Interview: Valve on the future of Team Fortress 2

Tom Francis


tf2 thumb 2

Team Fortress 2 has updated all nine of its classes now, so everyone's curious about the future of the game. I posted myself to Mann. Co headquarters in the hope of wrestling Saxton Hale for answers, but when I arrived I found only project lead Robin Walker, in his throne room at Valve. Disappointed, I clambered out of my box and interviewed him nevertheless.

He talked about where Team Fortress 2 can go from here, how many people are still working on it, what they plan to include in the next update, why unlocks work the way they do, why they might start a new TF2 beta, and what the point of hats is. Accordingly, the interview is long. This is part one - part two is here .

PC Gamer: So you've updated all of Team Fortress 2's nine classes now, I guess a lot of people are going to assume you're done with the major updates. Is that the case?

Robin Walker: The short answer is that to us there's not really any hard line between the last update and the next update or anything, you know, we're just going to keep going. We've ran into this with the classless update where we were like “well we have this big chunk of content and we want to ship it, it just happens not to be centred around a class”.

The centering of updates around a class initially was just, if you go through the history of the updates we shipped prior to the medic update, which you can do by looking at the release notes in the 119th update, we were trying various strategies, we were exploring various sizes. How do people respond when we ship something this size, how to people respond when we do something small? You're trying to find what's the balance between rate of release and size.

What we found is that if you go too fast then there's this continual level of noise and that seems successful at satisfying your existing customers but doesn't become visible enough to find new customers, so we kept trying a bunch of things.

Valve's definition of hats extends to Elvis hair.

Robin Walker: The medic update was the first time we really succeeded. It was like, 'okay, this is a way for us to do an update of a sufficient size that we can get press to talk about it and people outside the existing customers see it, and are interested by it and so on'. So, we stuck to the classes for a while because it made sense. As soon as it stopped making sense the classless update came along where it had a bunch of stuff that didn't have – it needed some other hook, some other press story that you could write about it, and so that's why we invented the classless update.

The 119th update was the same sort of thing where, here's another bunch of update content that, you know, it was supposed to be the hundredth but that's the way it goes. Anyway, so, the next update isn't a class update but it has another big chunk of stuff to it, we've got the large- people already know that the Polycount items, the results of the Polycount contest are going to be in that, so that's a huge bunch of content that touches six different classes, so there's going to be lots of different goodies in that stuff, too.

PC Gamer: It seems like you're going to run out of names. You can't call them all the Classless Update.

Robin Walker: We were pretty happy with the classless update. Hey we have no class around here, well that was well proven with the Jarate. Whatever the next update is, we'll have fun with it.

PC Gamer: So do you already know what the next update looks like?

Robin Walker: Yes. In terms of what we'll say at this point, it will have five winners of the Polycount pack, so that's somewhere around twenty items, I can't remember the exact amount, and then we're picking another one of the pack as a surprise bonus, now making it not a surprise! Fail. We're going to ship a new game mode and then we have a couple of other exciting features that we've been working on for a while that should be really really fun. So yeah, we'll probably start talking about that stuff soon once we've figured out how to talk about it.

The membership badges encourage players to take pride in being a hardcore fan.

PC Gamer: It seems like the significant thing about the class updates was that for the whole period of time from the Medic update to the Engineer one, people knew what the schedule was. They knew there were going to be a certain number of updates. Do you think that was important?

Robin Walker: I think it's important. Lots of customers seem worried that at any moment we're just going to say “bye!” I guess I can understand that philosophy given that that's how many other companies have worked. At some point we're going to have to stop on TF2 and do something else. We'll have to take a bunch of what we've learned here and apply it to other products. Probably the best answer I have to this is one I gave in an interview with a fansite recently where they asked “what guarantee can you give us that you won't stop after the next update?” I can't give them a guarantee, but the reality is that I couldn't give them that guarantee at any point in the past.

If it had made sense for us to stop after the first two class updates then we probably would have stopped then. It didn't make sense then and it doesn't make sense now. We're still learning and we're still growing, building useful things both for our customers and for us and our future products, so at some point we'll stop but for now we've got lots of ideas and lots of stuff we probably want to do.

