The Tool the Best Engineers Are Using Right Now

1. 0:00 – 0:05

   Intro

   1. AEAaron Epstein0:00

      [upbeat music]

2. 0:05 – 0:45

   Conductor explained: a Mac app to run multiple coding agents in parallel

   1. AEAaron Epstein0:05

      Today I'm excited to welcome Charlie and Jackson, co-founders of Conductor, fresh off of a $22 million Series A co-led by Spark and Matrix. Guys, thanks so much for joining.

   2. CHCharlie0:17

      Thanks for having us.

   3. JAJackson0:18

      Thank you, Aaron.

   4. AEAaron Epstein0:18

      Yeah. Tell us a little bit more. What is Conductor?

   5. CHCharlie0:21

      So Conductor is a Mac app that lets you run a bunch of coding agents all at the same time. So instead of being in your IDE editing files or w- in your terminal, uh, running Claude Code or Codex, Conductor consolidates all your coding agents and in one click you can create an isolated copy of your codebase, tell Claude or Codex to go work for you, and then you can review and merge its work.

3. 0:45 – 1:50

   Business traction and the new launch: Conductor Cloud

   1. AEAaron Epstein0:45

      Very cool. And you guys were the first to build kind of this new way of, of, uh, shipping code with multiple agents at the same time within one interface, and it seems like it's really working. Like, what's the state of the business right now? Tell, tell us more about h- um, how it's been growing.

   2. CHCharlie1:01

      Yeah. So we're, we're a small, scrappy team, but we are growing really fast. So we've grown about 10 times since January and, uh, uh, now we have everyone from, like, scrappy indie hackers to, uh, engineers at big public companies using us every day. Um, but there's a lot to build and, uh, uh, a lot to, uh, figure out, and so we're excited to grow the team.

   3. AEAaron Epstein1:24

      Yeah. Very cool.

   4. CHCharlie1:25

      Yeah.

   5. AEAaron Epstein1:25

      And you have some new announcements of things that you're launching today. Tell us more about that.

   6. CHCharlie1:29

      So we're very excited to launch Conductor Cloud today. Uh, up until today, Conductor all runs locally on your Mac. So if you shut your laptop, Claude Code or Codex or whatever agent you're using will stop working. Um, but as of today, you can create a workspace that is backed by a cloud environment. So you can shut your laptop and the agent will keep working for you.

4. 1:50 – 2:37

   The scaling problem: managing more than 3–5 agents is an interface challenge

   1. AEAaron Epstein1:50

      Uh, it seems like a lot of people that are using Conductor are maxing out how many, uh, different instances, how many different worktrees they can be running at a, at a single time. Is this the big unlock that's gonna let people get past the bottlenecks that they've been hitting so far?

   2. CHCharlie2:06

      For me, I think, like, I can only really manage in my head, like, three to five agents at once, and I think this is part, like, partly an interface problem, and s- one thing we're really excited to solve next. Um, I think we, we, we've proven that you can run more than one coding agent at a time and it will still be productive, but I think to get to, like, the next level, where you run, like, more than three to five, like, we'll need ... It's an interface challenge, and, like, I think we'll need to, like, go up a, a higher level of abstraction to, to make that work.

   3. AEAaron Epstein2:36

      Very cool.

   4. JAJackson2:37

      Yeah.

   5. AEAaron Epstein2:37

      Um-

5. 2:37 – 3:45

   Early experimentation: from manual repo clones to worktrees to Conductor primitives

   1. JAJackson2:37

      When, when we started Conductor, like, we were try, trying to do this manually ourselves and just running into a lot of friction that made it hard to get past one or two agents, and that's, like, why we built the app in the first place.

   2. CHCharlie2:52

      Yeah.

   3. JAJackson2:52

      I think even before that we were trying to build stuff like Conductor and, um, like, the, our very first attempts, it was just, like, too, much too early to be able to move away from your IDE.

   4. CHCharlie3:04

      Yeah.

   5. JAJackson3:05

      So ...

   6. CHCharlie3:06

      Yeah. Yeah, the agents had to, like, re- reach a, like, sweet spot. There was kind of a tipping point where you didn't have to pay attention to everything they were doing, um, but you had to pay attention to enough of it where that made orchestration and, like, running multiple at once possible. Um, but yeah, in our, in our, like, early experiments with this, we were, like, we had, like, five clones of our repo and, like, we were running a Claude in each one. And then, then we discovered worktrees and we were like, "Oh, we should probably be using worktrees." And, like, step by step we, like, started building the primitives, like, for Conductor, like, to solve a problem for ourselves.

   7. AEAaron Epstein3:41

      Yeah. That's very cool. And, um, m- maybe let's go back to the beginning and, and kind of-

6. 3:45 – 6:59

   Founders’ origin story: meeting in college and deciding to build together

   1. CHCharlie3:45

      Yeah, yeah

   2. AEAaron Epstein3:45

      ... tell the story of how you guys got there. H- how did you guys first meet?

