Curious how realistic the skills learned in such a program are still applicable in 1-2 years. This industry is moving very fast, it feels like the self motivated, self starter, self learner has been thriving. Probably a good thing to have such a program, but I have doubts.
The YC meta model is probably a better school than school now. I'm sure YC knows that.
If you need to be told what to do, it's a short step to your AI replacement being told what to do; money will ensure it. If your job is to discover and capture new things in an exploding possibility space, that's a career.
Schools should teach hacking, but it doesn't seem that's what this program is doing.
> "With so many companies worldwide faced with upheaval due to AI in software development, I believe this program will become a template for rapidly leveling up new developers across the industry."
This seems cognitively dissonant with the AI upheaval resulting in the industry needing fewer developers. Maybe they want cheaper developers?
If we freeze AI models at their current stage of development, then yes. If not, given the rate at which models are improving (compare Claude today to Claude 1 year ago), in 3 to 5 years software development could be fully automated.
Let's see, my grandparents were about 160cm when they were 35. My parents were about 170cm when they were 35 and I am about 180cm around 35. You are saying I can expect my grandchildren to be around 200cm at that age and their grandchildren 220cm?
Or we could assume since new data is increasingly AI generated then the next generation of models are inherently born out of incestuous relationships so I would imagine Claude's grandson is going to be a dwarf.
Of course it is possible also that Claude will marry a chinese model (since it's asian odds are it is also smarter) bringing some diversity into the "gene" pool. Or better yet, a Russian supermodel.
Are you sure that AI model progression and the height of your family members are 35 is related? It seems unlikely but I guess you'd know better than I.
"We're running out of data, so LLM progress is hitting a wall" was a prominent late-2024 narrative. Back then many experts including Ilya used the word "plateau" when describing LLM progress. No one is using that word now, and there is no evidence we are about to reach a plateau, even though we ran out of data long ago.
If we didn't have any humans before your grandparents, and these were the only data points we have to extrapolate from, then, yes, this would be a very reasonable prediction.
Your comment completely misses the point. AI is fully capable of replacing shitty developers that cannot program and cannot write original code of their own. This is not a technology problem, missing skills problem, or even an AI problem.
It’s a problem of missing training and missing mentorship coupled with weak leaders who would rather throw their people to the wolves armed with crayons and coloring books until they just figure it out.
Before AI was a thing once a large organization hired too many shitty developers you could occasionally clean house by laying some off and ordering the rest to own more responsibilities. Once AI starts replacing the weakest among us the scapegoats are gone. Then what? Eventually the finger pointers can only point at themselves once there is nobody else to point at.
Is there evidence that overall software developer positions are declining relative to other jobs? It wouldn't surprise me if Jevons paradox applied here.
Im sorry for being so negative but MS sold software engineers (and US high school and college students) down the river so long ago, the nerve of them now even pretending to care is just so offensive. This business has gotten so rotten, the best thing I can tell anyone is to find something rewarding to do with your life.
"software engineer" is likely a career long gone. "Product engineer" and architect is probably the only software-adjacent title that will have consistent demand for awhile.
Basically still just a prompt jockey but knows the difference between outputs that are good and bad.
I was a huge AI skeptic from the beginning through ~2025. But once Opus came out, it's pretty obvious writing code by hand is an obsolete skill on its own. You still need to know how to read code, how to know what good/bad code looks like, and if it matches what you're truly asking for. But there's really no reason to be typing code with your own fingers in 2026 and beyond.
2024 me would be shocked at current me saying this.
> writing code by hand is an obsolete skill on its own. You still need to know how to read code, how to know what good/bad code looks like,
It's my impression that the best way to be able to read code and distinguish good from bad is to write code. Get burned a few times and you know why certain patterns are a bad idea. You don't internalize that equally well from a powerpoint slide.
Compare this to math. Every successful mathematician has gone through the motions of proving certain basic theorems in math. Like Euclid's proof that there are infinitely many prime numbers. And then they went through related exercises. This was all learned with sweat. The statement above is akin to saying there is no point in doing any proofs by hand anymore, just read one and you will have sufficient understanding of mathematics and become a good mathematician. I don't think that's true. (And I also believe that many sw eng people would be better at their job had they groked the proof of the halting problem.)
Yep. If you think your job involves "writing code" as the core part of your responsibilities, write yourself off immediately. That career is dead and dusted. Sadly most companies still haven't realized this, given how few Architect openings exist (and those that do, most are asking for 20+ years of exp lol)
In my professional experience most people I encountered that held the architect title didnt really know anything about proper software engineering and just knew how to talk well and hold meetings with many different stakeholders - which tbf is not a trivial skill, but I can’t imagine everyone becoming architects and no one doing actual work they understand.
Definition of what the role actually means can change as well over time, in the same way that a civil engineer also isn’t physically doing the construction.
Most companies aren't gonna need an architect, but if you can tell and demonstrate a good story around quality/risk/stability they might be interested in that.
The assumption here is that senior engineers are still needed to review AI-generated code. That assumption might not be valid for very long. I find myself reviewing AI-generated code less and less because the code quality is very good -- and getting better every month. Soon reviewing AI-generated code will be like manually verifying the results of a calculator--it is a waste of time. At that point all you need is an architect. I predict senior software engineers will be obsolete within a year.
