>And outside of industries with government enforced perversions of simple "supply and demand" forces (we're not talking Nth level effects here) this means more access to the service or lower prices or both.
Greater access at the expense of the number of people employed is not necessarily a net positive. In the case of healthcare, the data is fairly clear that higher nurse staffing ratios lead to better patient outcomes (read: decreased chance of death or serious complications). This is particularly true for ED and ICU nurses.
Also, in the case of healthcare, I'm fairly certain that hospitals in the US are legally not allowed to turn patients away. So "lack of access" is not currently a problem for serious medical emergencies.
I found myself interested in the papers analysis of wages, so I did some digging.
The researchers model says the following about hourly wage growth in the substantial and extreme scenarios:
- the hourly wage* in the knowledge workers group will fall (-1.1% and 11.5% respectively)
- but the hourly wage* of labor across the rest of the labor force will grow (5.9% and 33.6% respectively)
*Important to note, AFAIK this hourly wage decline/growth is a not measured against 2026 dollars, but the counterfactual of what hourly wages would be in the absence of AI.
Okay, so my first take away from reading these numbers is that declining wages in one group appears to be offset by rising wages in the other group. Not terrible all things considering.
But then I started to think about the difference in hourly wages between the knowledge workers group and the non-knowledge workers group (which I'll now refer to as *K* *N*). I had a sneaking suspicion that the inter group wages gap could be large enough where growth in the non-knowledge share of labor doesn't offset the loss in hourly wages from displaced knowledge workers who find themselves migrating occupations.
Looking through the paper, it breaks these groups down as follows:
K — Knowledge / Cognitive:
- Management, Business and Financial Operations, Computer and Mathematical, Architecture and Engineering, Life, Physical, and Social Science, Community and Social Service, Legal, Educational Instruction and Library, Arts, Design, Entertainment, Sports, and Media, Healthcare Practitioners and Technical, Sales and Related, Office and Administrative Support
N — Non-Knowledge / All Other:
-Healthcare Support, Protective Service, Food Preparation and Serving, Building and Grounds Cleaning and Maintenance, Personal Care and Service, Farming, Fishing, and Forestry, Construction and Extraction, Installation, Maintenance, and Repair, Production, Transportation and Material Moving
So I had GPT Sol 5.6 take each of these occupations and find the mean hourly wage using BLS data [1]. The findings suggest that as of 2026, the simple group average for K is $45.17 vs. $23.67 for N. But this is short sighted as this doesn't take into consideration the different levels of employment in each occupation. I asked ChatGPT to account for this using the same BLS data, which gives a weighted average of $41.50 for K and $23.40 for N.
This means that knowledge workers are currently making ~double what non-knowledge workers make in the US. The crucial question is how does this change when considering the wage decline/growth between K and N across the papers three scenarios?
Here's the breakdown using the simple mean for hourly wage:
Scenario Knowledge Non-knowledge Gap K -> N penalty
Looking at these tables, both in the current moment and across all three of the papers scenario, there is a large wage gap between K and N. Outside of the extreme scenario, this gap is >40%!
My immediate takeaways having done this bit of research is that if Anthropic is right and we're on track for either the substantial or extreme scenarios, *then many of us in this thread could see a large loss in hourly wages in the short term future.*
NOTE: I'm not an economist. I push pixels for a living. But if you are an economist and you find gaps in the following analysis, please inform me.
I think you'd be better off expanding this into a blog post, where it can have proper formatting and you can take the time to come up with a main point and structure an argument around it.
Graphics programmer who still codes by hand here. I use LLMs for search, and for rubber ducking. I would not disrespect the years I've put into this craft by outsourcing my thinking. Nor do I want to wake up in ten years and realize I have willingly made myself stupid through overuse of these tools.
Sure about that? The effort put into a leetcode medium or hard is usually way above the kind of code i’m mostly doing at work — and realistically i only have at most 2 hours a day of that level of work in me, the rest is the autopilot code, reviews meetings, docs etc.
I’ve found aoc/leetcode pretty much reversed the accidental atrophy for me
Leetcode is about learning an algorithm/data structure. Programming in general is about much more than that. Knowing minimax and ring buffers doesn’t help much in knowing whether what your LLM does makes sense. For that you need to work on more open ended questions.
There is more to software engineering than solving code puzzles. What about debugging, writing clean code, writing effective tests, architecture, optimization, profiling etc.? None of these are exercised by doing leetcode. Use it or lose it.
>It should be noted that these stories were 1000 word stories
A huge limitation of the study IMO.
Most people who read are not reading 1000 word short stories. If they could replicate this story using novella length stories (25-40k words), I'd be curious to see the results. I would wager that a much larger % of people would prefer long form content from humans.
>Why don’t we pay people to dig ditches with spoons? It would produce jobs!
Be glib all you want, but research has shown that displaced workers do not have great outcomes [1].
It's all fun and games to dunk on a person online, but when an entire industry employing millions of people starts to dissolve do to rapid automation, people will have wished there were more calls for organization, social safety nets, etc.