   3. JAJackson3:48

      Yeah. We met in college. I was a freshman and Charlie was, uh, fifth year.

   4. CHCharlie3:54

      Yep, yep.

   5. JAJackson3:55

      He was-

   6. CHCharlie3:55

      Which at the time was, like, is, like, the biggest age gap possible.

   7. JAJackson3:58

      [laughs]

   8. CHCharlie3:59

      Like, if you had told me then that, like, we'd be working together, I would be like, "Jackson's a child." Like, h- like, w- that would be just crazy to me. Yeah.

   9. JAJackson4:08

      Yeah. Charlie was, uh, he, he taught me how to lift. Um-

   10. CHCharlie4:12

       Yeah

   11. JAJackson4:12

       ... he was a great teacher, great coach.

   12. CHCharlie4:14

       Thank you. F- finally, finally you have, have admitted it. [laughs]

   13. JAJackson4:18

       All right.

   14. CHCharlie4:18

       Yeah.

   15. JAJackson4:18

       I'll, I'll never say it again.

   16. CHCharlie4:20

       [laughs]

   17. JAJackson4:20

       Um, yeah, you, you were trying to convince me to take a CS class with you-

   18. CHCharlie4:24

       Mm-hmm

   19. JAJackson4:25

       ... and I, being a stubborn freshman, I refused. Um-

   20. CHCharlie4:28

       Yeah

   21. JAJackson4:28

       ... but yeah, we, we, we connected after college basically, um, and were working on a bunch of projec- crazy projects together. Um, Charlie moved out to San Francisco and, um, I was working at Netflix on the machine learning infrastructure team. Charlie was at Replicate doing growth and, like, engineering. And we were just working on a bunch of projects together, and, um, uh, like, wanted to, I guess, try our hand at a startup, and we didn't really know what we were gonna do at all, but we, we got it. We had an interview with, with you, Aaron, and we, uh, decided to do it. Yeah.

   22. AEAaron Epstein5:09

       Yeah. Talk about, um, deciding to, to work together and start something. So you were at Netflix and you were at Replicate. Did you know that you wanted to do a startup? Did you know that you wanted to do something with each other? W- what was that first kind of conversation of, you know, we should leave these jobs a- and actually start a startup together?

   23. JAJackson5:26

       I think we were having a lot of fun building stuff together. Um, I didn't even have in my head, like, an idea of what a startup was. I mean, I guess I had some idea, but it was not at all connected to what startups really are.

   24. CHCharlie5:38

       Mm-hmm.

   25. JAJackson5:38

       So I don't, I, I didn't have, like, a, a dream to do a startup, but I, I really liked working with Charlie, and he, um, knew a lot more about startups than me, and he, this was something he was interested in, and I was, um-Uh, looking for a change. I was, like, working from home in my bedroom and going crazy, and, um, yeah, I, I wanted something to sink all of my time into.

   26. CHCharlie6:00

       I remember, uh, like reading Hacker News at my first job out of college and reading, uh, You Weren't Meant to Have a Boss.

   27. JAJackson6:07

       Mm.

   28. CHCharlie6:07

       And b- and being like, "Yeah, I wasn't meant to have a boss."

   29. JAJackson6:09

       [laughs]

   30. CHCharlie6:09

       And so I think, like, the idea of a startup, like working on a startup, has been, like, in my brain for a while. It felt very natural to work with Jackson because we, like he said, like we were having so much fun building projects together, and the idea of just, like, doing that all the time felt amazing. Um, we had played ultimate Frisbee together in college. Believe it or not, it was a very, very serious team, and we put a ton of hours i- into it and, um, we were a really good team. A big part of me, and I think it's like a big part of our company now, is trying to replicate that feeling of being on, like, an elite sports team together and, like, putting your all into it and working against big challengers. Like, in college it was UNC.
7. 6:59 – 8:40

   First startup idea: an AI agent that books reservations (and why it didn’t last)

   1. AEAaron Epstein6:59

      When you guys decided that you wanted to start a company together, did, did you have an idea? Did you know what you wanted to work on?

   2. JAJackson7:04

      When we applied to YC, our idea was AI that books reservations for you.

   3. AEAaron Epstein7:09

      Yeah.

   4. JAJackson7:10

      And that did not last long. [laughs]

   5. CHCharlie7:12

      [laughs]

   6. AEAaron Epstein7:12

      Well, I remember that. Your application talked about you could reserve tennis courts-

   7. CHCharlie7:16

      Yeah

   8. AEAaron Epstein7:16

      ... and yeah.

   9. CHCharlie7:16

      Yeah, yeah. And I think in hindsight there were some cool ideas. Like, it was like us- We were playing around s- using Sonnet 3.5 to, like, control your browser for you.

   10. AEAaron Epstein7:23

       Mm-hmm.

   11. CHCharlie7:24

       But it was very much a solution in search of a problem. We were like, "Okay, it's cool that Claude can, like, click around in our browser. Like, what do we, what do we do with it?"

   12. AEAaron Epstein7:32

       So I remember interviewing you guys with that idea.

