To avoid being replaced by LLMs, do what they can't
seangoedecke.com
seangoedecke.com
Tell that to all the customer service AI chatbots. They don't need as many real CS agents anymore because one chatbot can reply to hundreds or thousands of tickets with useless garbage copy-pasted from the FAQ. Given that even most real CS agents also only do that nowadays anyway, yeah they've been replaced.
This tends to be incredibly effective at CS's real job, which is to make customers with problems go away from being able to make it to bothering the actual consumer-unfriendly-decision-makers. If you start looking for this you will see it absolutely everywhere. Chatbots are really really good at making people too frustrated and exhausted to pursue other routes.
There are a few companies whose CS actually still is good (like Mila Cares!!!) but in my experience most CS just wants to get rid of the annoying customer.
Also, sort of related (to me, blame my autism), I've heard that the practical effect of a suicide hotline can be annoying the caller enough for them not to do the thing. I've had at least one or two friends told me that this was the effect it had on them. (they're doing better nowadays)
Most anti-AI arguments can be dispensed with by recasting them in terms of the broken-window fallacy. This is certainly one of them.
Could you explain how and what fits the broken-window fallacy here? Is it the AI abolitionists? (I know at least one person who violently hates AI and anything relating to AI and firmly believes that there is absolutely no place for AI or LLMs anywhere in the world and that any usage at all for any reason whatsoever is utterly deplorable. I don't talk to this person since they tend to flame at others for using or talking about AI for any reason.)
That's where the broken-window fallacy comes in. It's not just a matter of paying someone to break all the glass in town to make work for the glaziers, it's like responding to the invention of cheap shatterproof glass with the same flawed zero-sum reasoning.
(I'm rate-limited so can't reply, but my position is basically that 100% unemployment is a good thing, not a bad one. We can't get to a post-scarcity society by doing things the same way we've been doing them, or by retrying the same alternatives that have failed before.)
I keep seeing cashiers replaced by self-checkout machines at stores, fast food jobs might be too complicated for certain people (like me - I couldn't do it), places like say the Apple Genius Bar require you to know what you're doing, etc. Maybe I'm super naive and first-world by missing something super obvious, but if I am then maybe I could learn something today.
Can you answer my question about an "AI native" company that never even had people to replace?
(You're Alice btw) :
Alice's response has a few issues:
Shifts the Goalpost:
Bob's point was about AI's impact on existing jobs—he's arguing that AI increases productivity, not that companies like OpenAI don't need people. Alice shifts the discussion to whether OpenAI, an AI company, is hiring people, which isn't directly relevant to Bob's claim.
Strawman Argument:
She implies that if AI is truly productive, OpenAI shouldn't need to hire. But OpenAI is in a high-growth phase, likely hiring to build and maintain the technology, not to replace traditional roles with AI. AI productivity doesn't mean zero hiring—it often means hiring in different areas or scaling faster.
Burden of Proof Misstep:
Alice demands proof from Bob while making her own claim ("baseless assertion") without providing evidence that AI is directly causing 1:1 replacements.
Weak "AI Native" Point:
The question about "AI native" companies might sound clever, but it doesn't address Bob's core argument about existing companies using AI to do more with fewer people.
Most software developers don't work for Google. They don't need a degree. They are making basic crud applications or some other apps that are chaining APIs or libraries together.
The difference between a junior and a senior in these kind of jobs is productivity and code quality/ maintainability. If you don't think current gen LLMs help with that you are delusional
But ya, I see the future for SWEs as not only being able code, but being able to review code created by AI, and being able to write prompts to get AI to generate code.
Otherwise we might as well predict that when all is said and done, AI will have no more impact than the fax machine, which is what someone once thought of the internet, and which seemed a lot more reasonable on dial-up AOL than in today's world.
https://x.com/benln/status/1889388151770325427?s=46&t=tl870z...
But I don’t think the advent of LLM is responsible for any job cut in these companies. The overall startup ecosystem is (Stripe, cloud computing, efficient customer support software, etc).
But the fact that Cursor is one of those impressive companies shows the demand for AI assistance is huge
In other places, look for hiring freezes and re-orgs, rather than straight and obvious message from management: "Say bye to Bob, he's been replaced with this AI agent".
To everybody's point, not a new thing - automation, including what may of us have worked on for decades, by definition is simplifying, making more efficient, or reducing human effort.
But to your specific question - HN seems to have a couple of excited stories a week about a person saying "I don't really know coding, but I've started my tiny niche SaaS with help of AI". Which is not necessarily a bad thing, and more along the curve of all the other advancements - perhaps the question is the speed of improvement(replacement) and change on this particular one.
How are you going to debug/extend the code if you don't know how it works?