Every single other respectable engineering specialty is licensed, bonded, and insured. Most have local unions as well.
We are not going to stop dealing with the bootcamp script-kiddie middle-manager-with-an-LLM slopcode until we have an efficient way to confer proof-of-skills during the hiring process. It shocks me that ABET discontinued the PE exam subspecialty for software engineering due to lack of interest. Maybe we'll get smart enough to circle back and demand it.
Next is bonds. Large high-dollar government projects that take years to execute because they build critical public infrastructure are often paid in terms of bonds. Wanna know why? Because if the work is shoddy or the contractor tries to skip town, the bond can be rendered worthless. But if the job gets done, and gets done right, an engineering firm can list it on the balance sheet and continue paying salaries even when projects might take years and millions of dollars to come to fruition.
Next is insurance. Hospitals already found this one out the hard way with ransomware. And for those of you who think "it's just software, it's not a big deal"... I hope you never have to bet your life on your code. I wouldn't.
Finally, we have unions. I've noticed that game dev got smart and there was a bit of organization at a recent GDC. The rest of us, I guess, have glided along on the salary bump associated with the difficulty in finding and retaining software engineering "talent". I guess we're finally seeing that disappear now that folks are trying to see how many lessons they can learn the hard way through nontechnical brute force and LLM slop. Yeah, sure, some engineers are using LLMs responsibly in a way that matches their capabilities-- i.e. using it as autocomplete to type code faster. But there's a reason why other easily-digitized knowledge-work fields like law and architecture haven't been subsumed by LLMs. It's because accuracy matters, and because those regulations have already been written in blood.
I mean, I don't think unionizing will make such a difference.
Maybe, going in the sense I mentioned above, the work will be more sprayed around in startups.
Since big companies just want to do what they've been doing but with less people, instead of doing more per person.
This surplus will work with startups and new businesses to disrupt the big ones
But solving hard problems and developing systems intuition was. If you're trying to tell me that people working on code bases predominantly written by AI are still solving hard problems themselves, or understand the systems the AI is building, I think they're lying to themselves.
I work on a codebase with a lot of math and I've both learned and implemented a bunch of things that were beyond my level before I could iterate on it with AI. Same for binary reverse engineering. On the other fixing tedious bugs by hand and typing the code in is not something I miss at all.
I'm also dependent on a bunch of advanced libraries and compilers I didn't write. I could reimplement them if I took years to do it but it would be a total waste of time.
>I'm also dependent on a bunch of advanced libraries and compilers I didn't write.
Two different kinds of dependence here: one is taking advantage of a pre-existing tool to perform some set of tasks. The other is there to outsource your thinking over concepts you are not able to reason about yourself.
I'm also outsourcing computational geometry to CGAL and CPU optimizations to GCC, so what? I can still reason about them.
For example for a problem I had recently Claude pointed out (after some prodding for me) that the determinant of the Jacobian of a function I'm interested in is a polynomial in polar coordinates, so I used a fancy root finding library (which I also didn't write) to solve it efficiently.
I have a blazor app in poduction, Codex made errors which it could not fix and it required me reading documentation. I won't argue that it didn't make my life simpler, tasks which would take me days are done in hours, sometimes without touching code.
Yea, i still deal with a ton of code in AI heavy workflows. If anything the frustrating part is absorbing the code quickly enough. AI (Claude for me) writes in cryptic text and the code flows can often be non-obvious.
I need (and am exploring) custom review tooling to improve this AI->Human code flow. Reviewing PRs were always the hardest part for me in programming. They were often full of the developers decisions and you have to rediscover those as you're reading code for it to make sense[1]. However i find this even more difficult to discover these decisions from AI.
However unlike human PRs we can ask more of AI. Rarely have i had a developer put on a presentation for a PR - but AI could right? AI could produce a guided walkthrough of the code. Not sure if it will help of course, but my thought is we're all stuck in the old "PR review flow" but instead of PRs it's AI - and the volume of them is far greater than anything prior. So i expect we need to tweak how we review, how we get information from LLMs.
[1]: I'm speaking generally, and about larger PRs. Not some small func where you can easily see what it does. Business logic and complex code can be difficult to decipher in PRs, imo.
From what I’ve seen they get to the level of understanding of systems way faster than we did because they iterate and experiment with this from day one. We had to spend years understanding code syntax, and other minutia.
Also, you need a different kind of toolset/skillset to grok a system that was vibe coded all the way - different failure scenarios. Older devs tend to just say “see, I told you, it’s just spaghetti underneath, you need to clean it up”, and the newer gen learns to work with that spaghetti.
I fear the majority of people in this thread who are joking about no longer being able to do their job while Codex/Claude are down aren't really joking.
>The question is whether this is a bad thing or a good thing.
Framing this as a debate worth having gives people who would argue on the side of "it's a good thing" a platform. We should not be giving people who would argue for the death of thinking a platform.
We are overly reliant on technology today. I don't see this trend slowing down in the future.
I think that a world where everyone is reliant on technology, and nobody understands how it works is a nightmare.
reply