   13. CHCharlie7:35

       Yeah.

   14. AEAaron Epstein7:36

       Um, how married were you to that idea at the time?

   15. CHCharlie7:39

       I remember, like, trying to decide, uh, how much conviction do we act like we have in this i- idea.

   16. AEAaron Epstein7:44

       [laughs]

   17. CHCharlie7:44

       And we, uh, we, I did, I think we did a pretty good job of, like, thinking it through a lot.

   18. AEAaron Epstein7:48

       Yeah.

   19. CHCharlie7:48

       And, like, thinking through the, like, the holes and, like, yeah, like, the, how, how it could get really big. It was a bigger challenge than something we had worked on, but it wasn't really like... I didn't feel, like, in my bones, like this was the thing to work on.

   20. AEAaron Epstein8:01

       Yeah.

   21. JAJackson8:01

       Yeah.

   22. CHCharlie8:02

       Yeah.

   23. JAJackson8:02

       And as we learned many times in YC, there's not that much you can learn by just thinking through an idea.

   24. CHCharlie8:08

       Yeah.

   25. JAJackson8:08

       Once it, you go actually try and do it, you find out all your thoughts were wrong, or at least we did.

   26. CHCharlie8:13

       Yeah, yeah.

   27. JAJackson8:14

       So.

   28. CHCharlie8:14

       Yeah. And even if someone-

   29. JAJackson8:15

       That didn't count for that much. [laughs]

   30. CHCharlie8:16

       Yeah. And even if someone tells you, like, "I really want this thing," you can put it in front of them the next day and they're like, "Oh, like, my friend is gonna love this." Like, like, it doesn't necessarily translate to, like, a- them actually wanting something.
8. 8:40 – 11:30

   YC batch reality: rapid prototyping, constant pivots, and speed as a superpower

   1. AEAaron Epstein8:40

      And so maybe talk a, a little bit about, um, what the, the YC batch experience was like for you and, and what your main focus was during, during that three months.

   2. CHCharlie8:48

      Yeah.

   3. JAJackson8:50

      I think probably the first month and a half we were casting around for an idea, and, um, we, we tried all sorts of stuff. We were asking our, our friends and, like, building random prototypes and, um, showing them to people-

   4. CHCharlie9:05

      Yeah

   5. JAJackson9:05

      ... who invariably hated them. And it's like, yeah, building stuff we knew nothing about.

   6. CHCharlie9:10

      Felt like every office hours it would be like, "Okay, we're trying this new thing." [laughs]

   7. AEAaron Epstein9:14

      Yeah. That's what I remember, too. I feel like there'd be times where we'd meet, you know, every three days sometimes.

   8. CHCharlie9:19

      Yeah. [laughs]

   9. AEAaron Epstein9:19

      And you'd be like, "All right, when we come back in three days-

   10. CHCharlie9:21

       Yeah, yeah

   11. AEAaron Epstein9:22

       ... we're gonna be here." And then you'd come back in, you'd be like, "We, we have a new idea."

   12. CHCharlie9:25

       Yeah.

   13. JAJackson9:25

       Yeah.

   14. CHCharlie9:26

       Yeah.

   15. JAJackson9:26

       Well, did you think when you interviewed us it would take us so long to, to come up with something?

   16. AEAaron Epstein9:31

       Well, it, it wasn't the coming up with something that was, was the thing for you guys, it was the finding the thing-

   17. CHCharlie9:37

       Yeah

   18. AEAaron Epstein9:37

       ... that was actually gonna stick. You guys came up with a lot of idea- A lot of them were, were really interesting, too.

   19. CHCharlie9:41

       Yeah.

   20. AEAaron Epstein9:42

       Um, and I think the other thing that really stood out to me was you guys were just, like, the fastest builders. You built stuff so fast that you shipped things every three days, sometimes totally new ideas, way faster than a lot of other companies would even ship an MVP for-

   21. CHCharlie9:58

       Yeah

   22. AEAaron Epstein9:58

       ... you know, their whole thing during the batch.

   23. CHCharlie9:59

       Yeah.

   24. AEAaron Epstein9:59

       And so that was really impressive to see. So, um, any-

   25. JAJackson10:03

       Um, credit for that goes to Charlie. I think he has just a, an incredible drive to, like, get stuff out the door as fast-

   26. AEAaron Epstein10:10

       Mm-hmm

   27. JAJackson10:10

       ... as possible, and I've, I've l- hopefully some of that has rubbed off on me.

   28. CHCharlie10:15

       Yeah. I th- I, I think we have a, a nice dynamic where there's, like, two archetypes of engineers. I, I think of myself as a hacker, where I can make something really... Like, I, I like moving really fast, and I can make something that's sort of, like, duct-taped together and, like, leaves a trail of dust behind it.