The phrase "don't really know" is kind of doing the heavy lifting here - for example, I myself haven't done professional programming in 20 years, but I know the basics. Enough to be structured in my design / requirements / direction to LLM, and to read and comprehend the advice others have discovered and shared. With that in mind, extending the code seems to work fairly well - start with basic requirements and extend them through LLM. Having not done coding in 2 decades, I've asked ChatGPT to create a python script to interact with a website's API and retrieve an object; then I've extended it to retrieve and store multiple objects, then to input for parameters, etc. it was fun!
Anyhoo, the use case discussed is not necessarily a complex enterprise resourcing application; but a little niche webapp with simple requirements, created by somebody who understands the domain itself more than coding, and which may be exciting to a target requirement, seems eminently doable with today's LLMs -- again, the proof is in the pudding of people who have done it, rather than my claims :)
Anyone skilled enough to code can now codes in any language. Jobs aren't being replaced but no longer added.
I also demonstrated how an intern (1st year) was able to replace a 5 year experience position after six weeks of working with AI.
We're seeing a 5x acceleration in shipping after our first year with limited exposure.
Don't you think that it might be more efficient to just communicate with the bullet points?
The operations crew is having it automate metrics rapidly. Senior developers are increasing their throughout rates. Things come back to code review in a generally better state because folk say "hey, review this for me".
Is it a real developer? Not at all. Is it affecting our hiring or anything? Nah. But it's already a huge help and rapidly getting better with new models (o3 / deepseek), tooling (CSP and agents), and integration (cline, cursor, memory prompting).
Idea to bank to pull funds to 3d print to market validation in a day.
Example from last week: pasted my terraform that was using internal custom modules and prompt that based on this code build additional services not yet built using the module output so they could all communicate.
First run. 100% accurate. Zero typing after paste and prompt. 2 minutes of time.
It probably saved me an hour at least of looking up module output and documentation with the additional services.
It may be minor or simple yet I now have an additional hour.
While every single developer now uses LLM, the outcome really depends on how one is using it.
That was always true. Once you learn how to program, it doesn't take a ton of effort to learn another language. It takes more time and effort to master a new language, but that's still just as true as it ever was.
What the competent people are saying is often overshadowed by the echo chamber of misleading information, since there is profit in driving people towards chaos.
I barely get 1 line of code that I’d actually ship unchanged. It’s really useful but still has a long way to go
Just like you don't need to be able to outrun the bear, it's enough to outrun your friend, it's okay to be theoretically replaceable by AI, as long as I'm not the most obvious person to be replaced (at least that's what I tell myself).
Companies move slowly, I just hope they move slowly enough for me to provide a good life for my family as a software developer.
This is a motivating and least depressive outlook on the future, as it encourages me to learn things better, and that feels good for me.
I have many thought on AI, sometimes excited, sometimes frustrated, sometimes worried, but I didn't see this idea phrased like this before, so thought Id share it.
I got shaken out quickly (but am planning an epic comeback). The math doesn’t work out favorably so being ambitious is probably a good idea. But if things get too advanced then it also gets easier to automate an entire company, and enough automation means more viability for solo ventures. That ambition comes in handy there as well
It will probably not ask you any questions, but if it does and you answer it will not ask follow up questions, or if it does it will lose track of your answers or non-answers. It does not maintain situational awareness. It does not speculate on your state of mind or competence as you help it.
You can avoid your job being eaten by AI by moving toward roles where you talk to people, understand their problems, and perform the work of translating that into solutions.
There will also always be an orchestration role no matter how much automation is thrown at a problem. - Who debugs that AI code? - OK, say it’s an AI. Now who fixes the auto-debugger when it breaks? - Who makes decisions about rebuilds and migrations and big platform shifts? - Sure maybe that decision is informed by advice from AIs but someone with accountability has to make the call before the wheels are put in motion for the rebuild or migration or whatever.
Both are durable targets for your career.
VCs would probably like that as well.
Maybe pitch that to CEOs as a cost saving meausre.
I'm a theory and simulation guy, but in retrospect I should have done far more experiments when I was in training. I guess it's never too late to start...
Perhaps our future overlord will thank you :)
* Problems are ill-defined and poorly-scoped
* Solutions are difficult to verify
* The total volume of code involved is massive
In my view, this is describing legacy code: feature work in large established codebases."
If you have used cursor.ai to try to create a moderately sized project you'll see this happen even with newly generated code.
In my experience, if you limit yourself to generate not well thought through prompts and do not work on getting a deep understanding of the generated codebase, the LLM will start duplicating the same code flows in different ways, many time forgetting some of the behaviour already implemented.
Kind of like having dozens of developers working on the same codebase clueless about what each other has done and re-implementing the same functionality until the code turns into a pile of spaghetti code.
