"Software is Over": Someone Vibe-Coded All of Adobe’s Products And Open-Sourced Them

[LevelUP]

The first shoe to drop in the gaming industry. 

Epic Games reportedly pivoting to “AI first” game development model
https://arstechnica.com/gaming/2026/10/r...aboration/

[NotJoeBiden]

I think there is something special and inherently human about art being created by people rather than AI. Games fall into this category for me.

[SteveFoerster]

Yes, I agree. This is definitely a revolution unlike anything we've ever seen before.  I think the AI revolution will unfold in two phases: a massive gold rush followed by rapid commoditization.  Take video games, for example. Roughly 20,000 games were released on Steam in 2025, with estimates of 25,000 in 2026 and projections as high as 60,000 in 2027!

When AI dramatically lowers the cost and time required to create something, the supply of that thing explodes. And when supply explodes, differentiation becomes increasingly difficult.

For a while now we've seen this in books, podcasts, and video, where the tools to create them have essentially been free for some time and the barrier to publish so low that there is now more of each of those than anyone could possible consume. The end result is that most books go unread, most podcasts unheard, and most videos unwatched. As you said, differentiation is key.

(And that was before LLMs!)

In a world where anyone can build almost anything, the scarcest resources won't be code or technical skills. They'll be good judgment, attention, trust, and distribution.

...and nepotism.

As for why AI struggled to recreate Star Raiders, I suspect the biggest issue was a lack of context. AI can analyze text and images, but depending on the tools being used, it may not be able to directly watch and understand videos. That means it's missing critical information about the original game's mechanics, controls, timing, and overall feel.

Mostly. There are some decent gameplay descriptions out there, but I didn't prompt the models towards them because I wanted to see how they would do with minimal hand holding.

What I got was recognizable and playable. I could have refined it, but I wasn't actually trying to play the game, I was just going through the process to evaluate various models' capabilities, so when I reached that goal I left it as is.

[Maniac Craniac]

AI-supported coding is getting better. So on the one hand, given that the software in this article is described as "pre-alpha" quality, I'll hold off a little longer on Adobe's eulogy. But on the other hand, Adobe and companies like them better have a plan for what's coming.

As an experiment, I tried to get the free Nemotron models to vibecode a browser-based version of the old Star Raiders game for the 8-bit Atari. It was... well, partially successful.

Ah, seeing a Star Raiders reference on this forum puts a huge smile on my face! Star Raiders is definitely impressive on the Atari 8-bit systems, but it's basically a wonkfest that requires a lot of imagination. The fact that it's supposed to be first-person and that the objects on screen hardly resemble what they're supposed to represent could be tripping the AI model up. You might get better results with nonscrolling one-screen games with straightforward visuals like Breakout. 

Or, since you're using Nemotron, it could be fun to make Robotron. You can call it Nemrobotron. 

[DegreeNerd]

Get yourself a Claude subscription (or anything else that gives you a generous quota of SOTA models, such as Cursor/GrokBot/etc)
It will do that then without breaking a sweat.

Thank you for assuming the model was the issue and not my prompting!  It did mostly work, and I could have refined it to improve but I wasn't really going to play it, so I was satisfied with it as a demo.

For now I'm happy playing around with the free ones to get a feel for what various models can do.

Besides, frontier models may be a little smarter, but I'm not trying to solve Millennium Prize-style problems here.

Even for simple problems such as say recreating Star Raiders, then having a good model / quota / harness makes a massive difference 


To give some analogies:

Imagine you want to do something easily and simple, like watching a 1hr documentary on WW2. Think about how easy this is to do with a fibre connection and YouTube. 
Vs trying to do this pre-YouTube with only a dial up internet connection. 


Both are "the internet" (i.e. just like how both free vs paid subs "are AI")
Yet you have a radically different experience with each!


Or to give the 2nd analogy:
Imagine writing a book on clay tablets vs writing it with a computer. 
Both are using the same tool of "the english language". But one of these approaches is going to be vastly more efficient than the other!

Spending $20/month for Claude Pro to dabble with it, or $30 for X Premium, is totally worth it. 


Or heck, even just spending $10/month for OpenCode Go or $8/month for X Plus is still worth it and much better than nothing!

https://x.ai/pricing
https://opencode.ai/go