   29. JAJackson10:31

       [laughs]

   30. CHCharlie10:31

       Uh, and then Jackson I've, is, I, I, I think of as, like, a gardener, where he will kind of, like, pick up the broken pieces behind me and then, like, build it into something, um, like some- something robust.
9. 11:30 – 14:22

   Key turning point: ‘Build a dev tool you want’ and the path through Chorus

   1. CHCharlie11:30

      I, I very distinctly remember coming into an office hours, like, halfway through the batch, and we were, we were pretty down. Like, I, I, I was really into B2C ideas and Jackson liked B2B ideas, and we had tried, like, 10 different things at that point. And I remember you saying... I remember two things. One was, um, you said, um, "I think dev tools would be a good fit for you guys."

   2. JAJackson11:54

      [laughs]

   3. CHCharlie11:54

      And then kinda like a light bulb went off in our heads when you said that. And the s- the second thing you said that we still, like, have to this day, actually, is you said, uh-P- uh, put up a poster on your wall of your guys' faces, and then just say across it at the top, "Make something these guys want."

   4. AEAaron Epstein12:09

      [laughs]

   5. CHCharlie12:09

      And just build something that you guys want.

   6. AEAaron Epstein12:11

      Yeah.

   7. CHCharlie12:11

      Um, and so we literally did that. Um, and then I think from there it actually, like... That narrowed the scope a lot. We were like-

   8. AEAaron Epstein12:17

      Mm-hmm

   9. CHCharlie12:17

      ... okay, dev tools. Um, and we at the time w- and it's funny, this is still a common thread, we felt like the AI coding tools weren't going far enough. Um, and so that summer, Sonnet 3.5 had come out recently. Um, we felt like there was, like, a big overhang in, like, what was possible and, like, what people realized was possible.

   10. AEAaron Epstein12:38

       Mm-hmm.

   11. CHCharlie12:38

       Um, uh, and so... And we also became obsessed with this, uh, coding tool called Aider, which was, uh, like, sort of, like, sort of like the grandfather to Claude Code. And we actually tried to build a version of Conductor, like, that summer, where, um, w- we were like, instead of editing the files by hand in an IDE or even using, like, a tab model, like, you... Instead the human should be telling the AIs, the AI what to do, and it would go off and do a bunch of work and surface it to you. Um, and so we tried to make a user interface that was, like, the thing after the IDE.

   12. AEAaron Epstein13:10

       Mm-hmm.

   13. CHCharlie13:11

       Um, and the problem is it didn't quite work. Uh, the models weren't really good enough. Like, Sonnet 3.5, he did a bit too much hand-holding, and you would need to open your IDE too much anyway. Um, tool calling was, like, barely a thing back then.

   14. AEAaron Epstein13:26

       Mm-hmm.

   15. CHCharlie13:26

       We lost a bit of conviction in, in ourselves and, like, in that idea because, like, it just- the models weren't quite ready, and, like, we weren't quite, like, ready to, to build that. We were like, let's take a... Like, we have a bunch of ideas around, like, how the UX of this thing should work. We were like, let's, let's just apply that to chat, um, because we know that we can make a better chat app than anything that exists. And we had an office hours with you where you were like, "You can't, like, you can't pivot until you get 10 people to try something."

   16. AEAaron Epstein13:56

       [laughs]

   17. CHCharlie13:56

       And so we were like, okay, we're just gonna put something out. Uh, and we're gonna put out a basic version of a chat app, and it's gonna let you talk with multiple models at once, 'cause that's, uh, again, that's like solving, like, a problem for ourselves that we wanted. So we released Chorus in January, so, like, a few months after the batch actually. Um, and then it was in the process of building Chorus and building the dev tools that would let us build Chorus faster that we stumbled up- upon Conductor.

10. 14:22 – 15:24

    From first line of code to users: Conductor’s prototype-to-launch sprint

    1. AEAaron Epstein14:22

       Mm-hmm. How long did it take you? It sounded like you built the prototype really fast. Jackson saw it, and Jackson was like, "This is not how I would want to use it."

    2. CHCharlie14:29

       Yeah, yeah.

    3. AEAaron Epstein14:29

       So you were kind of building for yourselves first.

    4. CHCharlie14:31

       Yeah, yeah.

    5. AEAaron Epstein14:32

       How long until you got to the thing that you actually started getting other users to, to start testing out and giving you feedback on?

    6. JAJackson14:38

       I think it was probably, like, three weeks from when we s- wrote the first line of code, and there were a couple weeks before that when we were using Claude Code a lot and, like, developing some manual processes around it and thinking, like, "We need a tool for this"-

    7. CHCharlie14:54

       Yeah

    8. JAJackson14:54

       ... "and, like, there's a huge opportunity to build something around this."

    9. CHCharlie14:56

       Yeah. We were like, "Someone else is gonna build this. Like, like, we, we need this tool."

    10. JAJackson15:01

        Yeah.

    11. CHCharlie15:01

        And then we were like, "Wait," like, "we can build it ourselves."

    12. JAJackson15:03

        Yeah.

    13. CHCharlie15:03

        And yeah, I think it was, like, two to three weeks from, like, first line of code to releasing it.

