How We Have Gone From Cloud to Terminal (And the Tech Continues to be Refined)

When I was first introduced to AI vibe coding, it used to be that you would have a conversation with ChatGPT—i.e., "Hi Chat, help me make a basic website." It responds with and you continue interacting with it in the ChatGPT chat. It was like a letterbox relationship; you’d peer through the slot, get a snippet of code, and then scurry back to your editor to see if it actually worked.

Then came about "The Platforms"—like Lovable.dev or Bolt.new—which simplified things. You could actually build cool apps and landing pages... whatever your mind could conjure and articulate on the platform. It was a lot better because context was retained and you just got a lot more made. Things like storage were a button away and completely abstracted. You didn't have to worry about setting up a database or configuring a server because the platform handled the heavy lifting behind the scenes. Again, the theme was that you could get away with knowing less... but The Platforms were a game-changer. They had their limitations, mainly cost, but things were better than they had been.

The Resident Agent and the Era of Swarms

But then... AI, which had largely been a cloud game—like an office you go and visit—changed. It was invited to live right on your desktop. What I refer to are terminal coding agents like Claude Code. This is AI that has full access to your desktop, able to interact with everything on your machine. Suddenly there was a new "intelligent roommate," and it was powerful. With full context of any project and the ability to move between files, the undying intelligence of the model could be felt—and intelligence being simply capability.

And now folks are creating swarms of these agents. We’ve moved from a single consultant to an entire automated department working in the background. In fact, you might find a good number of software engineers, or anyone in an adjacent role, doesn't write a lick of code or—at the very least—very minimal code. We have entered the "single-use plastic era" of software; if you need a tool for a specific task, you just generate it, use it, and throw it away because the cost of building has dropped to near zero.

The Realization: Coding Was Never the Job

The truth that this shift has revealed is that coding was never the actual job. many don't know how to code without AI, yet are able to produce outcomes that match those who spent years learning to code manually. The industry has shifted from "knowing the syntax" to "orchestrating the outcome." But this removed distance has invited increased danger. There is an old myth that a vampire cannot enter your home unless you invite them in; by giving agents terminal access, we’ve officially opened the door. If you have anything sensitive on your machine and you are not careful, you run the risk of leaking it. You do not control where your data goes, so it’s crucial you only expose the right files.

Looking ahead, the industry faces a growing "multiplier gap." If AI is a multiplier, the distance between those who understand the foundations and those who don't will only widen. Someone who understands the core logic—a 10—becomes a 20 (), while someone who only knows the vibes—a 4—becomes an 8 (). As this technology refines itself, the skill of understanding tech from 0 to 1 will likely become rarer. We run the risk of a future where everyone can "vibe" a prototype, but fewer and fewer people actually know how to fix the engine when the agents can't find the leak. The world has changed; we are no longer coders, we are supervisors of an intelligence we are only just beginning to understand.