Q&A: Valve's Swift On Left 4 Dead 2's Production, AI Boost

Valve designer Kim Swift gained fame when her first game at the company, Portal, became one of the instant classics of this generation -- particularly as Swift and her colleagues were hired directly out of DigiPen on the strength of their student demo Narbacular Drop.

Now, Swift is working on Left 4 Dead 2, the anticipated -- and initially controversial, to those who expect long gaps between Valve games -- sequel to last year's stand-out co-op hit.

Gamasutra recently had a chance to talk with Swift about working on the game, taking in discussions of both Valve process and the game's innovative AI Director, which controls everything from ambiance to enemies in the game world.

Valve joined development on the original Left 4 Dead late into the process with its acquisition of Turtle Rock Studios. How was integrating into a game that was already in progress at one studio, and how did that work?

Kim Swift: Well, it worked pretty well. They didn't want to move, and so we did a lot of video conferencing. For the most part, the gameplay was in place when we started to work on Left 4 Dead.

Let's see, things that we added sort of in the meantime were mini finales and finales. We worked on the pacing with them a little bit more, and the art changed quite a bit once we actually started working on the project.

I like how the art direction functions multiple ways. It hooks into mood and aesthetic, but there are very deliberate design-orientated functions to the art.

KS: Since we don't have a lot of HUD elements that are saying, "Go this way, and what not," lighting, and textural cues are really important to us to be able to guide the player through a level.

How does the interface between art and design work?

KS: Pretty well. I mean, we work in cabals. I'm actually in a room with 12 other guys, a mix of level designers and artists, and so we're always talking to each other and making sure we're on the same page. For the most part, we all design things together. So communication is a key.

The process at Valve is probably of great interest to a lot of people, because of the way Valve's games turn out. But it sounds like it is relatively atypical.

KS: It was definitely something that I really appreciate learning, coming right out of school and then working at Valve. We've got a very iterative process, and it's pretty democratic.

We all sit down as a group and throw out ideas, and write the ones on a whiteboard that we think are good ideas, and then start to elaborate on the design together as a group. I mean, two heads are better than one, right?

It has started to become quite more common for people to take feedback from all disciplines on design.

KS: Well, it's the best way to solve a problem, right? Because the best way to solve a problem may not necessarily be a technological one, maybe it's an artistic problem that we'll solve, or it's just actually a design problem. So the fact that all the disciplines are communicating together, we'll find the most efficient way to solve a dilemma.

So when you started working on Left 4 Dead 2, was the plan to do five campaigns?

KS: Well, originally when we started working on Left 4 Dead 2, we were just coming up with content that we really wanted to ship with Left 4 Dead 1, but we weren't able to get to it or it just wasn't flushed out enough by the time we shipped it.

So we were really excited coming off of Left 4 Dead 1, and just had a ton of ideas. Then we all sort of took off in different groups and worked on our own little ideas, and then presented them to each other. At the end of the day we just had so much content, that we decided that a sequel was the best way to go.

What's your planning process? It sounds like you guys didn't sit down and write docs. Was it your cabals that you used to come up with like a presentation?

KS: Sort of like on an individual basis. There's a time right as soon as the game goes gold, in between when it actually ships that you're just kind of sitting around twiddling your thumbs, maybe play testing the game a little bit. It's the best time to work on new ideas that you weren't necessarily able to get to in the previous game.

Before you roll into production?

KS: Yeah, before you roll into production. It was pretty organic how Left 4 Dead 2 all came together. We all presented ideas to one another coming off of Left 4 Dead 1. At some point someone told us that, "Oh, yeah. We should make a sequel out of this." We were like, "Sounds good to us."

It sounds kind of funny, just because it's really quick. A quick sequel, which is not Valve's well‑known trait.

KS: Yeah, definitely. We're pretty proud of ourselves for not moving at "Valve time". We practiced a new organizational tool. We used the Scrum method this time. We decided to give it a shot, and it's worked really well for us on the team.

And also, heading into the project, we were pretty sure about what our deadlines were, so we were able to try and be more deliberate with our planning to actually get the sequel out.

So the first game wasn't developed under Scrum and the second one is?

KS: Yeah. We decided to give that method a try.

So have you been seeing more productivity? Obviously, you have one more mission to do. Do you have a perceptible increase in productivity?

KS: I think more than anything it helps us prioritize what to work on. Since we knew that our shipping date was a lot sooner, we were able to sort out which ideas were actually doable in that amount of time, and that we knew were actually going to be successful, rather than trying a whole bunch of different stuff that wasn't necessarily going to be what we shipped.

How did you identify the things that you knew were going to work?

KS: Well, there were definitely components, like coming off of Left 4 Dead 1, one that we knew we wanted to get to and that would be a good thing to add, was mainly weapons, for one, that we wanted to add in Left 4 Dead 1, but we were unable to at the time because we just, well, ran out of time. And so right away we started to work on melee weapon ideas.

Did you improve the AI Director in meaningful ways between projects?

KS: Yes, there have been improvements to the AI Director. The AI Director is able to control the weather now, and we've also got components of certain maps that will change dynamically when you load the map. An example of one would be the cemetery area in the Parish campaign. There are different configurations that the AI Director will pick depending on how well the team is doing.

I love that idea, and it's a really intelligent idea to be able to do that, especially in a game with only a few campaigns, but which you want people to continuously enjoy the game. Was that the genesis of the AI Director?

KS: Yeah. We know people are going to play the same maps over and over again, and so if there's at least a little bit of a difference each time you play, it makes it more exciting and keeps it fresh.

So, the AI Director has purview over elements, level design, obviously enemy behavior, right?

KS: Yeah, yeah. It controls different horde structures. We've got ambient mobs now that we didn't have in Left 4 Dead 1, so it will control the pacing in certain areas. So in one area, you might come in and there's a couple of ambient mobs that are just kind of wandering around ignoring you, and then you might have a horde coming at you that's aggressive and is out for blood.

It's interesting because it can control things that are sort of seemingly more atmosphere‑related in terms of appealing to the player, but then there are things that are more direct.

KS: Well, yeah. We want that sort of pacing aspect that you get in a really good horror flick where there's moments of silence where it builds tension, and then it just lets it go.

Do you want to see that concept used in other kinds of experiences, do you think?

KS: We definitely like the experience of having the gameplay be a little bit different each time you play. I think it keeps the game amusing for a long period of time. I think in multiplayer games, it's a really strong idea.

This article was first published November 12, 2009. It has been updated in 2024 for formatting.