Gene Kim

Thank you to all the speakers who shared their amazing experience reports and raw opinions of this weird world we've entered, where the cost of code production is going to zero and happens instantaneously.

And I love what Charity says — unlike other sane fields like civil engineering, these transitions were at least grounded in reality before. I think a lot of us were genuinely surprised that you sort of conjure up code just by talking to your computer. That's just not supposed to happen.

So to set the stage for what's going to happen next, which I think is going to be so awesome, I want to tell you a story.

So it's 1992. I was a sophomore at Purdue University studying computer science. Like Steve Yegge, I was taking a graphics course where I spent an entire month writing polygon shaders with light sources — which no one does anymore, right? After GPUs came a couple years later, no one ever wrote polygon rendering code before.

But one of the cool memories I have was driving with my roommate, Tom Cronin, in his 1984 Ford Escort, with two other people, to compete in a gaming contest. And so this was actually being held at the University of Texas, Austin, by Dave Taylor, famous for id Software, and Lord British, famous for Ultima.

And so we went down there. We drove two and a half days to compete and to win.

And it was so cool. And for me, it was this kind of magical memory. We were all competing — we didn't know the rules. It was four people on a team, and we were competing in a taxicab battle contest. We wrote code that was a client. We'd connect to a server. We had to navigate a maze, pick up passengers. We also had to compete against other taxicabs. We could all throw grenades 360 degrees. You could fire a weapon outside the driver's side.

And I can't even remember how we did it. This was before HTTP servers — it must've been raw TCP sockets. Anyway, I wrote the fire control code. Tom wrote the movement code. Peter Rudd did something, and Brian Batchelder wrote the maze navigation code.

And I'll be honest, for decades I was so mad at Brian, because his maze navigation code would run for a little while and then it would segfault and kill our entire client. And so on the screen we'd see our cab driving around, and it would just stop — because Brian's stupid code made a segfault.

We got into the quarterfinals because my code worked, and as long as cars drove past us, we could at least get some points.

And this was obviously — we were using C, long before memory-managed runtimes, before Smalltalk, because we weren't smart enough to use Smalltalk.

Anyway, I actually ran into Brian Batchelder in 2015 when I learned he was a senior engineer at Google. He's like, "Hey, Gene." I'm like, "Brian Batchelder — if it weren't for you, we could've won." And Dave Taylor and Lord British would tell you, "Man, you guys are pretty good."

Anyway, so we were thinking about a fun activity to do tonight, and I just remembered that experience. And I was thinking about how wouldn't it be great to recreate that moment here — where we could build a bot, put it into an arena, have them battle each other, maybe like BattleBots.

And so here to tell us about what they created is Brian Wald and Brian Scott.

---

Brian Scott: What Gene did not tell you is that he still has not forgiven Brian, so he's enlisted two other Brians to build a new game so that he doesn't have to think about that anymore.

So yes, we're going to jump into game night and the party after here. There's going to be the bar back there, which is open for you to go get drinks. Over here there's the console gaming. And in the back corner there is all of the analog gaming and tabletop games — you can pull anything off that shelf and play those.

And then up front here we'll have the game that we've been talking about, that we've been building, and Gene had started a week or two ago. And so we wanted to just give you a little bit of an update on what it is, and we'll get into more details once we get everything set up here.

We'll have the six contestants up here who signed up earlier today. So all of you who signed up, come meet us up here so we can talk you through the instructions on how to do this.

But the concept is going to be: we have this BattleBot game, just like it is on the Discovery channel or whatever channel that is, where they put the bots in and everything just goes off and runs, right? So you will be given basically an instruction file. There is no repo, there is no nothing. And you will have your AI coder grab that down, give you the context, and you're off building your agent — using your agent to build your bot. And then you enlist it into the battles, and we'll do multiple rounds so you can iterate on that and monitor it and do all these things.

And then we've done some really cool things in the background to make that happen.

Brian, I don't know if you have anything else to say on this right now?

---

Brian Wald: Yeah. Just to give you a little bit of context — you can leverage any agent you want. You can leverage ChatGPT, you can leverage Claude. If you don't want to use any of those, the folks over at Sourcegraph also have Amp. So feel free to use whatever you want to use.

And with that, again, we'll give you everything that you need. And what's really cool about this is that you can change the tide of the actual winner in games. So in real-time, you'll be able to actually modify your bot to help you improve and change the tides of who's actually going to win.

---

Brian Scott: Yeah. And so we'll be doing some commentary on that. We'll be talking through what people are doing, making it very interesting. The game itself is interesting — a lot of great background into it you'll hear. So hope you all kind of just stick around to watch that, but also go off and play the other games as well. We'll do a few rounds of this, hopefully, if we have time.

If there are any questions or anything else, come up and have a chat with us. Thank you.