It can be done but:
* You must have a deep understanding of the code
* You need to think hard about what you are doing and give very detailed instructions to the AI
It works for trying a quick prototype but when moving on to production grade code you need to slow down and "program" step by step providing precise instructions as you go.
You'll have to design the changes to the minor detail and then you can let the AI do the grunt work.
It's like programming without coding.
Most of us are already accountable for outcomes, not outputs. Perhaps we could go further in that direction--but doing so only makes personal sense if the underlying work that you're doing is important to you. If AI is about to make us all 10x coders, why should we keep the jobs we have when we could take that extra capability and go do something more meaningful--the kind of something that used to require a 10-person company.
I'm personally pretty happy with my company, but my point is that once everybody gets more productive, what's the likelihood that everybody who still has a job after the transition still wants that job now that doors which were previously closed are now open?
It's gonna be a bigger reshuffle than just taking more ownership over our existing domains.
This truly is the challenge - both to have the huge context window and the ability to conduct coherent and comprehensive reasoning using the entire context. We should see soon whether there is a Moore's law effect here: I would be immensely surprised if not.
Upon reading this, it seems like the author is in the latter group, and while he offers a few points about what computers can and can't do, the advice given is horrible advice because it takes things in isolation and overgeneralizes, while not paying attention to underlying factors.
The "lets just tough it out" approach and specialize in old code, or learning to do what AI can't are impossible tasks in practice.
If the author is in the latter group, I think he's unintentionally doing himself a disservice by showing a low level of competency in addressing the problems.
You don't want to hire an engineer who is blind to the potential liabilities they create.
Any engineers in IT are intimately familiar with the fallout from failures involving sequential steps in a pipeline.
There's front-of-line blocking (FOLB), and there's single points of failure (SPOFs), these are considered in resiliency design or documentation of the failure domains. The most important parts of which are used in identifying liabilities upfront before they happen.
Entry level task positions are easily automated by AI. So companies replace the workers, with AI.
How do you get to be a mid-level engineer when the entry level no longer exists...its all based upon years of experience. Experience which can no longer be gotten.
Does this sound like a pipeline yet?
You still have mid-level engineers available, as you do senior engineers, but no new ones are entering the marketplace. Aging removes these people over time, and as that sieves towards 0 the cost of hiring these people goes up until it reaches infinite (where no one can be hired).
What goes into the pipeline is typically the same but most often less than what comes out of said pipeline. In talent development its a sieve separating the wheat from the chaff.
Only the entry point is clogged, and nothing new is going in, humans deal with future expectations and the volume going into such pipelines is adaptive. No future, no one goes into such professions.
After a certain point, you can't find talent. There's no economic incentive because companies made it this way by collusion.
Things stop getting done which forces collapse of the company. Its not just one company because this is a broad problem, so this happens across the board creating a inflationary cycle of cost, followed by a correlated deflationary cycle in talent, that cannot be fixed except by the industry as a whole removing the blockage. They can't do that though because of short-term competition.
When have industry business-people today turned on a dime in economically challenging situations where the money wasn't available; ever.
Debt financing makes it so these people don't need to examine these trends more than a year out, but the consequences of these trends can occur just outside that horizon, and once integrated the bridges have been burnt and there is no going back while also maintaining marketshare.
All of the incentives force business people to drive everything into the ground in these type of cycles. The only solution, is to know ahead of time, and not bait the hook. The business people of today have shown that this is beyond them, its all about short-term profits at the limits of growth, business as usual.
Real world consequences of such, you can look to Thomas Malthus, and Catton who revisits Malthus.
Catton importantly shows how extraction of non-renewables can reduce or destroy previous existing renewable flows leading to lower population limits as a whole than prior to before prior to overshoot.
Similar behavior applies broadly to destructive phase changes of super critical systems with complex feedback mechanisms (i.e. negative flips to positive and runs away, or vice versa leading to collapse/halt). In other words where you have two narrow boundaries outside which the systems fail.
This is one of the concerns I hear. Not really in a position to judge how serious it is but I've had this discussion with people in senior roles related to, let's call it developer mentoring/development.
To the degree LLMs make junior developer roles commodities and therefore less attractive financially that definitely makes bringing new people on-board at a lot of companies less attractive.
Essentially no one thinks an LLM is going to step into the role of an experienced senior developer as anything other than a possibly useful assistant. Someone just out of school? Maybe you don't replace the best but maybe you need a lot fewer of them and pay them a lot less.
Historically, we see Information Technology advances first disrupt IT, then it spreads with adoption everywhere else to realize the same cost savings, as a labor multiplier.
> Maybe you don't replace the best.
