February 15, 2025

AI replacement theory

As developers realize the superpower bestowed to their productivity by AI, there has been rising sentiment to the idea that they’ll no longer have jobs. I think that’s not quite right.

Imagine this hypothetical situation: AI can build any app, of any scale, with ease. It can fix bugs, deploy infrastructure, do QA testing, build beautiful user interfaces, meet SLAs, and so on. It can, for all intents and purposes, replace every developer. What happens then?

Evernote

As of writing this I pay $10.83/mo for Evernote. It organizes all of my notes, is multi-platform, and I find the UI to be adequate. But in this hypothetical AI future, ChatGPT can design, build, and manage my own version of Evernote seamlessly. Infra costs aside for now - why would I continue to pay Evernote?

Evernote is gone. But it’s not just Evernote. Every person and business now runs all their software in-house, AI-powered. Their CRM, their office products, their chat app, their website everything is now bespoke, in-house developed-and-managed by AI.

Salesforce is gone, SAP, Dropbox, Slack, Zoom, Microsoft Office, Service Now, ZenDesk - hell, perhaps even operating systems - all go the way of the dino. Trillions of dollars of market cap, hundreds of billions of revenue, and millions of jobs gone. And those are just B2B products - product development tools: Adobe, Figma, LucidChart, Ableton, all of these one off development tools - antiquated years before.

Going meta

Digital technology has always been extremely meta”. It is, in fact, the definition of meta. It is an abstraction unto itself - that’s really what a programming language fundamentally is. A bit - the lowest level programming primitive - is an abstraction, denoting if something is, or isn’t. Digital technology has always been about going meta, and AI is just the latest, and perhaps final abstraction of it.

We went from distributing goods in person, to doing it on online storefronts, to doing it through Amazon. All of this is going meta”. And now, the only thing that’s left, is the distribution mechanism itself to become automated.

Look at the education economy as another example: in eras past, we used to read books. But the textbook was tossed aside in favor of the tutorial. Today there are entire companies focused on standardizing and making knowledge more comprehensible. There are tens of thousands of courses on Udemy and Pluralsight which are nothing more than executive summaries” of information that is already freely available. But with AI at the stage that it is today, all of that value is already gone. I can have ChatGPT give me a TL;DR of any subject on earth. It is not only cheaper than the existing methods, it is better.

Aside from a few programs, I think education is done as we know it. Imagine being a freshman in a CS program in 2025.

It’s not so much that AI will replace developers. The way we interface with technology will shift at a fundamental level. The interfaces will be aggregated at taken over by AI, as we see being done already. If we want to buy or sell goods to each other, there needs to be some interface allowing us to transact. In many cases I suspect these will be the usual players: Amazon, Visa, etc. But I also suspect there will be many more opportunities in this space. The yet-unsolved problem is the interfaces that our AIs use to communicate.

If you want to buy bedsheets, you’ll tell your AI butler that you want new bedsheets. He’ll take your preferences into consideration, doing his best to find the best sheets at the best price, utilizing every preferential dimension in his calculation. He’ll then utilize some sort of interface like Amazon but for AI, where he can make the best purchase calculation possible, come to a conclusion on a price with the seller’s AI, and transact via some crypto-style currency. The future here feels very distributed! And on the producer’s side, they’ll have an AI provide them with minute to minute data of the best types of sheets to produce based on consumer preference. Commoditization will know no bounds.

Wrapping up

One thing is clear: digital artisans have lost. Learning optimized workflows in Blender, programming subtractive synthesizer harmonics, and understanding esoteric of programming language philosophies - all of these will be no more than a few bytes of data in a nameless server rack. Tears in rain. Enjoy the gains of your esoteric mastery while you can friends, we are not long for this world.

Perhaps this is punishment for man’s hubris: sacrificing gaia for a false digital replica. Or perhaps this is a renaissance of sorts; man’s return to anima, unburdened by the divinity (and insanity) of abstraction.

If any of this comes to pass, which I think we are still a very very long ways away from, then my dear developers, there are much bigger concerns than AI replacement theory. To replace developers” is to reduce our digital surface area altogether.



Copyright Nathanael Bennett 2025 - All rights reserved