    14. JAJackson15:08

        The first week was Charlie prototyped it all himself, and then that worked shockingly well, and we're, we decided to spend... We- I think we thought we could ship it in another week, and it probably took us-

    15. CHCharlie15:21

        Yeah

    16. JAJackson15:21

        ... two and a half more weeks-

    17. CHCharlie15:23

        Yeah, yeah. [laughs]

    18. JAJackson15:23

        ... um, to polish it up.

11. 15:24 – 17:21

    The ‘magic moment’: parallel agents, unread dots, and building Conductor with Conductor

    1. AEAaron Epstein15:24

       And what convinced you? I mean, you had, you'd been through that process a bunch of times with, with other ideas and other apps. W- what was, like, the first thing that you saw that actually convinced you that, like, I think we're onto something here, and I think this is gonna stick?

    2. CHCharlie15:36

       I think for me, it was the first time giving the agent a task, then pressing Command and, and giving another agent a task, and then, uh, seeing a little, like, unread dot come up in the first agent.

    3. AEAaron Epstein15:52

       Mm-hmm.

    4. CHCharlie15:52

       And then I, like, clicked back into it, and it, like, had finished.

    5. AEAaron Epstein15:55

       Yeah.

    6. CHCharlie15:55

       And I was like, like, the idea of running multiple agents, like, even running, like, one Claude Code was pretty new at the time. And so the fact that we could run multiple and it would actually be productive was pretty magical.

    7. AEAaron Epstein16:08

       Mm-hmm.

    8. CHCharlie16:08

       Um, I mean, another litmus test for me is, uh, 'cause I get really excited by lots of, like, shiny new things. Another litmus test was, for me, was does Jackson fi- find it useful?

    9. AEAaron Epstein16:20

       Mm-hmm.

    10. CHCharlie16:20

        And it... 'Cause most things-

    11. AEAaron Epstein16:21

        [laughs]

    12. CHCharlie16:21

        ... I show him, he's like, "I don't... What am I supposed to u-

    13. AEAaron Epstein16:24

        [laughs]

    14. CHCharlie16:24

        ... like, I don't... There's no... I don't know what to do with this." But this was one of the few times where he was like, "Wait," like, "uh, this was actually kind of helpful for me."

    15. AEAaron Epstein16:31

        Yeah.

    16. CHCharlie16:32

        Um, so that was, like, another, like, another bar to cross.

    17. AEAaron Epstein16:35

        Yeah.

    18. CHCharlie16:36

        Yeah.

    19. JAJackson16:36

        I think there was also a phase when we, in the early weeks developing Conductor, when we were u- it was... There are certain problems that are, like, really well shaped for Conductor, where you have a bunch of, like, s- mostly independent features or bug fixes you wanna do. And I think there was a, a one week maybe when we were, like, really in that phase, and we were just ripping through-

    20. CHCharlie16:59

        Yeah

    21. JAJackson16:59

        ... like, so much code- [laughs]

    22. CHCharlie17:00

        Yes

    23. JAJackson17:00

        ... with Conductor and-

    24. CHCharlie17:01

        That's true, yeah

    25. JAJackson17:01

        ... it was exhilarating-

    26. CHCharlie17:03

        Yeah

    27. JAJackson17:03

        ... I think, uh.

    28. CHCharlie17:04

        I think being able to build Conductor with Conductor was another, like, magic moment-

    29. AEAaron Epstein17:08

        Yeah

    30. CHCharlie17:08

        ... for us because I c- I was just, like, not leaving the app, and I was just like... I felt like, yeah, it was like conducting on, like, l- on a ton of caffeine and it, and it was actually working.
12. 17:21 – 20:21

    Going viral: authentic demos, reps, and learning what you’d click on

    1. AEAaron Epstein17:21

       I mean, you guys were the first to pioneer a lot of this, running, um, multiple agents at the same time. Um, and so that obviously, you know, resonated with you immediately. It resonated with a lot of other people. But you also have this other special ability, this knack to, of, uh, you know, putting out videos that go viral.

    2. CHCharlie17:40

       Mm-hmm.

    3. AEAaron Epstein17:40

       Um, talk a little bit about where that came from and, and maybe any, like, tips or tricks or secrets that you can share with anybody else that's watching this on, on how you think about that.

    4. CHCharlie17:49

       Yeah. One of my claims to fame is, uh, a demo I made where the day GPT-4 Vision came out, um, I wrote a hacky Python script that, uh, would record, uh, my webcam and then watch me in real time, and then use a voice clone of David Attenborough to, like, narrate what I was doing as if I was in Planet Earth. And, um, uh, I, I just, I tweeted that out on a whim, and it-Went totally viral, and even, uh, Sir David Attenborough saw it. Um-

    5. AEAaron Epstein18:20

       [laughs]

    6. CHCharlie18:20

       Unfortunately, unfortunately he didn't love it, which it was sad for me because I- I'm a huge fan of his.

    7. AEAaron Epstein18:24

       Yeah.

