On June 16, 2026, SpaceX announced it would acquire Anysphere, Inc, the parent company of Cursor, for $60 billion in stock. 

A lot of people argued that Cursor wasn’t worth anywhere near that amount. They were right Cursor, the app, isn’t worth anywhere near that. 

But that’s not why Elon Musk wanted Cursor. He wanted the people. Exceptional people who know how to effectively use AI tools.

We have entered into a new era:

Output = Capability X AI Productivity. 

And both Capability and AI productivity are distributed wildly unevenly and both have exponential consequences on output. 

AI Productivity is only starting to be understood and by the time it is understood, the world will have changed. The first industry that will really feel its effects is the game industry and it will mean the end of most of the AAA studios.

Ah, another Visitor…

AAA studios operate much like any other IP based business. They succeed because they have capital and accrued experience. 

Big projects required a lot of human beings which required a lot of capital and accrued experience means they already have the existing tech, assets and workflows in place that made their positions secure. AI tools change the equation.

Within 18 months of this writing, I predict that we will see the release of titles made by relatively tiny teams that will exceed the quality and depth of what would previously required hundreds of developers made over multiple years. And this content will not appear as “AI slop” because the people making these games will have such expertise that the output will be indistinguishable from the finest AAA artists and designers.

In other words, the big AAA studio is going to go extinct.

That’s a pretty radical claim and I’m not sure how I feel about it. But let’s walk through it together and you can tell me where I’m wrong.

The Premises

I am relying on 4 premises.

1. Human capability is unevenly distributed.
2. Capability translates to outcomes exponentially
3. Effectiveness with AI is also unevenly distributed
4. Effectiveness with AI translates to outcomes exponentially.

Part 1: Capability and the historical AAA advantage

I don’t want to focus on intelligence because intelligence isn’t capability. But most of you are familiar with this curve:

This curve is useful because its distribution applies to the other factors that determine capability in a given field:

1. Cognitive ability
2. Conscientiousness
3. Industriousness
4. Resistance to Stress
5. Domain Knowledge 

Take each of these factors, square them and add them together, then multiply the result by the 5 factors also multiplied together and you get a big number (so divide by a big number to get a non crazy number).

Here’s your average person:

The average person represents 1X output. He or she is our standard bearer here. 

But we aren’t talking about average. We are talking about people who make it through the interview process at AAA studio. We are talking about...Ralph.

Ralph

Let’s talk about Ralph. 

• He works at a major game studio. 
• He has 10 years experience working on AAA games 
• He’s pretty impressive scoring a 7 in cognitive ability (he’s really smart) and 8 in domain knowledge and a solid above average in the other categories. 

Someone this solid across the board is a rare find. But for the sake of argument, let’s say our AAA studio has managed to fill their company with people like this.

He’s 7X more productive than our average person. But look yourself, how much more productive are you now than when you started your career? 5X? 10X? 

Admit it. You reading this right now, I bet you there are many tasks that you could do in a day that you have seen people close to the average spend a week on. And not even in your area of expertise.

Ralph at a big company

Ralph works at a big game studio. His team has 250 people working at it. 

There is a cost for having such a big team and these costs are well understood and researched.

Per-person output falls as headcount rises, eaten by coordination. A century of evidence says so: the Ringelmann Effect (1913 — in a rope-pulling experiment, each added person pulled less than the one before), Brooks’s Law (1975 — communication paths grow as n(n−1)/2, so adding people to a late project makes it later), and Price’s Law (√N of the people do half the work). The standard way to model it is sublinear scaling: a team of N runs at N^(β−1) efficiency per head, with β around 0.85.

The TL;DR is that the more people in a group, the more of a output tax you pay.

Here is how that looks:

 

Part 2: AI Effectiveness

We don’t yet know what makes one person so much better at using AI than someone else. At least not definitively. 

One person using AI ends up with a bunch of AI slop. Another person using it can “Vibe code” some amazing things right now. What separates these two people? Here is what I have observed: It’s the same factors.

