ej88

It's not about whether they make mistakes (they do! although the exact definition of a mistake is nuanced), but whether they can take accountability if the software fails and millions are lost or people die. A large part of the premium paid on software engineers is to take accountability for their work. If a "business person" directs their agent to build some software and takes accountability -- congrats! They are also now a software engineer :)

The lines between a software engineer / business person / product / design and everything else will blur, because AI increases the individual person's leverage. I posit that there will be more 'software engineers' in this new world, but also more product people, more business people, more companies in general.

I am personally of the opinion that ML will end up being 'normal technology', albeit incredibly transformative.

I think you can combine 'Incanters' and 'Process Engineers' into one - 'Users'. Jobs that encompass a role that requires accountability will be directing, providing context, and verifying the output of agents, almost like how millions of workers know basic computer skills and Microsoft Office.

In my opinion, how at-risk a job is in the LLM era comes down to:

1: How easy is it to construct RL loops to hillclimb on performance?

2: How easy is it to construct a LLM harness to perform the tasks?

3: How much of the job is a structured set of tasks vs. taking accountability? What's the consequence of a mistake? How much of it comes down to human relationships?

Hence why I've been quite bullish on software engineering (but not coding). You can easy set up 1) and 2) on contrived or sandboxed coding tasks but then 3) expands and dominates the rest of the role.

On Model Trainers -- I'm not so convinced that RLHF puts the professional experts out of work, for a few reasons. Firstly, nearly all human data companies produce data that is somewhat contrived, by definition of having people grade outputs on a contracting platform; plus there's a seemingly unlimited bound on how much data we can harvest in the world. Secondly, as I mentioned before, the bottleneck is both accountability and the ability for the model to find fresh context without error.

They do, in the paper they mention they evaluate the LLM without tools

ai skeptic fanfic evolves in fascinating ways every day

huh? this is completely inaccurate

Im starting to believe that the biggest moats will be in the application layer, and that people are starting to realize traditional saas moats apply to those companies too

network effects, distribution, proprietary data, systems of record

companies like opencode have none of the above

cursor's distribution has been faltering and they're hard pivoting to training their own models with their proprietary data to try to build their moat back

yes

am close with a few employees there

1. on prem 2. extremely strict data controls, if one of palantirs big customers found out data got leaked people are going to prison

assuming no asi, the market value of thinking without accountability trends to zero, the bottleneck will be thinking + accountability, at least for knowledge work

if ai truly solves novel thinking then nothing is a barrier. the physical world is downstream from robotics which is downstream from software. itll be able to persuade nation states to collect data for itself etc etc (insert sci fi ending)

there doesnt seem to be a limit in terms of the ceiling of what companies can do with software, probably the most elastic demand out of any industry ever

the swe role is going to change but problem solving systems thinkers with initiative won't go away

I feel like this article doesn't really contribute much to the discourse and is somewhat spoiled by the author's biases.

I think the point about lacking precise language to describe LLMs is reasonable, then the author follows it up with claims that the machines can't count and that they are incapable of math (easily disproven). Then says "talking rock" is a better alternative, which to the average person would be even more confusing. Then says AI researchers tend to not consider LLMs AI (like.. what?)

The point on Turing's Imitation Game was reasonable too, then confidently proclaims LLMs are not doing anything intelligent and are pure mimicry. Intelligence is notoriously poorly defined, and the stochastic parrot meme has already died now that RL enables out of distribution behavior.

The chat point and talking dog syndrome are both reasonable and I generally agree with them.

the role of a 'developer' is ever-shifting, even now a lot of swe roles expect a good amount of product or customer-facing skills

its the opposite, majority of users prefer to get support via chat or phone

navigating ux is still difficult in 2026

the average hn user is leagues above what the average customer or even smb knows about tech and ux, just not realistic for them to redesign their apis

most support volume comes through voice, and you need a layer to interpret what the customer intent is

additionally for many use cases it's not feasible from an eng standpoint to expose a separate api for each entire workflow, instead they typically have many smaller composable steps that need to be strung together in a certain order depending on the situation

its well fit for an llm + tools

the majority of everyday customers have never heard of an API and prefer to call in via phone

in that medium, llms are so much better than old phonetrees and waiting on hold

either way, it's clear they didn't build enough

Accounting for inflation, rents barely moved from 2022-2024 in D.C.: https://ipropertymanagement.com/research/average-rent-by-yea...