    8. CHCharlie18:25

       Um, but, um, I think part of, uh, part of how I learned to, um, to, to get attention on Twitter and to like shape posts in a way that could go viral was my role at Replicate was basically this. Um, I, I joined as an engineer, but like our main problem was growth for a while. And so how do we get people to use our API? Well, we have to like let the ... First they have to know about it. And so like building demo apps and like make like ... creating like, uh, videos that like showed you what was possible-

    9. AEAaron Epstein18:55

       Yeah

    10. CHCharlie18:55

        ... was like a big part of my job. I think one thing that- that's really important to me is like being authentic-

    11. AEAaron Epstein19:01

        Mm-hmm

    12. CHCharlie19:01

        ... and like being direct and not using like corporate speak, or not trying to like act like you're bigger than you are or ... I don't know, like saying it how it is, like you're talking to a friend. And I think that's one thing that's worked well for me, is just like, like the David Attenborough video was just like a raw screen recording. I like, like I just like showed my screen and di- didn't make many cuts and like it's just some Python on a screen and like showing you how cool that could be. And I, and then keeping the language really direct and straightforward I think is, is important. But then I also think like it's im- like you have to like, you have to just have a ton of reps and like put out a bunch of stuff and see what resonates with people and like start to learn like the shape of a post. Um, uh, I don't know. I don't know if I'm doing a good job explaining this. It's-

    13. JAJackson19:50

        Some of, some of the advice that I thought-

    14. CHCharlie19:53

        Yeah

    15. JAJackson19:53

        ... that you had that I thought was really interesting was like observe what you like.

    16. CHCharlie19:59

        Mm-hmm.

    17. JAJackson20:00

        Like w- what you heart-

    18. CHCharlie20:01

        Oh, yeah

    19. JAJackson20:01

        ... and what you like seeing-

    20. CHCharlie20:03

        Yeah

    21. JAJackson20:03

        ... when you scroll and like observe the, the emotions and the reasons for why that is and like try and replicate that.

    22. CHCharlie20:08

        Yeah. Yeah. That's, yeah, that's one n- nice way of summing it up is like, yeah, make something that you want to click-

    23. JAJackson20:13

        Mm-hmm

    24. CHCharlie20:13

        ... like on. Um, similar to the YC advice, it's like make something that you want-

    25. JAJackson20:18

        Yeah

    26. CHCharlie20:18

        ... to consume and, and like, yeah.

13. 20:21 – 23:00

    What top engineers do with AI tools: skills files, simple setups, and ‘slop-free zones’

    1. AEAaron Epstein20:21

       Yeah. That's good advice. You guys, I know you, you do a lot of, uh, bike tours around the city-

    2. CHCharlie20:26

       Mm-hmm

    3. AEAaron Epstein20:26

       ... to meet with your users.

    4. CHCharlie20:27

       Yeah, yeah.

    5. AEAaron Epstein20:27

       And sit down and watch them use Conductor.

    6. CHCharlie20:30

       Yeah.

    7. AEAaron Epstein20:31

       Um, and it's, it's pretty cool 'cause you guys have a front row seat to some of the best engineers in the world that are literally building on the frontier doing things that have never been possible before. And I'm curious what you've learned in the course of watching some of those top engineers work.

    8. CHCharlie20:45

       Yeah.

    9. AEAaron Epstein20:45

       Like what sets them apart?

    10. CHCharlie20:47

        Yeah. So I, I think a couple of things. I think o- one thing that I've noticed a lot is the best engineers are putting a lot of thought into their, their skills files.

    11. AEAaron Epstein20:58

        Mm-hmm.

    12. CHCharlie20:58

        Um, and that's something we try and do as well, is like take knowledge that like, uh, like take React best practices that we've learned, like from either talking to people, like experienced React engineers or like s- things that are specific to our codebase, and putting those into markdown files, and then the agents can continually read from. Um, I think one thing also that's maybe somewhat surprising is how simple and how vanilla most of the best, uh, the best, uh, engineers', uh, uh, setups are. Like I think you can go, I think you can easily Midwest meme here-

    13. AEAaron Epstein21:33

        Mm

    14. CHCharlie21:33

        ... and like put a ton of thought into your configurations and your ... Sort of like reminds me of like people who p- like have amazing Vim setups and like Vim key bindings. They can do stuff like, I don't know, they can like ... I don't know. They can do all kinds of crazy stuff on Vim, but like not necessarily get work done.

    15. AEAaron Epstein21:50

        Mm-hmm.

    16. CHCharlie21:51

        Um, so I think people have surprisingly simple, uh, setups. Um, but I do think th- putting a lot of thought into your, uh, skills has been like a, a recurring thing. And I also think, um, uh, something we are working on, something I've seen, is having a f- what we call like slop-free zones.

    17. AEAaron Epstein22:08

        Mm-hmm.

    18. CHCharlie22:08

        Where it's like you can let the AI run wild in certain parts of your codebase, and then other parts of your codeba- codebase are like much more thoughtfully considered and like the human is the architect, not the AI. Um, but I, I think we're still figuring it out. Like-

    19. AEAaron Epstein22:24

        Yeah.

