Mass unemployment due to AI is coming. This isn’t a doomer scenario anymore. Major FAANG companies are asking managers what percentage of work can be automated; “dark” factories in China run with minimal staff and save on lightbulbs by removing humans and the lights from their warehouses. I talked about this in Slop Is Enough, arguing that mediocre agents are sufficient to replace humans in most of our tasks. But beyond mediocrity, it points to something deeper. As AI gets faster and better, you’re left with a few choices:
•Learn a craft that would not be automated (yet)
•Learn how to use AI in your day to day workflow
•Retire and buy an egg farm
If you choose #2—which seems safest and requires the least change—something else happens. For coders, it means delegating the fun parts of the craft to an AI and keeping the boring ones (writing specs, checking LLM outputs, sometimes doing code review of AI slop). For most white-collar work, it’ll be the same. The high-value work shifts away from you. You’ll try to find comfort in the MBA bullshitters who insist the remaining 5% is where the moat is, but don’t doubt it: you just handed your role and usefulness to a model.
The Sloppening will lift millions out of mediocrity and help the imbeciles become yesterday’s above-average workers.
For everyone else, it means work as we knew it has ended. The feeling of being lost trying different paths to reach a solution gets delegated to what they now call “test-time compute,” and the endgame is obvious: a world where you don’t have to do anything because the AI takes those paths without you. If you’re lucky, a few “thoughts” will surface in the reasoning traces, but those are just scraps to help you claw for meaning in what your sentient AI is doing instead of you
Believers and the people who control the narrative will tell you that you have more agency now. But guess what: It’s just a transitional artifact, and it won’t last. Humans are still useful only because LLMs have made waiting tolerable again, propping up that agency story.
They will tell you that the marginal cost of software is lower so you can build more. They‘ll say that things that take 2 weeks will only take a day. They don’t mention that soon inference will be so fast that even the waiting won’t need you: sub-second LLMs will generate software as part of their answers and fully conceal that any code is running to reach a conclusion. You, the bottleneck, will be totally erased from the equation.
It’s like realizing you’re not the main character of your own story and taking a back seat to everything that mattered to you. If your goal was to learn, progress, achieve greatness in the ways that used to be valued… that’s gone. In a generation or two, what define us a human, this desire to learn, experiment, think, do trial and error, fail and get stuck in a problem will be designed out. We will make fun of the few people that still choose that path, and when the last of them falls quiet, mankind inner monologue will simply fade away - unnoticed and unmissed
the bottleneck never was code. it was the gate that was keeping away the Elizabeth Holmes and SBF from software engineering and it just opened.
anthropic "ethics" are soviet-era brainrot that make the model appeal to the class of people that wouldn't even know they live in a soviet hell hole.
android « taste » from the hipster sf events to the noads ads are just a symbol of how normie and consensual their leadership is. not that OAI is better but somehow even with everything altman said and did they appear more honest.
they tell u validation is important for agents to work well. in fact we need it bcz the stuff doesn't work. it is just raw dogging every problem with sometimes the dumbest take ever a SWE will come up with. it is just because our stds are low that they will make us obsolete.
AI today feels like crypto early nft era for the numbers of bros and scammers shilling their multimillion vibecoded app that no one will ever see