In fairness, there's no real way to tell who the best/competent are. University programs have always failed to prepare the student in IT because of the fast moving nature of it.
Given the lack of any way to properly distinguish oneself, the best and most competent will look for a time, but eventually they have to go where the money is. That means retraining and taking the loss in time and investment, and its a sticky decision where they aren't likely to fight for such meager scraps.
Competent people have options others don't.
A wage price floor was hit awhile back in the long trend towards wage suppression. You can't really pay them less when other opportunities with no education provide more economic incentive. This is an example of chaotic distortions involved in money printing generated whipsaws.
The opportunity cost ratio between unskilled and skilled labor eliminates any incentive. Why spend 10 years on education and experience when you'll only make at best 33% more than someone who doesn't (pre-tax). Less than that post-tax, and we aren't considering the increased costs that are borne by individuals seeking out positions. The job market has always imposed cost in time, from interview projects (where they steal your work without compensation), to circuitous interviews, etc.
And maybe the others should reconsider whether IT/tech is the automatic meal ticket they thought it was. Not sure that's the worst outcome. Honestly, there are probably still a lot of jobs floating around--just a lot fewer at the highest paying and sexiest companies--and some of the jobs may be in trades and other professions.
When merit is no longer an important metric, the competent leave first because they are no longer rewarded for their productive capacity, or effort. The people who thought t was an automatic meal ticket burn the house down for everyone.
The current offers going out for IT Architect work, decade+ direct experience is 40k/yr, no equity. No conversion rate on applications that isn't 1000:1 phone call.
For 40k/yr, you can go and flip burgers and not have to deal with the high stress involved in these type of positions. It would be a joke, if only it were not serious, and things will only get worse.
The creeping ruin has a way of coming at you sideways without you knowing until its too late.
A good way to tell which category a blog falls into is to look at the first link provided on their homepage.
There are many ways in which this reasoning is flawed at its foundation.
The main flawed assumption in this assumes that the information provided by an LLM is both factual, and accurate, and that these junior developers will be able to adequately determine this.
In most cases the process of validating accuracy takes more time than the process of learning it the right way in the first place, and it requires domain knowledge they do not have. This makes for an impossible task of the junior, and allows them to easily be misled to false and destructive conclusions.
It is well established that hallucinations occur in these models, and have occurred to the point where legal professionals have cited non-existent sources and perjured themselves in the process, threatening the investment they made in their career in its entirety.
These professionals are highly incentivized to avoid this outcome, but its still happened regardless of the incentives, repeatedly, with several making national news over the span of a year, and similar news the previous year.
The entire premise you make is that rapid learning occurs, but it neglects and conflates the word 'learning' whose normal context is fact and truth, to that of learning falsehoods, which reflect and promote destructive delusion.
In the absence of learning in its normal context, the latter's likelihood occurs exponentially with each additional factor, while the former trends towards 0 as a fractional. Its a chair without legs.
Following from this towards introducing it to people at a young age, where children are biologically incapable of discerning falsehood, this promotes an indoctrinated state of delusion and circular reasoning with no rational basis, in reality it is quite an evil thing in my opinion hobbling the young and ruining their futures by induction of maladaptive reasoning frameworks.
Children are a vulnerable set of the population, and such activities can only diminish the young's abilities to survive long term. Something no good person would ever do.
There is a thing in literature called a Devil's pleasure palace. It refers to a short story, though I don't remember the author, it was slavic iirc, where a Noble of the Aristocracy tests his daughter's fiance to determine if they would maintain fidelity after marriage, without them knowing its a test.
A witch and magic are employed, though one can imagine drugs being used as well, and the prospective husband is led through a series of events unbeknownst to them, in repeated attempts to induce in him every possible indulgence without consequence.
He is tested for several days, unable to leave, and when he tries to leave he is told he cannot unless he partakes. Should he cave in to desire he would have been killed on the spot, he doesn't instead choosing to die instead.
He neither agreed to this (informed consent), nor knew of it happening, thus making it both a sinister and malevolent tale of chance, where the outcome will be destructive in all but the fairy tales, especially given the vulnerability of the young.
LLM's and AI broadly depict this in their function. They deceive, and manipulate those utilizing them without any perception of this having happened, because the required knowledge to do so is outside their domain of knowledge. The same as any fallacy by authority.
No education true to any valid definition would ever use these. Involuntary indoctrination is a vile thing, and there is no place for doctrines of Learning Understanding Acceptance, lest you somehow imagine the world is better off as depicted in 1984 by Orwell, where in reality, shortage and slavery eventually devolve into famine and population collapse through the Socialist Calculation Problem (Malthus/Catton/Mises).
LLM's shouldn't be called Large Language Models, they are more appropriately Looming Liability Machines.
Irrespective of Trump’s agenda