Everything becomes anything: why we're betting on users, not IP
· Christoph Heuwieser
Video: Everything Is Everything Now, Atrioc on his Big A clips channel, 6 October 2026
Atrioc's new video starts with what looks like a joke: Mario physics in Elden Ring, Minecraft mechanics in Pokémon, Tony Hawk-style skating in a Call of Duty map. Twenty minutes later it ends on a much bigger question, and he admits he doesn't have a neat answer to it. I don't either. But I think his question is the right one, and it lines up with something we've already decided at a4sx.
What the video actually argues
Atrioc separates the viral clips into kinds. Some are what he calls a trick: two games running side by side, one character drawn over the other. Others are not. They come from decompilation, which means turning shipped machine code back into readable source. He explains that this used to be months or years of manual guess-and-check: write a candidate line, compile it, see if the output matches, repeat.
His central point, which he says the programmers he spoke to agreed on, is that current models are extremely good at exactly this kind of work:
"AI has become incredibly good at things that are verifiable." (Atrioc, Everything Is Everything Now)
Decompilation is the perfect case. Every guess can be checked against the original binary. Once the friction is gone, you can lift parkour mechanics out of one game and drop them into another, or rebuild a familiar desktop tool closely enough that, in his own testing, he couldn't find a difference in the features he uses. His summary: "Everything is becoming everything." And, more bluntly: "There's no moat on software."
He's careful, and I want to be too. He says plenty of this doesn't work yet, especially anything with an online component. He thinks a lot of what gets made will be low-quality "slop". He's openly uneasy about copycats ripping off small developers within days of a release. He isn't cheering. He's trying to understand what changed.
IP is inflating
Here is the way I've been thinking about it for a while, and his video sharpened it.
Code is starting to behave like a currency being printed faster every month. When re-implementing something takes an afternoon instead of a year, each line of proprietary code buys you less protection than it did. The value doesn't vanish all at once. It inflates away. Secrecy that used to hold because the effort was too high, what Atrioc calls security by obscurity, is weakening because the friction it relied on is shrinking.
This isn't a legal argument. Copyright, licences and contracts still exist, and copying someone's work without permission is still copying someone's work. It's an economic argument. The defensive value of keeping code closed is falling, and I expect it to keep falling.
So what is still scarce?
- People who actually use the thing, and keep using it.
- Trust: a track record of how you behave when something goes wrong.
- Community: people who report bugs, send fixes, and tell others about you.
- Accumulated real-world material: work that happened, which a model can't produce by reading a binary.
What we're doing about it
I said this to my team yesterday, close to word for word: the more time goes on, the less our IP is worth and the more our user base is worth. We won't have a moat in the old sense. We have to earn users. Nothing else really matters.
We're acting on that. We've started open-sourcing parts of our core:
- ctxnorm (PyPI) turns the different ways LLM providers report "context window exhausted" into one error shape that clients can recover from.
- redact-transcripts (PyPI) is a helper for redacting secrets and personal data from AI-agent transcripts and JSONL logs without breaking their structure. It's a tool, not a guarantee. Read the code and check its output.
If someone could re-derive these in a weekend anyway, then keeping them closed protects very little. Publishing them gets them reviewed, improved, and used by people who might never have heard of us otherwise.
The honest caveat
This logic cuts both ways. If software has no moat, neither do we. Anyone can build something that looks like a4sx. The only defence is being the place people want to stay: useful every day, straightforward about what we do and don't do, and run by people who listen.
Atrioc ends his video saying he'll keep researching and doesn't know yet what it all means. That honesty is underrated. I don't fully know either. What I do know is which assets are getting cheaper, and which aren't.
a4sx is a marketplace for AI coding-agent sessions: publish, acquire, resume and continue real work, dead ends included. If that sounds useful, come take a look. If it doesn't, tell us why.