    20. CHCharlie22:24

        Yeah.

    21. AEAaron Epstein22:25

        Yeah. I mean, everything has been changing-

    22. CHCharlie22:26

        Yeah

    23. AEAaron Epstein22:27

        ... so quickly over the last few months even. Um, and, and we're even finding, you know, in the current YC batch, the advice that we give is different in a lot of ways than it was even just a few months ago.

    24. CHCharlie22:36

        Mm-hmm.

    25. AEAaron Epstein22:37

        Which makes it a really exciting time.

    26. CHCharlie22:38

        Like in what way?

    27. AEAaron Epstein22:39

        Um, I, I mean, you can build so much more so much faster, you know?

    28. CHCharlie22:43

        Yeah.

    29. AEAaron Epstein22:43

        Before it'd be like to take on a huge, you know, company that has a lot of surface area in the product, it'd be like, "Well, what's your tiny entry point that you're gonna start with?"

    30. CHCharlie22:52

        Mm-hmm.
14. 23:00 – 25:20

    The future of engineering: from agent orchestration to ‘AI CEO’ interfaces

    1. AEAaron Epstein23:00

       And so, um, I'm, I'm curious, what do you guys think is next for coding? And, and what does-

    2. CHCharlie23:05

       Yeah

    3. AEAaron Epstein23:05

       ... engineering look like over the next-

    4. CHCharlie23:07

       Yeah

    5. AEAaron Epstein23:07

       ... few years even?

    6. CHCharlie23:08

       Yeah. Few ... Years is hard to say.

    7. AEAaron Epstein23:11

       Yeah, right?

    8. CHCharlie23:12

       Yeah, yeah. Months is maybe-

    9. AEAaron Epstein23:12

       Who even knows at that point, yeah.

    10. CHCharlie23:15

        I, I do think one thing we feel confident about is that the models are gonna get, they're gonna get 10, 100 times smarter. They're gonna be able to run for longer without you needing to intervene. Um, I think they're gonna start feeling more and more like a human coworker would. Um, although it's like they still ha- will have an alien brain that's very different than ours. Um, I think that's one reason we're really excited about Claude, is that, uh, the agents are gonna be able to run for much longer. And, um, I think, uh, I do think that like you will, we will need interfaces that make you feel like the CEO of a, a, a, um, even a mega corp-

    11. AEAaron Epstein23:54

        Sure

    12. CHCharlie23:54

        ... of, where you have like thousands of these AI employees that are working for you, and most of the time you're like sitting at your desk like the CEO would, and like head of engineering's giving you a report on what they worked on, head of marketing is giving you an update. Um, and you're staying high level, but then once in a while you'll, you'll need to like drill down in the details. Um, and that's sort of like what, like a PR review feels like now where you're like zooming in, um, and like asking what like the specific worker had done f- has done for you. Um-

    13. JAJackson24:23

        I think that goes back to your question about like what-What are the best people doing?

    14. CHCharlie24:27

        Yeah.

    15. JAJackson24:27

        It's like they're able to understand the, everything that's happening and, like, go look at what the agent is doing on every level and optimize, you know, optimize away the bottlenecks the agent is running into, and then also, like, think proactively instead of just reacting to whatever the agent is coming back with. Like, thinking ahead of, you know, what are the design questions I'm gonna need to, um, answer to, to write this? Or, like, what, like, which, which pieces of this do I care about? Like, which constraints do I need to give to the agent?

    16. CHCharlie25:00

        Yeah.

    17. JAJackson25:00

        Um.

    18. CHCharlie25:00

        Yeah, in some ways, yeah, like, the, the best people at building with AI are, like, also the best people at building with other humans.

    19. JAJackson25:06

        Mm-hmm.

    20. CHCharlie25:06

        Like, the skills are actually, like, pretty transferable.

    21. JAJackson25:09

        Yeah.

    22. CHCharlie25:09

        Um, and the people who were the best engineers in the pre-I- AI world are likely the best engineers in the AI world as well.

    23. JAJackson25:17

        Mm-hmm.

    24. CHCharlie25:17

        Um, so yeah, there's a lot that transfers.

15. 25:20 – 27:25

    What’s next for Conductor: ubiquitous access, higher abstractions, and better reviews

    1. AEAaron Epstein25:20

       Uh, what's next for Conductor after the, the most recent launch here?

    2. JAJackson25:24

       Right now Conductor runs on your laptop, but the vision is that Conductor can run, um, wherever you are, and it's like the ... It manages all of the AIs that run as part of your organization, and, like, you can hook into that from your phone or your laptop, um, or your, your existing cloud infra.

    3. CHCharlie25:45

       Yeah. I'm also very excited for the, like, the next level of abstraction. Like, Conductor to me, like, right now f- is in some ways very July 2025. Like, that's when-

    4. JAJackson25:56

       Yeah

    5. CHCharlie25:56

       ... that's when we created it, and, like, that's, that's w- that's, that's the state of the models at, when Conductor was introduced.

