3/3
Your point about the bubble economics is spot on. This is super loose but interesting:
I did a little back of the envelope estimate of the real cost of LLM-enabled engineering, removing the subsidy and taking into account the token consumption trends for real programming. (Yes, I used an LLM.)
I think meaningful programming work will converge on 10-15k per month token burn and produce a new elite class of engineer/architects who are effective enough to justify the burn. And that will be another skill added to the stack.
LLMs may shrink the ranks of middle managers, UI devs and corporate digital ditch diggers massively, but my strong intuition is that it will NOT commoditize real software, quite the opposite. That subset who have a true gift for the work, will thrive as much as ever, or more. We will end up with high demand for the top tier who can produce 10x value.
I've actually seen this play out with previous hype cycles and this is always how it has gone. As Cory Doctorow recently put it (and even though I can't stand his progressive values and socialist whining, he is right) the bosses always hate the engineers and the salesmen of the new hype always sell the fantasy of getting rid of them. I've seen it too much to deny it.
In the long ago, I saw chuffed up managers gloating that with UML diagrams and WYSIWYG editors, they wouldn't have any more engineers in just a few years. I've seen Eric Schmidt's abominable talk at Stanford. Same dumb antihuman managerial fantasy. Never gonna happen.
The UI/app/framework dev stuff keeps getting automated. It sprawls out when capital is cheap. The managers get slaughtered when capital tightens. But the core of people who really were built to program computers just get more valuable as computing becomes more important. It's about judgement, creative problem solving, metabolizing complexity and systematizing. And the nameless quality. LLMs don't actually produce intelligence, they just imitate it, and real living intelligence is what programming actually is.
Nobody that loves real programming should be worried about LLMs. They are databases and code generators. Brainrot and slop have always been problems with databases and code generators. We've complained about stack exchange cargo coding copy pasta and IDE boilerplate and framework bloat and JavaScript package sprawl... And rightfully so.
Heck, Socrates made the same observations about the written word itself.
We will get through it with human discipline. It's NOT really the tool itself, even though it feels that way. It's the fact that we are new to it. Human beings magically lose all their common sense and discipline every time the find a shiny new tool. It's reasonable to get disgusted, and I am with you. I can imagine how some Babylonians may have felt about bricks.
Anyway, that's my take. This was a great article. Thanks for writing it.
