Agents are eating the software development lifecycle from the bottom up — code generation already gone, review and ops following fast. What remains is figuring out what people actually want, which can't be automated or scaled. That job is called product engineering, it pays $240k, and nobody's training for it.
Dude seems to be high on AI propaganda and is completely ignoring that running any model has some cost. Sure, right now many companies are burning through tokens like wildfire, but that won’t last forever - it possibly won’t even last thru 2027 - so calling any part using AI “free” is just dumb, even for people who run models locally.
He mentions how frontier models, the most expensive ones, are doing the best job. Guess what, those are the ones that aren’t on huggingface and joe schmoe isn’t going to pay up for.
I think the author gestures at a pretty solid argument that most of the “mechanical” arguments for LLM capabilities have a clear trend line towards “can do thing”.
Strictly speaking, the job he’s describing even falls into that category, there’s just not a lot of training data for it.
Anecdotally this has matched what I’ve seen. Which is unfortunate because the part of tech I like the least, regarding software development, is when the thing that was built was actually used for anything.
I got into this specifically, and exclusively, for the craft. I don’t like making other’s lives " better", I don’t like “solving problems”, I don’t like the customer. But here we are staring down the barrel of a career making “products”.
This is a really great point. AI tokens are expensive for quality today, and that will continue. We’re already seeing a bit of a plateau, and this whole “let’s slow down so we don’t kill ourselves” is propaganda, I think. AI companies that leverage “price to value” are going to succeed. I see Google Gemini fitting this mold long term, GLM, Kimi did fit this mold but their later models seem to jump in price pretty fast chasing Opus capability. If you take “capable models” as the standard for price to value and then charge what it actually costs (not VC subsidized), then you’re going to see a world where Software Engineers are probably back in high demand. I think the article points out the hard part: how do you grow new Sr. Engineers if you’re replacing Jr’s with Claude. (I think the obvious answer is that we shouldn’t be replacing Jr’s with Claude)
Dude seems to be high on AI propaganda and is completely ignoring that running any model has some cost. Sure, right now many companies are burning through tokens like wildfire, but that won’t last forever - it possibly won’t even last thru 2027 - so calling any part using AI “free” is just dumb, even for people who run models locally.
He mentions how frontier models, the most expensive ones, are doing the best job. Guess what, those are the ones that aren’t on huggingface and joe schmoe isn’t going to pay up for.
I think the author gestures at a pretty solid argument that most of the “mechanical” arguments for LLM capabilities have a clear trend line towards “can do thing”.
Strictly speaking, the job he’s describing even falls into that category, there’s just not a lot of training data for it.
Anecdotally this has matched what I’ve seen. Which is unfortunate because the part of tech I like the least, regarding software development, is when the thing that was built was actually used for anything.
I got into this specifically, and exclusively, for the craft. I don’t like making other’s lives " better", I don’t like “solving problems”, I don’t like the customer. But here we are staring down the barrel of a career making “products”.
This is a really great point. AI tokens are expensive for quality today, and that will continue. We’re already seeing a bit of a plateau, and this whole “let’s slow down so we don’t kill ourselves” is propaganda, I think. AI companies that leverage “price to value” are going to succeed. I see Google Gemini fitting this mold long term, GLM, Kimi did fit this mold but their later models seem to jump in price pretty fast chasing Opus capability. If you take “capable models” as the standard for price to value and then charge what it actually costs (not VC subsidized), then you’re going to see a world where Software Engineers are probably back in high demand. I think the article points out the hard part: how do you grow new Sr. Engineers if you’re replacing Jr’s with Claude. (I think the obvious answer is that we shouldn’t be replacing Jr’s with Claude)