I suspect most of the rent increase is due to inflation (sticker price). Nonetheless rents are still rising, which means 60k units were not enough to outweigh demand.

This is exactly why customer service is ripe to be decimated by AI agents that can actually interact with systems

ed finally coming around to the coding use case, and even then he believes they're mostly used for summarizing and searching

we are so early

Thanks Max! This was a really interesting article and closely matches my own experience with how the agents have been progressing

one of the takeaways I get when reading skilled engineers' experiences with these tools is that they essentially offer leverage, and the more skill someone already has the higher their ceiling will be

The only reason why have an ad budget is because buying ads was effective. If they were no longer effective there would be no ad budget

i guess it depends on what you define as bad and what that threshold is

https://trueup.io/job-trend

this tracker shows continuous improvement since 2023

is there any data about the overall job market on whether it's been good or bad? genuinely curious the most recent data point shows a rebound https://www.citadelsecurities.com/news-and-insights/2026-glo...

and fwiw i dont know any swes struggling to find work personally

swe is so broad and in bubbles its hard to get an objective analysis

i feel similarly. suppose ai makes people more productive:

1. companies that are not doing well (slow growth, losing to competition etc) or are in a monopoly and are under pressure to save in the short term are going to use the added productivity to reduce their opex

2. companies that are doing well (growth, in competitive markets) will get even more work done and can't hire enough people

my hunch is block is not doing as well as they seem to be

obviously he's going to posture his company as growing and doing well, but clearly not enough for the board and shareholders given their headcount growth from zirp

some companies are in the position to go for moonshots and block hasn't panned out

trimodal swe compensation (elite, big tech, everyone else) extends to the job markets too

> write an article dismissing ai

> usage is copy pasting code back and forth with gemini

the jokes write themselves

"CEO Dario Amodei predicted last March that in six months AI would be writing 90% of code, and when that didn’t happen"

I mean, a lot of developers have 90% of their code being written by AI (myself and my friends at the labs included). Obviously YMMV depending on your codebase and individual skill.

"Software engineers will at times overestimate their capabilities, as demonstrated by the METR study that found that developers believed they were 24% faster when using LLMs, when in fact coding models made them 19% slower. This, naturally, makes them quite defensive of the products they use, and whether or not they’re actually seeing improvements."

I wonder what he thinks about the new METR update that showed a net speedup as a lower bound (due to participants literally not wanting to even tackle tasks with AI due to how slow it would be), with the returning devs having the greatest improvements in speedup?

"for one of Anthropic’s greatest lies: that AI can “work uninterrupted” for periods of time, leaving the reader or listener to fill in the (unsaid) gap of “...and actually create useful stuff.”"

We're probably at the beginning of the S curve for long-running tasks that create useful stuff (https://ladybird.org/posts/adopting-rust/) but it clearly needs hand-holding and a way to self-verify work.

"No amount of DarioMath about how a model “costs this much and makes this much revenue” changes the fact that profitability is when a company makes more money than it spends."

Feels like he's being dishonest here because the economics of the labs are unique (and precarious). Each model (revenue - cost to train and serve) is profitable. Labs invest in the next model to maintain their advantage, otherwise people will stop using their latest models. This probably doesn't go on in perpetuity (which is what Ed should've analyzed more). To his benefit, he's right that CC subscriptions are currently being subsidized.

[Insert quotes of Dario saying models will be smarter than most humans or Nobel laureates]

I mean, he's not wrong in certain definitions of "smart". They're already well above the average human in terms of testable world knowledge, math, coding, science, etc... but obviously fall short in other ways compared to humans.

not enough people look at the slope, just the coords

Really interesting updates to their 2025 experiment.

Repeat devs from the original experiment went from 0-40% slowdown to now -10-40% speedup - and METR estimates this as a 'lower-bound'

more devs saying they dont even want to do 50% of their work without AI, even for 50/hr

30-50% of devs decided not to submit certain tasks without AI, missing the tasks with the highest uplift

it also seems like there is a skill gap - repeat devs from the first study are more productive with ai tools than newly recruited ones with variable experience

overall it seems like the high preference for devs to use AI is actually hurting METR's ability to judge their speedup, due to a refusal to do tasks without it. imo this is indirectly quite supportive for ai coding's productivity claims.