Making things got cheaper, deciding did not

On the mobile side we put out close to 250 prototypes in two years. We doubled down on 17 of them and increased development, 9 of those went to soft launch, and only one went global. As I have described before, we had made prototyping incredibly fast and got up to 2 games a week. But deciding which one would live never got cheaper, we held the same meetings, looked at the same numbers and decided in the same amount of time. On paper we looked much faster, but the most expensive and slowest thing in the games industry is deciding.

Now the same thing has started happening across the industry with AI. A designer can put out a prototype in two days that is playable and looks 80% finished, work that used to be a month of five people from different disciplines. While putting out a prototype gets this cheap, deciding which game you will finish still costs the same.

There is a rule all of us in the games industry know, the last 20% of the game eats 80% of development. Where AI speeds things up is that first 80%, so you get to the point where the real work starts sooner. The remaining 20% sits there as it was, and most of the weight of development is in there.

The answer I usually hear is this, we already use AI, we put out prototypes ten times faster. Good. Ten times faster prototypes means ten times more decisions and ten times faster decisions. Producing can be handed over to the machine, filtering cannot, and your filtering capacity is the same as it was two years ago. The bottleneck left production and came to you.

Your definition of done does not work at this speed. Something that comes out in two days and looks 80% finished is not 80% finished, it only looks that way. There is as much work left as there always was and most of it is the part you cannot see, optimization, edge cases, the save system, console buttons and dozens of small improvements made for the player. Yesterday I wrote that nobody on your team owns the calendar, and there I also said that being visible in the game and being finished are not the same thing. As the speed goes up, the gap between those two opens right up.

Your working rhythm is the same. The weekly meeting was designed for a world where a week of work could be looked at in a week. If two days produce a month of output, in your weekly meeting you are looking at four months of work at once, and in that meeting nobody really looks, everyone just approves. The same thing happens with the pull request, the more there is that needs reviewing the closer the reviewing gets to zero. I see it in myself too, I used to read every piece of code AI wrote, now I do not even open most of it.

The real cost is still killing. Killing a prototype asks for approval from the person who made it, it weighs on the team psychologically, and it usually means burying an idea somebody defended for weeks. So in most teams nobody holds that authority, and because the prototype does not die it counts as having lived. This cost does not get cheaper because production got cheaper. We had the same problem in mobile, when we had to kill a game that was doing well, one of my developers working on it was so affected that he asked for three days off.

The most valuable thing now is taste. The person who knows taste and can manage it becomes the most important piece of the team, and Rick Rubin in music is the clearest example of this. The games industry always had people like this too, but they will stand out even more as AI develops, because they will be the biggest bottleneck. On a small team that person is usually the only one and you cannot parallelize them. If two people on a three person team produce ten times faster, the third person cannot have taste ten times faster or decide ten times faster.

Do not take slow down as the conclusion here. As long as it is managed, speed itself is a good thing, but you have to stop working with a process that was not built for that speed. The old process does not slow the speed down, it takes the time you gained back at another bottleneck.

Speeding up production is a matter of tools, speeding up filtering is a matter of capability.