1. Cognitive ability determines how many different plates they can have spinning at once.
2. Conscientiousness determines how good they are at weeding out AI slop.
3. Industriousness determines how creative they will be at finding new uses for AI.
4. Resistance to stress determines how well they will deal with the anxiety of not always knowing how something is being done under the covers.
5. Domain knowledge determines how well they will be able to make use of the AI training. The UI expert who knows what a glyph is will get things done a lot faster than the person struggling to describe what they want on that button.

So now Ralph, using AI isn’t 7. He’s 7 X 7. He’s a 50! Now, before someone think’s I’m overblowing it by multiplying the two together, go back to the calculator. Change Ralph from a bunch of 6’s and the one 7 to a average of 7.5 and you get 20. 

In other words, we’re not turning Ralph into some sort of god. We are leveling him up. Take that 50 year old engineer in their prime and his output compared to someone fresh out of school is going to be a lot more than 50X. 

But the fact is, we don’t really know yet what the AI multiplier is. A year ago I would have said something different than today.

Part 3: The nine-person team

The global leaders in the tech industry didn’t start out huge. They started out very small. Steve Jobs, Wozniak, Bill Gates, etc. were all 10s. And their core start-up team was up there too. To succeed, they had to grow. And as they grew, the regression towards the mean begins to come into play.

Early Disney may have had world class animators, but when they decided to do underwater scenes, someone has to draw those bubbles. And a 10 isn’t going to stay long drawing bubbles. So let’s return to our calculator. Nine people:

1. Biz Guy
2. Creative Director
3. Lead Designer
4. Lead Developer
5. Lead Systems Engineer
6. Art Director
7. Lead Artist
8. UX Artist
9. Tech 

You can fiddle around with the positions as you see fit.

Now, let’s talk about this group of 9. I’m going to base the numbers on what I’ve seen at small but brilliant start-ups:

What I want you to take away from this is that I didn’t make these guys insanely great. They’re really capable. But not all 10s. I think, if anything, I underated. Anyone who starts a company has to be basically a 9 on resistance to stress. 

Without AI, they can’t compete:

The 9-person team gets crushed. 239 to 747. That’s why up until now, the AAA studios didn’t have to worry that much about say an Oxide Games coming in and stealing their lunch because as brilliant at Oxide was when it was founded, the seed company was only 9 people. 

I can only imagine how things might have gone when I cofounded Mohawk Games with Soren Johnson. Even without AI, it still made Old World. 

Now let’s look at what AI does.

This makes a big assumption that all 9 of these founders are masters of using AI. A big assumption today.  Most developers I know, even ones using AI, are only using it a little bit.  So, for now, you can take that AI effect and multiply it by 3% or 5% or 10% as the industry rearranges itself.

But what about a year from now?

Now, someone in the comments is going to (or should) argue that these gains are ridiculous. Except I am confident that someone else out there knows that one expert AI user can outproduce entire teams today. We’re already seeing it. 

Even if you want to scale back the effect of AI to handle all the grunt work that developing IP of any kind involves, you would have to scale it back a lot before the AAA team becomes competitive again. And remember: We aren’t even comparing the cost element. We are comparing sheer outcome.

Let me repeat that: We aren’t even comparing the cost savings in this argument. We are saying that the 9-person team will actually out produce the 250 person studio in sheer quantity and quality without taking cost into account.

Destroy him, my robots

So what’s going to happen?

The data is still early. But it tracks. Game development often involves dealing with a lot of custom made (i.e. janky) software. Setting up a “scenario” or a new character or adding a feature can be extremely time consuming that even the most talented developers and artists have to deal with. 

And I should note, this scenario doesn’t rely on AI art generation to be true. It turns out, artists are actually pretty safe. But the days of an artist having to futz with rigging or setting up UV maps or tweaking their model to lower triangle count are coming to an end. Instead they will, gasp, get to focus on making art.

The same story is true of writers. Instead of a writer spending 90% of their time trying to mess with how to get their text to actually show up in the game, the AI will handle that allowing the writer to be far more creative in their writing.

Regardless, whether AI makes a huge difference or only a modest difference, the math still reaches the same inevitability: The AAA game studio is no longer viable.