I have found that the attention moves to thinking about the things I want done and planning, reading and iterating over the specs and other artefacts that will be part of running the agents. I still need to understand the code and iterate over it to get to a usable and maintainable point.
I find the problem is we are reaching the top of the slop curve. I will subside because it's impossible to actually do anything useful with all the output. There will just be a ton of half-finished and abandoned projects. Whatever gets into production will require more eyes on it.
I just think a lot of people are still stuck in the "holy f** I'm so productive" and working themselves into the ground being productive pumping out code. I think it's a phase that will pass.
This 100%. Maybe I have a prompting skill issue, but without my guidance Opus (and now Fable) writes some gnarly stuff with tons of small bugs, and weird design decisions.
It most certainly will replace software engineers. What's missing is, as the article suggests, the "Delivery" bit. But that's not the realm of software engineers, that's the realm of DevOps/SRE/Cloud engineers.
I work as a cloud engineer and have been contacted by multiple non-engineering friends who have now been able to create their pet projects from scratch in different languages and have it running locally, as webapps and native apps. So what they are missing is a platform to easily deploy and maintain their projects, much like a "normal" developer would. Right now it's quite tedious to set up this scaffolding, but it's absolutely possible with AGENTS.md, skills and rigid hollistic tests. Once done, non-technical people can continue developing independently without hiring any software engineers by simply telling claude/codex what they want. Claude/codex will then be able to make judgement calls based on the preset architecture, which will guide the non-technical user.
So in my anecdotal case, AI has already replaced several software engineers. Once scaffolding like this is productized, I suspect that greenfield projects can be managed entirely from a product standpoint using agentic coders + platform engineering. And that is today. Imagine in 5 years.
If it's good enough to replace plain Software Engineers it's good enough to do DevOps/SRE too. There's no meaningful distinction between those categories in this context.
I would be more worried about the viability of these business cases. If I see someone crowing about how they vibe coded some service in a weekend and now they expect $X,000,000 in ARR, I just think these type of people are not going to make it. At some point, the customer will realize that there is nothing preventing them from replacing your offering with a few dollars in API fees. Software vendors need to focus on support quality, provide some unique insight, guard proprietary data, or have some economy of scale on the hardware side. Being the first person to write that particular prompt is not a business plan.
Not buying it. The idea that deciding and delivering are things only humans can do with their intelligence seems faulty.
As it stands AIs today are not always great at making decisions (but they're getting much better), and orgs of today still trust people and hold people to account, rather than their AI systems.
Neither of these are strong moats. It's a moat only while AI systems have some limitations vs an expert human, and corporate processes are still extremely human-centric.
> Among the 270 jobs in the 1950 U.S. census, only one job was automated away — elevator operator. But many others were rendered obsolete by new technology, like the job of telegraph operator.
In that same time farm jobs went from 15% of the workforce to 2%.
yep. selective usage of stats at their best. how about factories too? conveyer belts? people losing their jobs all the time whenever automation comes in. and we just "hope" for the best they can find jobs or delusional hopefulness swinging into extremes ("be generalist!", "be specialist!", "work in service!", "learn to code!", "learn to mine coal!"), all incoherent. just listen to @pmarca to see how totally lost and incoherent tech leadership is.
They seem to address that: "This is sharply different from occupations such as agriculture in which labor demand was famously decimated due to mechanization and automation. The difference is that the amount of calories people consume is relatively fixed — even a 25% increase led to the obesity epidemic — whereas the amount of software produced has grown a millionfold."
It literally has and will even more in the future. It won't replace *all* software engineers but once the genie is out low effort low risk stuff will be done by an AI. Loveable and such have so many live projects, the alternative was a human building those.
The question is, are executives willing to give up all their power and status to an LLM or will “industry” just use AI to invent more bullshit jobs to keep everyone, including the exec relevant.
The reason humans haven’t been replaced in many areas entirely is because humans like being someone’s.
It’s not really about replacing software engineers. But about commodifying it. More software engineers (or roles responsible for code) that work for lower pay might be the trend. Or to maintain a high level of pay you wear many hats, including software.
AI won't be put in important positions of responsibility within an organization because AI providers will never accept liability for bad decisions. You can't fire Claude if it fucks up, and it's got very limited ability to learn from its mistakes. It's also incapable of making good decisions where doing so requires synthesizing more than a few hundred thousand tokens worth of domain knowledge/experience in something that doesn't have an infinite amount of synthetic verifiable training data like code and math.
