In 3 Years There Will Be No Programmers

than by AI*

A non software engineer wouldn't know what to speak into the microphone to do what they want. Basic things have so many random requirements that the ordinary person doesn't think about. AI is just a tool for software engineers to be more efficient. And AI actually gets things wrong quite a bit, so the software engineer has to be able to know when AI did it wrong.

When the economy is crumbling 6 months from now, we’ll still be praising all the amazing advancements the great ai has brought us. Well, that’s it for me. Back to chat-whosit and sora-whatever.

I work at a very large tech company. When someone joins my team, it takes 3-6 months for someone to build up enough historical and organizational knowledge to start making an impact. Even then, it can take another 1-2 years to fully make an impact. The current LLMs are strong but they will never have enough context to do that. There is only so much you can fit into a context window and, to replace us, you'd need to fit orders of magnitude more in there.

That being said, AI makes some tasks a lot faster. So, maybe you need 1 or 2 fewer engineers on a 12 person team. Thats hardly enough to make a dent in the industry. Economic downturns have a much larger impact on job security.

My brother shared an incident that happened while he worked as a military contract employee. He was computer programer/code writer responsible for writing the code which helped pilots fly jets (I can't share too many specifics about the jets because he could not either) But the incident he shared was that a young coder that just started working on his team couldn't understand why his code for a specific operation was rejected. My brother explained that the code had an unacceptable failure rate. The young man argued that the success rate was over 90%, as if he earned a A for his assignment. Fortunately, a real world explanation of death to the pilot is unacceptable finally registered and the code was improved upon by the young coder. Why is this relevant? Who is going to explain to AI that protecting human life trumps cost effectiveness in rewriting code until it is perfect.

Well yeah. Problem is as soon as something becomes common knowledge, it's already too late.

When you were being told to "get a trade" you should have been going to college.

And when you were being told to "go to college" you should have been learning to code.

And when you were being told to "learn to code" you should have been working out how to do social media.

Software engineering is arguably much more complex than other types of technical desk jobs like architecture or accounting. Those other jobs would effectively be automated away first in a world where automation of desk jobs is actually happening (which is not the world we live in).

The main reason why desk jobs can't be fully automated away is domain knowledge. Inside companies there is a fragmentation of domain knowledge. Joe the accountant in office C0126 is the only guy who really knows how to do forecasts for budgets with involving Quendelton Energy Services.

"Well, we'll just train some AI on all domain knowledge across the company." I'm not going to explain all the reasons why this can't work.

Furthermore, people massively overestimate the priority companies place on optimizing their workforce. Usually they have more important business problems to solve, such as getting some new product to market before a competitor does.

Software engineering is in fact much more difficult to automate today compared to 20 years ago.

(1) Many roles have been consolidated into the catch-all role of software developer

(2) Software today involves many integrated services; the total code to create a system is spread across many different code bases, multiple API versions running side-by-side, etc.

Let's see AI attend my 6 hr/day of meetings

Cute how people think AI is only going to eliminate programmer jobs. Most people sit behind a computer: doing various tasks: calculations, moving data, emailing, designing, editing. AI can/will do all of this. Programmers or the people who write code will likely transition to prompt engineers and robotic technicians. But almost all jobs will be affected.

M67: Former network engineer. Actually, you can do that right now. But this is purely for personal use and for fun home projects. But the truth is for business use, at the moment, it is actually slowing down the programming process because it makes so many mistakes and is very insecure.

It’s very likely that for finance and security some experienced human programmers will be needed to check and verify, but they won’t be programming, just using AI tools looking for vulnerabilities and checking it against their historical knowledge.

Don’t forget AI cannot think, AT ALL, it cannot reason AT ALL, and it makes mistakes. Will it get better? Yes. Will it be able to anticipate vulnerabilities in an anthropomorphic real physical-world environments? That is not certain and that may never happen.

Also, since this is a running site, don’t F’n ask it for training advice; it will just reward and cheer-lead you on whatever stupid plan YOU seem to like. Don’t ask me how I know this……

I write a lot of computer code for my job. I also teach others how to use data. All I can say is if you have not tried claude code, you should before opining on the future of coding and AI. What claude code does is simply amazing. I can now accomplish a week's worth of work in a few hours. Is it perfect? No. But Debugging in hours what otherwise takes days is more that worth the monthly cost to access claude code. Plus, AI gets better by the week. If you do not think that is going to have a major impact on the programming industry in general, you are in the same boat typists were in in the early 1980s. And yes, I was around then too.

AI is in its infancy and a lot of people are making broad brush evaluations on its efficacy, based on how well it performs in the moment of today with complex systems that were built by humans, for humans over many years. I’m old enough to have known many technological champions and naysayers from the past. They all made pronouncements that have not aged well. If there’s one thing I’ve learned, it’s to wait.

AI will replace typing in your own code. It will not replace software engineers any time soon. At least for serious software. Some one who knows software still needs to instruct the ai what to generate. Still needs make judgements about what is generated, then instruct the AI how to adjust the code. Sure a lot of silly websites can now be generated solely by AI and maybe AI will eventually fix the mess that is know as let’s run.com with requiring a SE to do this, but software is more than silly websites (I’m a SE with tons of experience who is retiring soon and uses AI every day in my job and loves it).

I’ll give an analogy, I write a lot of native code (c/c++, zig,rust,golang,etc). Compilers came along 60 years ago so you never had to write assembler anymore (for most jobs), but even if you are not writing an optimizer for compiler you still really need to know assembly today to write really good native good. AI is going to be like that compiler, you may write less and less actual code, but you will still need to understand it.

Always get a kick out of people standing up for AI, like they’re defending their golf buddy or something.

AI will only get worse. As time progresses, people with actual programming knowledge will become less and less. Eventually, AI companies will be forced to use code generated by AI to train their models. AI code that is already notoriously bloated and buggy. Companies will be forced back in to hiring programmers that won’t exist (have plenty of “prompt engineers” though) to check buggy code produced by AI trained on buggy code. Sounds great. No worries though, economy will be in shambles by that point anyway.

Why do people see this and think, "We can downsize (fire programmers)"? Why are critics seeing this as a cost savings (leading to job losses) instead of a way to keep your staff and exponentially improve productivity and theoretically increase profit?

I know this is just a programming test that AI totally fails, but I think the bigger issue is to step back and ask, "why would we need a program that can recreate TurboTax?" AI just helped me sort out an uncommon and potentially contentious land issue that impacts what I will pay going forward for both my property and income taxes. I threw a mess of a word salad at it, describing various land boundaries, things that happened in the past, and associated issues, and AI understood me and ran with it. It outlined three strategies and the general odds of each of them working, it gave me things to think about that were not on my radar, gave me some advice on what to say/not say to another landowner (and what language to use), it drafted a legal filing formatted and tailored to the laws of my state. And a few other things. Now, my CPA would ask me "how are your taxes coming?" And he's right. But at the end of the day, my taxes are all just information. Numbers and forms and associated written laws. In 5, 10 years I can't imagine that there won't be AI-based solutions that see both the forest and trees in my life and help me bring order to it all. Sure, my taxes will be part of that, but a small part. To the point others have made, it's not just programmers we're talking about. This impacts a staggering amount of jobs.

Not just programmers ...

Books were dead 20 years ago.

Now Spotify is selling physical books. Who knew?