. But
1:31:47
Unidentified speaker (UNKNOWN): no,
1:31:47
Ron: you have been given the keys. What do you do?
1:31:49
Chloe Casale: If I was given the keys, I would give I would come up with some interview questions that I thought were like testing things that people were going to be good at. And then I'd give every single person this would be so technically difficult. I already feel bad saying this.
1:32:04
Unidentified speaker (UNKNOWN): Like
1:32:05
Chloe Casale: every single person, especially experienced hires, are coming through one interview where I expect them to crush it. Because I think the thing that I want is to just give every single person the opportunity to show off what they're really, really good at when they come on site at Jane Street. And so I think that would mean if someone has like a bunch of like user experience in their maybe they're the CTO of something in their last job, it's like I
1:32:25
Chloe Casale: would want to be throwing them some. very high-level question that feels kind of CTO-y at Jane Street. Like there are some like problems that we tackle at Jane Street that I feel like I wish I could go and throw at people and be like, this sounds similar to your job. Here's a Jane Street problem. What would you do with it? And like give them the opportunity to shine that way. If they were very performance-minded, give them like a very specific question that like kind of like tackles
1:32:45
Chloe Casale: this thing. It's like I think that there's not enough opportunity for people to show off their specializations.
1:32:51
Unidentified speaker (UNKNOWN): And
1:32:51
Chloe Casale: I'm not saying we need to hire people based on their specializations, but I do want to see them when I'm making a decision as to whether or not someone should come to Jane Street. So if I could do anything, I think I would add that interview.
1:33:03
Ron: So like an attempt to be more adaptive to people's background, Yeah. essentially of like try and suss out like what's the most likely thing that they're like especially good at and see if you can have a question that like lets them ring that bell.
1:33:19
Chloe Casale: Yeah. I'd give one facet interview. I don't think I'd go all the way back to Palantir style. I think that was too much. And I think James Street, I really like what we do with our interviewing process here. But I think I would love to make sure that like every candidate has like at least one opportunity to show off something that I think they're they're gonna be good at based on their resume.
1:33:35
Unidentified speaker (UNKNOWN): And
1:33:35
Chloe Casale: I think that would be like the one change I would make to just try to capture those. people that maybe are showing up as just fine across all of our interviews, but could actually be excellent when they get here.
1:33:46
Ron: Awesome. All right. That sounds like a fun place to stop it.
1:33:48
Chloe Casale: Yeah. Thank
1:33:49
Ron: you so much.
1:33:50
Unidentified speaker (UNKNOWN): Yeah.
1:33:50
Chloe Casale: Thanks for having
1:33:51
Unidentified speaker (UNKNOWN): me.
1:33:52
Ron: You'll find a complete transcript of the episode along with show notes and links at signalsandhreads.com.
This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights