Thanks But No Thanks on the Claudeswarms, Kevin Roose

It’s generous of Kevin Roose, New York Times tech columnist and co-host of the Hard Fork podcast, to pity people who are toiling away without the benefit of claudeswarms. 

i follow AI adoption pretty closely, and i have never seen such a yawning inside/outside gap.

people in SF are putting multi-agent claudeswarms in charge of their lives, consulting chatbots before every decision, wireheading to a degree only sci-fi writers dared to imagine.…

— Kevin Roose (@kevinroose) January 25, 2026

 

In a January 25 X post, Roose said that he has “never seen such a yawning gap†between Silicon Valley insiders like him and outsiders. He says the people he lives near are “putting multi-agent claudeswarms in charge of their lives, consulting chatbots before every decision,†and “wireheading to a degree only sci-fi writers dared to imagine.â€

Hard Fork involves a great deal of guffawing from Roose—mostly directed at his more comedically nimble co-host Casey Newton—so it’s not lost on me that Roose is trying to layer some irony and exaggeration on top of his condescension in this post. He takes that mask right off, however, in his next one, in which he says he wants “to believe that everyone can learn this stuff,†but frets that perhaps, “restrictive IT policies have created a generation of knowledge workers who will never fully catch up.â€Â 

Recent Hard Fork episodes have been unusually enthusiastic about vibecoding—using AI tools to perform speedy software engineering. Once upon a time, Github Copilot and ChatGPT caused software engineers’ eyes to bug out because they could write code like a person, and you could run the code, and the code would work. Since around 2021 AI’s knack for coding has been steadily improving, and steering certain software engineers toward prophesies of various forms of Armageddon.

For instance, Dario Amodei, the CEO of Claude parent company Anthropic, published one of these earlier today in the form of a 38-page blog post. “Humanity is about to be handed almost unimaginable power, and it is deeply unclear whether our social, political, and technological systems possess the maturity to wield it,†Amodei wrote. 

Roose and Newton are not, first and foremost, software engineers, but Roose recently used Claude Code to make an app called Stash, an experience he talked about on Hard Fork. Stash is a read-later app like the discontinued Pocket, or the still-extant Instapaper. Stash, according to Roose, does “what I used to use Pocket for. Except now I own it and I can make changes to the app. And I made it, I would say in about two hours.†Well done. Sincerely. 

In another episode of Hard Fork, listeners provided their own stories about what they’ve been vibecoding. Presumably these people didn’t used to code, and now they’re coding, which is admittedly kind of cool. One built a tool for wallpaper clients to calculate how much wallpaper they need to buy. Another built a gamification system for his kids’ housework. 

With all due respect to these people and the neat stuff they’re pulling off with vibecoding, this is just people giving themselves busywork for fun. There’s nothing wrong with that, but that’s what it is. 

It’s true that most people don’t have the knowledge to perform software engineering tasks, and it’s intriguing to try vibecoding if, like me, you’ve never coded anything. I’ve had LLMs make some rudimentary side-scrolling games, build ray-traced 3-D environments in javascript, and perform some other little experiments that glitched out. I learned a little about LLMs, but it didn’t change my life.

Then again, I, like many people, am bored by optimization and productivity hacks, and it’s not in my nature to have software ideas that are purely software. In rare cases where I feel a creative spark that involves coding, the coding tends to be a small part of the idea, and the rest of the idea tends to involve a lot more engaging with the world than an LLM can do. For instance, I live in one of those neighborhoods where people go nuts with their Halloween decorations, and I’ve daydreamed about setting up festive lawn animatronics, but vibecoding a control system would only get me so far in the process of configuring my monsters. Most of the actual work would be me out in my yard with a power drill, wires, and stakes, futzing with my werewolf dummy, and Claude Code isn’t on the verge of getting that thing to stay upright on my lawn. 

Roose and other AI fanatics are talking lately as if It’s. Finally. Here. They make it sound as if AI is really about to take off, and the normies need to strap in. 

Next 6 months are going to be really weird. https://t.co/TAtAomZQzb

— Alex Graveley (@alexgraveley) January 25, 2026

 

When Roose talks about these benighted “knowledge workers†outside of San Francisco, if he exclusively means software engineers struggling to accomplish tasks that could be performed by claudeswarms (Claudeswarms, in case you’re wondering, seem to be little virtual coder hives that carry out complex coding tasks), I suspect his pity is misplaced. If AI-inclined coders are not allowed to use the latest AI tools while they’re on the clock, and they’re also software engineers in their spare time, it stands to reason that they’re playing with AI toys at home if they want to. 

And there can be little doubt that, half-joking or not, Roose’s experience of people in the Bay Area “wireheading†and constantly asking chatbots for life advice is real. That’s to be expected. They have a lot of other problems too, like a horrifying new habit of injecting themselves with peptide solutions they bought online. 

It’s not at all surprising that people in San Francisco think AI is about to become the closest possible thing to a god, because it feels like it’s close to being the thing a lot of people in San Francisco think is a god: a software engineer. An understandable mistake. 

But the rest of the pathetic knowledge workers who aren’t blessed to be in the AI haven of San Francisco don’t necessarily believe software engineers are all that powerful, and some of us are counting the months until next Halloween, and AI isn’t going to be much help getting our latex clowns to look scary by then. It probably never will, and that’s fine.