PC Gamer: How's it doing commercially? I spoke to you around the time of the Sniper/Spy one, and you said your maximum concurrent users had doubled as the result of that update. Is that still happening?

Robin Walker: We vary across different updates. The Engineer's probably the most successful update. That's if you ignore the Spy/Sniper update, when we made it so that no-one could get their items through achievements. Spy/Sniper was the first update when we said “here are a set of new items but you don't get them, you have to wait for them to drop randomly”.

We changed our mind on that, because it was kind of cruel. We saw a massive inflation of the player numbers there, players who we don't think were having fun in the way that we like them to have fun. Once we added achievements for the Sniper/Spy items pretty shortly afterwards the numbers dropped down dramatically.

If we ignore that specific spike, the Engineer update is our most successful update that drove a huge number of concurrent players and we're quite happy with the response across the board on that. The reality is, any updates like this over time are hard to compete with the amount of money we can make on Portal 2.

For us it's not about money, right? There's a lot of other value we get. We really like the fact that we're making a lot of customers really happy. We think that has a huge amount of value. We like the fact we're exploring. We've learned a ton about how to do this kind of thing. This sort of process is  important long term for maintaining the kind of products that we think are going to be very important in the future. Both for us, and for other people.

The community now design more of the hats Valve add than Valve do.

PC Gamer: If you were a new company and TF2 was your first game, and you'd made a decent amount of money from selling it in the first place, would you be able to sustain yourself with free updates, just from the sales that come from that?

Robin Walker: It's a hard question. Valve can do that. We made more money than it costs us to maintain TF, but we have a lot of other methods to make that money. You need to be able to generate visibility at the level that we're able to generate, and you need to be able to translate that into sales. It's a hard question. I think that's true of any business proposition. If you look at any way that someone else runs their business, you need to be very careful about assuming that if it succeeds for them, it'll work for you, right? That's just the reality.

PC Gamer: Are there any class updates that you've done that you feel you need to revisit?

Robin Walker: I don't think there are whole updates or anything. There's always things we'd like to tune a little better. The shortlist of them right now is probably,

1) I think Natascha is a little too powerful right now.

2) we're still not super happy with the Sandman but we might just call done on it. Lesson learned and move on.

There's still things to try with various numbers here and there. But we have a set of larger scale things that we're interested in trying out that we can't try out internally.

An example of that would be something like, what happens when everyone has ten times as much health, or everyone has ten health? We'd like to know, not necessarily because we think that's the way it should be, but what do we learn when we do that? What happens when combat takes minutes to resolve instead of seconds? What are the effects of that kind of change on all the other systems in the game?

There are systems in the game that are tuned to specific numbers. The amount of damage the sniper rifle does fully charged, for instance. On a headshot it's specifically tuned to the health of the other class, so there would be damage numbers we'd need to change. It's hard for us to run that test and get good data, so, we're probably going to gear up the TF2 public beta again soon and make it publicly available to everyone who owns TF. We'll start running some more wide scale experiments there.

I should say now: we're not proposing this as a change to TF2. It's more about how can we use this system we have to teach us some other things that might be useful in TF and might be useful in other games.

PC Gamer: So, game design lessons in general then?

Robin Walker: Yeah. One of the things we're about to deploy is a large scale overhaul of the stat tracking in TF. We'll be collecting way more data and have a much much better way of combing through it, and we'll ultimately be providing public API's to that sort of stuff so anyone out there who wants to go stats crazy about TF can do that. There are all kinds of interesting questions you ask yourself as a designer. When you have other products that have been successful it's hard to know what are the pieces that made that product successful, and one of the ways you can do that is to start messing with them and trying pieces out.

Currently which hats you own is random: trading will make preference a factor.

One of the characteristics you see amid some of the very successful FPS's like Counter-Strike and Modern Warfare is that they tend to have very lethal combat. More lethal than we have, but we never have a customer say “we wish the combat was more lethal in Team Fortress”. We do occasionally get customers saying they wish it was less lethal.