In theory continuous learning (live weight updates) could help to some degree. But there's essentially no progress towards that because it requires solving a few hard, currently completely unsolved problems. 1. Weights drift over time and there's no way to re-merge them after a few tens of thousands of updates, so when a new model version was released there'd be no way to update existing continuously-learned models to that. 2. It'd allow permanent jailbreaking. And 3. A model can't learn new things without forgetting existing things, unlike humans brains which have hardware plasticity (like London taxi drivers having larger hippocampi due to having to memorize so many streets).
What's the difference between "firing" Claude vs moving to a model from a different provider? The latter seems very analogous to firing an employee for performance and backfilling with someone new.
Re the rest, it's just not my experience that models become incapable of making good decisions in cases where input token count > the context window, but ymmv based on domain.
A very extreme example of this: a couple years ago when GPT 4 was state of the art and the 32K context variant was gated to design partners I worked at an EdTech company in the college admissions space that wanted to produce quarterly reports on student progress for parents. That involved crunching a LOT of data (multiple hours of meeting transcripts per week, very detailed notes about student activities, their general profile - UK and US admissions function very differently!)
It was a difficult problem, but we _did_ manage to produce these reports 4K output tokens at a time at a level of quality that exceeded what humans could do internally, and models+the surrounding tooling have only gotten better since then.
The only bit AI can't replace is probably the need for a 'fall guy' or someone to take responsibility for something. This, however, will obviously not be sufficient to prevent job losses.
there are systems that can very well operate without "fall-guy". example:
A) diffuse responsibility. nobody is responsible. — this is what typical beruactacyl operates in.
B) everybody is self-responsible. end users use AI, they are responsible. this is typical MIT licenses. "use at own risk", literally first statement in any open source.
Just look and see what Cloud did for software engineers? It pushed us one level higher and lowered the demand for "db experts" and "low level systems people". The only ones who remained were the strong ones who were hired into the cloud companies. The rest moved up and changed careers.
Why would anyone think the same thing won't apply here? If you are still a Typescript bunny who fiddles with some newly learned React tidbit -- this won't cut it anymore. The market won't need you. Move up and adapt or move down and become an expert (harder).
I think it's people who were sloppy about programming are more interested in vive-coding. Because now they can make something without the mental rigour needed. Engineering as it should be is play of rigour. Those who value understanding system will continue the human aspect of it
The whole field of engineering set to disappear and be replaced by contractors. Of course this is what they've always wanted. That's why they do outsourcing and the whole point of AI so, basically instead of getting paid a small salary to maintain someone's money-making machine, people will bid for jobs. They'll be more and more layers of abstraction that business owners will have to pay rent to. Until it's just basically socialism.
"In this essay, we argue that there is enough evidence to reject the narrative that once AI capabilities reach a certain threshold, it will cause mass layoffs." - too late, it already did
"Can the sandwich be further compressed? We don’t think so. At one end of the pipeline, development teams need to decide what to build."
I mean, but this is talking about the process as a whole, not individual jobs.
"Farmers won't be replaced by combine harvesters - we still need someone to decide what to plant and to harvest it". Sure, but if you used to have 10 labourers in a field manually ploughing with a pair of oxen and now you have one guy driving the machinery it absolutely has replaced jobs.
Companies are already talking about "1 person teams" to deliver projects. We'll still have _some_ jobs but the ratio will change dramatically and engineering will move a lot closer to "team lead" role (and maybe even Product Manger role to boot)
> software development, as a “decide-execute-deliver sandwich”. AI compresses the “execute” layer — the middle of the sandwich — but the other two layers resist automation in a way that will not be overcome by capability improvements alone.
I really struggle to see why improved capabilities cannot deal with those other layers. I do not believe you have substantiated this claim about not being possible as capabilities improve.
> At one end of the pipeline, development teams need to decide what to build.
Developers are not the ones that do this largely. This role is far more on the side of "Product Owner". Sometimes your job covers both, but this is not the majority of the work and does not mostly require SE knowledge - some input usually.
> This layer is hard to automate because it requires thinking about user needs, market signals, organizational priorities, and in some cases regulatory constraints.
Hmm, these are language models that can talk through much of this already - but more importantly none of what is mentioned there requires software engineering. For parts that do (I'm sure someone would come to correct me if I said that there was none or seemed to suggest it is never ever ever relevant) this is a much smaller slice.
> As AI capabilities improve, the kinds of decisions that can be delegated to AI increase over time. But this does not make the “decide” layer thinner — once a decision can be delegated to AI, it is no longer a source of competitive advantage, and the value of human decision-making migrates upward. Software increases in complexity over time, so there is no ceiling to this process.
