Don’t you dare rewrite the code? What do you mean?

Indeed, the code for the Social Security service is written in COBOL, and to find programmers who can support it, one has to search long and hard, paying them half a million a year.

You know, a sign of a person's rationality is their ability to see the difference between things. The more differences you see, the better. In 2003, when every programmer was on record, when to start working you really needed a diploma and a higher education, when you were paid a lot and asked for a lot, no one wanted to rewrite software for no reason.

But now it's 2025.

Let's wake up and look around. We have LLMs. And they are already really good. Some of them can handle contexts up to a megabyte. This means there's a chance that a hypothetical Claude could rewrite the entire codebase of the US Social Security service from COBOL to Go in one sitting.

You see, code itself is something that has long since lost its value.

In the old days, you could ask people to send you 20 bucks in an envelope, and in return, you would send them a floppy disk with a game or some useful program. I don't know if punch cards for computers were sold this way — probably not. But in the 1980s, you could make good money selling code you wrote over the weekend at home by placing an ad in a newspaper.

Then the internet came. People started to learn a lot more about everything. People began to sell software for platforms. If you owned Windows, you would pay anywhere from 10 to 30 bucks for a good program. People bought WinRar, Windows Commander, or game discs. At some point, software prices started to drop.

In 2006, the first public Git repository appeared. By then, Linux had already grabbed nearly all server operating systems, quickly pushing out Windows and artifact mainframes. Open-source became a viable thing. Over time, even big market giants, such as Photoshop and Dreamweaver, were replaced by something less polished but completely free.

By 2015, the market had changed again. Large corporations realized that code could be not only free but also profitable for them. VSCode emerged. The project was a hack. An ugly den of endless crappy code, which, nevertheless, was well supported, and money was poured into it. Eventually, this mess became the de facto standard in web development. (By the way, once again — https://zed.dev is years ahead of VSCode.) But this product also became the de facto standard for leaking your information, rights, and everything else. Software became free for advertising purposes.

You don't need to develop a new utility. It has already been developed. If you need video — you do it through ffmpeg. Or you go and work with DaVinci Resolve. A wonderful studio created with one purpose — to make you a BlackMagic customer.

Welcome to the latest milestone in the evolution of the software industry. Today, code is just a wrapper that requires you to subscribe to an LLM. Photoshop, Zed, DaVinci and everything else — it doesn't really matter. What matters is that you’ll be buying a subscription to their service because they have the assistant you need.

Programming by itself brings in 0 dollars. 0 cents. And in 2025 it can’t cost anything at all. Unless your code is a really intricate bytecode for a proprietary processor — then it might have some value. Otherwise, it’s worth 0.

Even if you slap a license on it and say you want to be paid for it, a few waves of an LLM will quickly create infinite clones. And if that feels too brazen, you can just ask the LLM to rewrite this code based on input and output data.

You see, the definition of “Computer”: a device that stores, processes, and transmits data. Data is the important part of this definition. The manner in which it changes this data — not so much.

Your language, compiler, or stack — none of it matters in the realm of business logic in 2025. In 2020, it was still somewhat relevant. Today, a good engineer can feed a monorepo with a couple hundred microservices into an LLM and get a functioning monolith on the output. You can also do the reverse: turn a monolith into microservices.

Are you struggling today because you lack static typing in JavaScript? Then go ahead — switch to Golang. Tired of Golang? Rewrite everything in JavaScript. It’ll cost no more than 10 bucks. Got more than a megabyte of code? Alright, you’ll have to work a week and pay a hundred bucks. How terrifying. Hook up tests to it, described by that same LLM — and you’re done.

(This doesn’t mean an LLM shouldn’t be managed by a good developer. Far from it — without this developer, the LLM will hallucinate and produce total garbage. But today, a good developer can output code a hundred times more than a 10x developer could ten years ago.)

Yes, in every particular case, a sensible developer really does need to sit down and properly analyze whether it makes sense to rewrite a codebase in a week or not. In some cases — you should, in others — you shouldn’t.

Now, let’s get back to the US social security system.

This country desperately needs to rewrite EVERYTHING from scratch, start to finish. Most government software is written for mainframes in the 1990s and still runs inside a wrapper of sixty different emulators that launch it on modern processors. They’re using Informix databases everywhere, sometimes Prolog, and here and there you’ll see FoxPro.

Do you know why you can’t transfer money in the US by card number or phone? Because most banks use ACH, which runs on TELEXES. Don’t know what a telex is? Well, you either didn’t serve in the US Navy, or you’re under the age of 80. Telexes are those machines that look like typewriters. You’d plug one into a phone line and connect with another telex. When you typed on one machine, the other would print the same thing. Of course, the machines are gone now (oh, how I hope), but banks still use software that works inside DosBox, which shrouds CSV shrouds in batch files for the telex emulator program.

Don’t believe it?

Go and try to initiate an ACH transfer between banks and include more than 80 characters in the transaction description. All ACH processors and gateways will reject the transaction. When you start digging through the depths of your logs with tech support, you’ll eventually get a photo of a CRT monitor screen, with white text on a red background: “Text does not fit in the field.” Why white on red? That’s how we liked to do it in DOS.

This system needs to be shaken up. It should have been shaken up ages ago. There’s no complexity in it. There’s absolutely no complexity. The entire US social security payments database could easily run on a single iPhone 16 with one terabyte of memory. This system is tiny in terms of implementation. But since we don’t want to update anything, we’ll spend millions of dollars writing emulators, releasing LPT printers, looking for programmers in COBOL and Prolog, and doing other such nonsense.

There’s nothing complicated about this system. We learned how to process financial transactions back in the 1990s. For 30 years now, we’ve been able to handle thousands of requests per second with ease. And 20 years ago—millions. And 10 years ago—billions. We can do all of this. It’s all in open-source code, already written and documented. We have LLMs that can rewrite COBOL into Golang. We have equipment that’s much cheaper than dot matrix printers. And it’s all readily available. All that’s left is to just do it.

But instead, we’ll sit around and complain, going on about how nothing can be rewritten because what if something breaks.

To really understand what you’re rewriting, you have to demand data. And the data—it’s available. Go here—https://www.ssa.gov/OACT/ProgData/tsOps.html, request the data—they’ll show it to you on the screen (by the way, the site is designed in the best traditions of 2006). Look at how much money comes into their funds and how much goes out. Check how much was coming in and going out in 2005. And you’ll realize that not only do we need to rewrite everything from scratch, but we also need to raise the alarm in every direction. Because no matter how many times someone says everything will be fine—you can’t fool math. Negative numbers speak for themselves. And the abundance of those negative numbers says even more.

So let’s stop pretending we’re living in 2003. Rewriting business logic code in 2025 costs next to nothing. You’re not writing video codecs and you’re not growing LLMs. You have strictly codified procedures, which can be translated to another language with your eyes closed.