Those are interesting pieces of competing data. Now it might be completely irrelevant to the success of Counterstrike and Modern Warfare and their success is completely due to different things, but if you had a large scale system you could test that change in, you'd get a whole lot of data back, both from from customer feedback, but also data back in terms of what happens to the average time it takes for maps to resolve? What happens to stalemates? We'd  love to be able to run that experiment, we're at a point where we can run that experiment, and we should run that experiment, so let's run the sucker and see what happens.

PC Gamer: It seems like the tension's between: combat with loads of hit points gets more strategic, the difference between the power of different weapons is much more significant, but the trade off is that it's incredibly unconvincing and it doesn't feel very satisfying.

Robin Walker: Theoretically, the longer combat takes to resolve, the more affect skill can have on it, right? If you think of every input into the game is an opportunity for you to demonstrate higher skill than me, then the greater the time, the more input, the more opportunity for the greater disparity we should see between low and high skill players in terms of combat outcomes.

This is all theoretical. The interesting stuff is often the stuff you can't see on paper, when you run it through the actual system of having real players playing with it that you'll start to see the interesting things. Most likely what will happen, based on my experiences in the past, is that you won't really find out overall whether it's a good change to make, what you'll find out a bunch of places where parts of the system change, hopefully in ways that are interesting to observe.

PC Gamer: I think that's why Half Life 2 Deathmatch is the best game Valve's ever done. Because when you get into a duel with someone with a gravity gun you can catch what they throw at you and throw it back to them. If that's all you're playing with then that can go on for five minutes.

Robin Walker: We'll have to add Gordon to TF2.

PC Gamer: Yeah, that'd be great, he could be the tenth class.

Robin Walker: It'll be like Half Life 2 CTF all over again.

PC Gamer: TF2 is a lot less lethal than most of the really popular shooters, and the time it really shows is when you're fighting a Heavy, or you are a Heavy. It does a have a really distinctive feel. When you're close up it's fast enough that it's convincing, but it doesn't feel like any other weapon because it's just this roar of noise and you have an instinctive sense of how long it's going to take to kill someone.

Robin Walker: That's another good example. Combat in TF2 is deliberately a lot more lethal up close. We scaled down damage over distance, and that was a deliberate attempt to make combat at distance to take longer and allow you the choice, and give you more time to make the choice of closure or retreat and so on. Up close it resolves very quickly. Was that a really good decision? It's debatable, it certainly helped to resolve some problems we saw with playtesters. Whether the game overall could be better without it is an interesting thing that we could test.

PC Gamer: It seems with TF2, every time you create a differential like that you create an opportunity for a class to exploit it, so the sniper bypasses the damage dropoff. And the fact that you can't always kill someone in one hit means you can have classes that are all about killing someone in one hit, like the Spy.

Robin Walker: Yeah, exactly.

Seriously though, this game has a lot of goddamn hats.

PC Gamer: The item system seems to be struggling to make sense to me without trading or some kind of intentionality to it. The difference between my backpack and someone else's is random stuff like hats, and dupes. Is the plan to go beyond that and have a situation where my Spy is completely different to your Spy in some way that I've chosen?

Robin Walker: The short answer is we have pieces of that coming, and trading is in the next update, so that's going to resolve that. Crafting at a high level was our way of doing, a selection of drops. One of the things people asked for is “I want to be able to work towards something”. The problem is poisoning the game in some way. If we say, ' if you want more soldier items you have to play soldier' than we start to affect the gameplay in a negative way.

PC Gamer: Isn't that what you currently do? With achievements, I mean.

Robin Walker: Sort of, but the achievements are completed rapidly, right? People assume that, and it is a factor, one of the reasons that we can't do performance related ties to item drops of any kind is that they'll get cheated. People will farm them, they'll farm them incredibly efficiently, and that's a big negative. The other problem is that it would affect the game itself, right? No matter what metric we chose there would be some part of the gameplay that was more efficient at generating that performance than others.

Let's say we pick kills. Payload maps contribute less kills than Capture the Flag maps. Now everyone feels dumb for not playing CTF. By always sticking to play time we say “you continue to play the game however the heck you want to play it”. Then  we said 'how do we get you to feel like you have some influence over your drops?'