As a principal, I still catch people submitting shitty AI code on a weekly basis. Their brains are rotting to the point that they can't tell if the code is doing the correct thing or not.
Just left my programming job due to this. Stopped thinking, stopped caring. AI destroyed my coding joy, so I’m finished with that. Leadership challenges next - hope to use my brain a bit more.
I've found myself engaging more with AI writing code. There is no easy time during the work day. I'm talking to product about what they want, communicating with AI to write it, and verifying that it covers what was asked for. And the better I get at it, the more of the product I need to learn at a faster pace. If I find some downtime, I work on additional features at the same time, something I wouldn't do on my own.
Quite simply, I'm more mentally drained at the end of a day working with AI than I am a normal day of coding. There is still a lot of energy and thought that needs to go into what needs to be written.
Anecdotally, I find the flow-vs-drained state often depends on whether I know the provenance and trustability of what's in front of me.
1. [Good] I know what I wanted, I have a concept of what "correct" looks like. I ask the LLM to do the thing, and I check if it's the thing I intended. Even if it's annoyingly wrong, I can fix it up.
2. [Bad] Someone else had a concept once. They put it through an LLM to generate something that reaches me, a ticket, PR, or review comment. However now I have to constantly guess which parts of the thing are "real" and which are nonsense the human never actually intended or even noticed.
The latter is worsened by how reading means inferring/simulate the mind of the writer, but LLM output is not from a mind we typically simulate. Subtle persistent wrongness is exhausting.
The issue is the higher ups don't even seem to care. You just turn into the guy that's yelling at the clouds. I don't know what the future of this industry is.
As of now it is insufficient to outline the end-user requirements, AI will ask you a bunch of questions that requires you to make architectural decisions. I think within 3 to 5 years that might change as well.
If you say "yeah go with your own recommendation" every time you'll get something that'll at least kinda work sooner or later. Not sure you need an "architect."
But is that workflow enough in a competitive world anyway, or will there be an edge for people who do more than "outline the end-user requirements" like it's the waterfall 70s?
If you need to be told what to do, it's a short step to your AI replacement being told what to do; money will ensure it. If your job is to discover and capture new things in an exploding possibility space, that's a career.
Schools should teach hacking, but it doesn't seem that's what this program is doing.
This seems cognitively dissonant with the AI upheaval resulting in the industry needing fewer developers. Maybe they want cheaper developers?
If we freeze AI models at their current stage of development, then yes. If not, given the rate at which models are improving (compare Claude today to Claude 1 year ago), in 3 to 5 years software development could be fully automated.
Let's see, my grandparents were about 160cm when they were 35. My parents were about 170cm when they were 35 and I am about 180cm around 35. You are saying I can expect my grandchildren to be around 200cm at that age and their grandchildren 220cm?
Of course it is possible also that Claude will marry a chinese model (since it's asian odds are it is also smarter) bringing some diversity into the "gene" pool. Or better yet, a Russian supermodel.
It’s a problem of missing training and missing mentorship coupled with weak leaders who would rather throw their people to the wolves armed with crayons and coloring books until they just figure it out.
Before AI was a thing once a large organization hired too many shitty developers you could occasionally clean house by laying some off and ordering the rest to own more responsibilities. Once AI starts replacing the weakest among us the scapegoats are gone. Then what? Eventually the finger pointers can only point at themselves once there is nobody else to point at.
... in Github Actions? What a great way to discourage them from entering the industry.
Basically still just a prompt jockey but knows the difference between outputs that are good and bad.
2024 me would be shocked at current me saying this.
And how you would know to do that, if you're not writing code?
It's my impression that the best way to be able to read code and distinguish good from bad is to write code. Get burned a few times and you know why certain patterns are a bad idea. You don't internalize that equally well from a powerpoint slide.
Compare this to math. Every successful mathematician has gone through the motions of proving certain basic theorems in math. Like Euclid's proof that there are infinitely many prime numbers. And then they went through related exercises. This was all learned with sweat. The statement above is akin to saying there is no point in doing any proofs by hand anymore, just read one and you will have sufficient understanding of mathematics and become a good mathematician. I don't think that's true. (And I also believe that many sw eng people would be better at their job had they groked the proof of the halting problem.)
Quite simply, I'm more mentally drained at the end of a day working with AI than I am a normal day of coding. There is still a lot of energy and thought that needs to go into what needs to be written.
Anecdotally, I find the flow-vs-drained state often depends on whether I know the provenance and trustability of what's in front of me.
1. [Good] I know what I wanted, I have a concept of what "correct" looks like. I ask the LLM to do the thing, and I check if it's the thing I intended. Even if it's annoyingly wrong, I can fix it up.
2. [Bad] Someone else had a concept once. They put it through an LLM to generate something that reaches me, a ticket, PR, or review comment. However now I have to constantly guess which parts of the thing are "real" and which are nonsense the human never actually intended or even noticed.
The latter is worsened by how reading means inferring/simulate the mind of the writer, but LLM output is not from a mind we typically simulate. Subtle persistent wrongness is exhausting.
What value would an architect provide that a SOTA LLM can't?
But is that workflow enough in a competitive world anyway, or will there be an edge for people who do more than "outline the end-user requirements" like it's the waterfall 70s?