Peter Fedichev@fedichev
You’ve probably heard the rumors that agentic AI has come for software engineers. We’re not exactly young anymore, so we’re retraining on the fly: at the office, agents first refactored our code—now they write it.
Today in Singapore, OpenAI ran a workshop for kids. They showed school students Manus, asked them to call their parents over, install everything, and then come up with something and “code” it (our try is aliendesign-w4zrvguh.manus.space). The mentor offered a simple recipe: think of an idea, ask ChatGPT to write a prompt for Manus, paste it in—and enjoy your finished app.
It’s been predicted for a while that from here on out it’ll be only fun and scary.
On the one hand, building something like this by hand would take me a week or two—and yes, it would work, but I still wouldn’t make it look this nice. For someone else, maybe a couple of days. For most people: a couple of weeks of Googling, hacks, and endless fixes, despite the triviality of the idea itself. And honestly, that’s great news: this kind of soul-sucking work is finally fading into the past. That was the “fun” part.
When the fun ends, it gets scary. You start to realize that instead of painfully polishing a paper for the last couple of months, in a couple more months you’ll be able to say: “Write a Science paper—here’s the data,” and get a very decent draft. And a year later, probably something close to final. I won’t continue—last week there were posts with millions of views laying out our near-term, not-so-much-cheerful prospects in detail (read here: x.com/citrini7/statu…).
What can you say—long-term, we’re all cooked. But as Tyler Durden said in Fight Club, the probability of our long-term survival is zero anyway. So thank God the problem is temporary: we’ll adapt.
LLMs, it seems, are by definition superhuman at supplying us with the most likely correct answers. On the other hand, LLMs don’t know how the world works—but they understand extremely well how our description of the world works.
And as we know, the most glory and money are exactly where our understanding/description of nature diverges from reality (from another movie: the key question is “what do you know that others don’t?”). That’s also, by the way, one of Peter Thiel’s questions to founders.
So modern AI will likely make infinitely cheap the work where you just need to follow consensus precisely—and sharply increase the premium on orthogonal, but possibly true, judgments.
Welcome to the brave new world. Support the younger generation in the contest: everyone—generate one alien (the image is from the app), and as usual, like and repost.