Well, crafting was the solution to that, you can work your way towards something you care about over time, and it also gave you a reason to care about dupes and so on, so that's the history of it, I guess. We've done a bunch of exploring on this system, going back almost a year and a half. The community's found pieces of this, data files we've accidentally shipped here and there and so on.

We've done Diablo style random attribute generation and all kinds of items and we never managed to figure out was how to do it in a way that we felt didn't make the game incredibly complex. We like large scale differences. The Demoman running around with the sword and the shield next to a Demoman without it is, they're like archetypes, subclasses that are easily understood, that are visibly understood. It gets a lot trickier when we get to subtle things. We'll try that, we're still trying lots of things. It'll be interesting to see, my suspicion is there's some kind of balance between it all that's interesting.

PC Gamer: If you're tying time played to random item drops, and you're tying random item drops to intentional items in that you craft them to get what you want. Why can't you just have the time played be points you spend to unlock something.

Robin Walker: I mean, we can, right? But the short answer to that is, why doesn't an MMO let you get the item you want? That's the reality of it. The truth of the matter is that players play longer when we give them incentives. We have to find a balance between giving them what they want and the degree to which we give them what they want. This has been well studied in terms of psychology. There's plenty of studies that'll show you how efficiently.

People will eat all the sugar they can until they're sick. They won't balance it, they can destroy the entertainment value of something if you give them the lever that lets them destroy their entertainment value, so we have to find a balance between that.

PC Gamer: But it seems like, for items, your philosophy has always been that everyone should be able to get the items quickly.

Robin Walker: Not necessarily.

The Sam & Max accessories were a test to see if players would object to other games' art in TF2.

PC Gamer:You said you didn't want to poison the game by tying unlocks to a particular type of performance, and I said you are doing that with achievements in a sense, and your answer to that was that achievements are over very quickly.

Robin Walker: Yeah. But not all of our items are tied to achievements, and we've been slowly growing that number. There are a couple of things to think about.

One: the set of items that are most valued in the world, hats, are not tied to achievements, right? And the other thing is that we're slowly introducing the community submitted items that are not tied to achievements. When we do PR around a set of specific new items, the cruel thing would seem to be, okay, update's out and now you don't get it. Given our old system, you may never get it, a year from now or five years from now. That wasn't very nice, so we fixed that.

Now, when we do PR around a set of items, the achievements are a way of saying, the items we've talked up a bunch, you're going to get them. The items that the community is obviously valuing more, items like hats and so on, cosmetic stuff, those are harder. There's a direct tie between the rarity and the value of those things. It's not subtle, it's very direct. If we let people get them easily, they would be valued very little, that's just the fact of the matter.

PC Gamer: I was just saying it seems like you could tie play time to earning weapons that you've advertised and leave hats as random, and leave community items as random.

Robin Walker: How's it different from what we have, sorry?

PC Gamer: Right now you have to use achievements to work towards an item when it first comes out.

Robin Walker: Or you can get it through drops.

PC Gamer: Yeah, but that doesn't happen in general.

Robin Walker: Yeah, the achievements are the primary way people do it, yeah.

PC Gamer: It means everyone plays that class immediately to get those items, which seems like, even if it's for a short time, you're poisoning the experience.

Robin Walker: Yeah. We've gone back and forth on that a bunch and interestingly we've received feedback from customers over all the months we've done this arguing both cases. During the Spy/Sniper we tried to measure how long takes to settle down. It looked like within a week it was back to normal. What we felt was, it's like a party that lasts for a couple of days, you know, TF is different for a week, and we felt like it's a benefit to us that the spike in the update, around the PR and excitement and the new stuff always is followed up by this spike in the gameplay. Everyone goes a little crazy and has fun. We did leave that in but it has been debated about much.

PC Gamer: That makes sense.

Continue to part two, where Robin talks hats, movies, meeting the team, and facestabs. And let us know what you think of the Valve's plans for Team Fortress 2, now that the class updates are complete.

Around the web