https://www.youtube.com/watch?v=X2cFbpQCZec

Sasquatch B Studios just published the most honest 15 minutes on AI and game development I have seen, and it is not really a video about a tool. It is a video about what happens when the means of production stop being scarce.

What the video actually shows

Brandon reverse-engineers two games with a repo called "reverse engineer anything", currently the number one trending repo on GitHub. It is not magic: it is an agent wired into the reverse-engineering toolchain that already existed, pointed at a running game.

First test: the Ori games. He isolates the variable that controls how "squishy" the ground feels and toggles it ten times over, in real time, in minutes. A playable mod, authored by conversation.

Second test: Hollow Knight. He had spent months hand-tuning his own platformer controller. Claude reads the movement code out of Team Cherry's game, rebuilds it, and the scripted player-motion test runs identically in both builds. Then he has Claude draw frame-by-frame animations through a small "easel" program, then environment art, and plugs all of it into the clone. Movement, animation, art, in a little over an hour.

His own video is the proof of concept: Claude composed the soundtrack and edited the footage — screen recording, audio edit, cuts. He wrote the script and spoke into a microphone. The rest of the labour in that video was not performed by a person.

And the surrounding news he reads out is the same story at industrial scale: Anthropic going from roughly 2,500 code merges a day in May to over 30,000 a day by September, to the point where git could not keep up and they had to write their own system; seven Adobe applications rebuilt in Rust; PS5 and Switch 2 jailbreaks landing in weeks after years of "impossible"; indie studios pulling their Steam Next Fest demos because someone could clone them before launch, with one studio adding the word "original" to its store images to prove which game was which.

The part the discourse gets wrong

The genre response — "AI makes impressive tech demos but not scalable code" — is the coping mechanism of people whose moat just drained. Brandon says it plainly: that is false if you know what you are doing, and there are principles, right and wrong ways, things to learn.

He is right, and the reason is material rather than mystical. What the model did here was not invent Hollow Knight's movement. Hollow Knight's movement was already public knowledge. Anyone could buy the game, feel it, measure it, study it, describe it. The information was social. What kept the knowledge locked up was the cost of applying it: ten thousand hours of labour, a proprietary engine licence, a toolchain, and a working week.

That combination — social knowledge plus privately owned means of applying it — is exactly where rent lives. Collapse the second half and the rent evaporates. That is why the loudest anger in the game dev scene is not about bad art. It is about a barrier that used to protect a position.

The reply is about property, not about craft

Notice what the defensive moves are. Not "this model output is worse" — it demonstrably is not, in the narrow, mechanical sense. The moves are: pull the demo so it cannot be studied, watermark the store page, threaten the repo, lean on the platform holder. Every one of those is an appeal to ownership, not to quality.

This is the same reflex as Adobe's licensing panic and Sony's and Nintendo's console lockdowns. None of those companies are defending art. They are defending the terms on which art is allowed to be made — the terms that route a cut of every project through their accounts. When a toolbox that used to cost a subscription starts costing an afternoon, the objection is never phrased as "we are losing revenue". It is phrased as ethics, because ethics travels further.

Brandon's line that "anyone can make games now, but not anyone can make good games" is correct, and it is also a class analysis he does not quite finish. Design, story, atmosphere, mood — those are not scarce because they are rare talents. They are scarce because they cost time, and time is what most people do not have to spend speculatively. Lowering the tooling cost does not create taste. It does redistribute who gets to try.

The bad news he is honest about

Here is where the video earns its keep. The same collapse of required labour-time that lets a hobbyist prototype a platformer in an afternoon is what devalues the professional's afternoon on the market. "Anyone can do it now" is a wage statement as much as a creative one. If the work is cheap to do, the work is cheap to buy. The great humbling of game developers that he names is not a crisis of competence. It is a repricing of labour, arriving the way it always arrives — as a personal feeling first, and an economic fact second.

And the productivity gain does not stay with the person who gains it. It flows to whoever owns the machine that produced it: the model, the tokens, the compute, the licence terms. He is paying per hour to a company to replace his own editing labour. That is not liberation from work. That is a new landlord on the same plot.

So the honest summary is neither the hype nor the doom. The forces of production got dramatically better and the relations of production did not move an inch. The model did not socialise anything. It just made the existing property arrangement cheaper to operate — and more visible.

Luddism, done properly

The instinct to break the machine is not stupid, and it is not new. The original Luddites were not afraid of looms. They were against machinery introduced to cut wages and dilute skill while the productivity gain went to the owner. The analogy holds almost too well: the fight is not over whether the agent can write a movement controller. It is over who captures the hour it saves.

Which leaves the actual question, the one the video circles and never asks directly: if a machine can now do the labour of a game in an afternoon, why should the product of that afternoon be private property at all? Why should the knowledge of how Hollow Knight moves be a thing you get sued for studying, when the model that studied it is trained on everything everyone ever published?

What I take from it

The tool is real, the speed is real, and pretending otherwise is a losing strategy for anyone who needs to eat from this industry. But "adapt" cannot mean only "buy tokens and go faster". It has to mean being clear about the two things that changed and the one that did not. The cost of applying knowledge collapsed. The cost of acquiring knowledge collapsed. Ownership of the machinery that did both is as concentrated as it has ever been.

That is where the next fight is. Not in whether Claude can clone Hollow Knight in an hour — it can, and the video proves it — but in who gets the hour back.

Watch it: https://www.youtube.com/watch?v=X2cFbpQCZec

---

Tags: AI, Game Development, Automation, Political Economy, Marxism, Intellectual Property, Labour, Class Analysis, Means of Production, Hollow Knight, Indie Games, Claude, Reverse Engineering