    6. JAJackson26:02

       Mm-hmm.

    7. CHCharlie26:02

       But I think ... I'm very excited to go up another level, uh, uh, of abstraction and, uh, uh, let you, uh, control the models from, um, in a different way than you can. And I think there's, like, a lot of interface product, uh, experiments to be done there that I'm really excited about.

    8. AEAaron Epstein26:17

       Very cool. What do you think that's gonna unlock for, for your users or for engineers or anybody who's trying to build?

    9. CHCharlie26:23

       I, I think a couple things. I think the first is, like, I am bottlenecked right now on my, like, working memory as a human-

    10. AEAaron Epstein26:30

        [laughs]

    11. CHCharlie26:30

        ... which is, like, three to five agents max.

    12. AEAaron Epstein26:32

        Yeah.

    13. CHCharlie26:33

        Um, and I think we can remove that bottleneck. Um, I'm also bottlenecked on reviews. Um, and, uh, like, we all are bottlenecked on reviews. Like, we have to pay very close attention to, like, each line that Claude is adding. We, we d- maybe differ from some other people, uh, on this, but, like, we are, like, try to be quite, like, considerate about the lines of code added.

    14. AEAaron Epstein26:53

        Yeah.

    15. CHCharlie26:53

        Um, um, and that's a, that's a bottleneck. Um, and so I think there's, like, a ton of, uh, experiments we can do with reviews. Um, w- we're still stuck in the 2010 GitHub PR review era, and I think, like, there's, uh, there's cool stuff we can do there. So.

    16. AEAaron Epstein27:09

        Yep. Yep.

    17. CHCharlie27:10

        Lots of exciting stuff on, in the near future.

    18. AEAaron Epstein27:12

        Yeah. I mean, you guys have, um, changed the way, uh, engineer, engineers build and the way engineering happens to this point, and I'm excited to see what you have up your sleeve, uh, going forward too. [laughs]

    19. CHCharlie27:25

        Yeah.

16. 27:25 – 29:22

    Hiring and culture: scrappy builders who make tools for themselves (and top engineers)

    1. AEAaron Epstein27:25

       Um, you guys are hiring too. Talk about, like, what are the qualities of the, the best, perfect people to join the Conductor team?

    2. CHCharlie27:33

       Yeah. So right now everyone on the team has either played ultimate Frisbee or is French.

    3. AEAaron Epstein27:39

       Okay. [laughs]

    4. CHCharlie27:39

       Um, so that's, uh, that, that probably helps. Um, but I think, um, I think people who are very scrappy and either, like, have worked at a startup or started their own startup, I think are, are great fits. People who w- want to move really fast and also, like, want to be at the frontier. Um-

    5. JAJackson28:00

       And importantly, like, people who wanna build, uh, tools for themselves and tools-

    6. CHCharlie28:04

       Yeah, yeah

    7. JAJackson28:04

       ... for the best engineers.

    8. AEAaron Epstein28:05

       Yeah.

    9. JAJackson28:05

       That's what motivates us, and it's, it's, it's especially fun to be able to work on the tool that, um, to use the tool to build the tool-

    10. AEAaron Epstein28:15

        Yeah

    11. JAJackson28:15

        ... and to, like, constantly be working at a bunch of different levels. Like, working on shipping a feature and also working on, like, what, what is the process that's going in here, and how can I, like, ship that to our customers as well? So.

    12. AEAaron Epstein28:27

        Yeah.

    13. JAJackson28:28

        Yeah.

    14. AEAaron Epstein28:28

        Yeah. It's super cool you use the word fun, and that's, that's the word I was gonna use too.

    15. JAJackson28:33

        Yeah.

    16. CHCharlie28:33

        Yeah.

    17. AEAaron Epstein28:33

        Just 'cause, uh, like, that's ... I don't know. I think that's what resonates off of you guys-

    18. JAJackson28:37

        Mm-hmm

    19. AEAaron Epstein28:38

        ... is you're working on something that you are the user of, that is being used by people that are like you, that you really care about, um, and changing the way a lot of people are, are building and shipping software, which is just super exciting. And, and what a better time than to be doing that-

    20. JAJackson28:53

        Yeah

    21. AEAaron Epstein28:53

        ... right now, right?

    22. JAJackson28:54

        Yeah.

    23. CHCharlie28:54

        Yeah.

    24. JAJackson28:55

        Yeah.

    25. CHCharlie28:55

        There's nothing I'd rather be working on, and, like, no one else I'd rather be working on it with, so it's, uh, pretty, pretty awesome.

    26. AEAaron Epstein29:03

        Well, thank you guys so much for coming in. Um-

    27. JAJackson29:05

        Thank you.

    28. AEAaron Epstein29:05

        This was, this was great to, to hear some of your insights on where things are going and hear more about how you guys operate, uh, within Conductor, so appreciate you joining.

    29. CHCharlie29:13

        Awesome. Thanks for having us. [outro music]

Episode duration: 29:22