As anyone who reads this newsletter, listens to our podcast, or overhears me yelling at strangers on the train knows, I’m excited that tens of millions of people could suddenly become software developers thanks to Claude Code (and its inevitable competitors—there’s a rumored DeepSeek coding tool in the pipeline, which would be chef’s kiss). But I also know that whenever I feel particularly hopeful about technology, I should grab a huge ice bucket and stick my head into it, because enthusiasm often doesn’t translate into social acceptance, utility, or positive outcomes. 

Having learned this lesson too many times, I’ve been trying to figure out what I’m not seeing. Four tech-world visions haunt me: The four vibe-coding horsemen of the coming AI software apocalypse. They’re not so much about the social harms, because we actually talk about those a lot. Instead, I’m thinking about the ways that processes could be upended when everyone can produce code.

“We can deploy anywhere.” I wrote about this last week. Over the past decade, deployment has become extremely cheap—what used to cost hundreds of dollars a month now costs pennies—but incredibly complex because everything needs to be secure and backed up and provisioned across multiple clouds. In the effort to simplify everything, we created a whole new discipline, “DevOps,” to make things work. I’m not saying DevOps goes away; I’m saying that it’s a lot easier to deploy anywhere, any time, simply by telling an LLM, “Here’s a root password; see what you can do.” You can go ahead and make every possible rule against this, but then look into your heart and ask yourself: Will engineers continue to fill out forms and write playbooks for deployment, or simply tell a bot to do it?

“Just clone it!” I’ve seen this one up close: People call us and say, “Hey, I pay $100 a month for this very specific commercial real-estate platform—and I hate it.” What comes next is a litany of frustration, followed by a desire for something that works really well, just for them, along with a desire to spend exactly what they’re paying today, but for something custom. Other people have asked: “Can’t we just use AI to clone Quickbooks?” Here’s what’s dangerous: You…can. Oh, maybe not today. But in six months, I will be looking for lots and lots of clones. CRMs for $2 a month? Accounting tools for free built by teams of five? Internal copies of accounting suites? Sure! It’s all on the table. And when you say, “Hey wait, how are we going to maintain that, and make sure it integrates seamlessly with third parties, like, say, Stripe payments?” People are going to just offer that AI will maintain it, too. If you don’t see the problem…well…I’m jealous.

“Framework-first programming.” A good rule of thumb: 80% of engineering time is spent configuring your IDE or code editor; 20% is spent writing code. There’s a whole engineering glossary around these tendencies: Yak shaving, bikeshedding, architecture astronauts. AI enables this in absolutely unimaginable ways: You can create configuration languages that write configuration files. You can port code in one language to another. You can run Linux applications in the browser. You can build frameworks all day long, without ever writing code people use. Things that used to incur vast friction are now easy as pie. This is very, very good and will lead to absolutely nothing ever shipping.

“Salesforce mobile gaming.” That isn’t a real thing, at least not yet, but people are going to realize they can ship just about anything, and are going to start slipping outside of the bounds of good sense, taste, or logic. Salesforce could decide to gamify CRM. Epic Games could decide to build a CRM that runs inside of Fortnite. Facebook could decide to make its own Google Docs, so that you can share docs on Marketplace—or Google could decide to embed a social network into Google Docs comments. These companies have had the ability to ship any software they wanted for some time, but it’s always been expensive, and hard to get the extra humans you needed. Now you might find yourself in a different position: Forced to justify your headcount, and capable of shipping tons of product in record time.

That’s just four—and there are many more to come. I’d love to hear yours. I should also say that I have very mixed feelings about the “democratization” of publishing over the last few years, and the way it led to consolidation within a very few media platforms. I have a deep distrust of AI around content generation. But the fundamental “deal” of the software industry has always been that technology can shift under your feet, and you have to work it out.