Now this is rather hidden but a huge leap in logic. The decide layer does get thinner for all the same projects, and then you simply assert that software will get more complex and so this cancels it all out.
A team of 5 may end up being able to ship what a team of 50 used to, and maybe now there are 10 teams outputting more - but is there not a clear limit to this? At some point do we not just need 45 fewer people? That there needs to be some engineers is not the same as needing anywhere near as many as we have.
For a time I think we will see increased output meaning more software, but that tails off as they get better.
> At the other end of the sandwich, human teams need to be accountable for what they deliver.
Why? And if we assume so, why does that need a software engineer?
> It is possible that some day in the future teams will ship mission-critical code without fully testing and understanding it,
You don't need to read code to test it, and people choose to ship products without fully understanding the code all the time. Literally any decision maker who is not a software engineer who knows the entire codebase does this. Companies fully ship systems that are far too complex for any single developer to even understand.
And much of software isn't mission critical. Or at least, if you want to say it is then the mission is low stakes.
> today’s AI is so unreliable that such haphazard practices would represent an existential threat to software teams and their customers.
I'd argue for a bunch of stuff this isn't true, and the whole point of the article is "never even if they get better" which is different.
> A central insight of AI as Normal Technology is that we can collectively choose to keep humans accountable through shared norms, law, and policy.
Sure, we can ban AI writing code, but will we? Is there a huge collective concern for all us high paid engineers being replaced by AI?
Sadly I think this post will mislead people, bc the difficult truth (for many) is that software engineering isn't that hard and that's why AI can easily substitute that layer (lower barriers to entry than widely believed).
We have been aggressively and enthusiastically automating away software engineering for the entire history of the computer industry. Every time we do so, we are able to build bigger, better things more quickly. When this happens, our work becomes more valuable and expectations rise to match. The world’s appetite for software has been insatiable so far. AI hasn’t replaced software engineers because every time we become more productive, the goalposts move.
There’s two things that could put an end to this. Firstly, we might finally become productive enough to exhaust the world’s appetite for software. I don’t see any evidence of this happening, but if somebody wants to make this argument, they should be clear about why this time is different to the entire history of the computer industry so far.
Secondly, if AI becomes superhuman at software engineering when acting autonomously. Specifically, AI+human developer no longer outperforms AI alone. So far, all the available evidence seems to show AI as a force multiplier for developers and that for good results, at best you can have AI doing 90% of the work as long as an expert developer is driving things.
There isn’t strong evidence that either of these situations is going to happen in the near future, so I think software engineers are safe for now. But if you have a narrow skill set and you are focused in particular areas (e.g. front-end web development), then I would worry more, because even if AI cannot replace software engineers in general, it’s quite likely to be able to completely consume specific domains with generalists holding the reins.
Quite a few developers will likely loose their jobs. In particular the ones who don't have mental capacity to work with models - the ones who are forever junior.
The engineers who can manage large scale projects using agents will, on the other hand, probably get a hefty salary bump.
In every regime where we have meaningful longitudinal data about the long run outcomes of introducing technology that is superhuman at some or all of a human's job, the combination of the machine and an expert human outperforms either alone, with probably the clearest parallel structurally being chess, though this is true of all of the hard sciences and all of frontier engineering (semiconductor design).
There will never be a human who can beat Stockfish ever again, digital intelligence has simply accelerated away from human intelligence in that regime.
There is no other human alive who can beat Magnus Carlsen. Stockfish crushes Magnus.
But Magnus and Stockfish playing together crush any conceivable combination of just human or just computer. And no serious chess player would train without a computer or decline the assist if the contest mattered.
And this is in a regime where the dominance of the machine is total, structural, and permanent, far more so than any existing AI's impact on the outcome distribution of any recent development on any white-collar knowledge work include even the most sophisticated software engineering done anywhere. The demonstrated as opposed to completely conjectural lift on SWE outcomes with Claude Code (and even that's controversial, let's take the high end of the claims) is real and changes the geometry of the situation not at all.
Nor is there any apparent limit on how much arbitrarily sophisticated software the world has an appetite for, you could take someone at the absolute top of the field (I'm a big Carmack fan let's go with Carmack), and give him cutting edge AI assist, and the best program a person can write just got better. Sweet!
And this applies anywhere from junior to Carmack: however good they were, they're better now. We can build harder things. We can have extreme quality software where previously we were stuck with some Electron jank, across the board. Does anyone think Slack would lose market share if it went back to its gaming roots and was gorgeously 3D accelerated on every surface against a backend that could instant and perfectly synchronize an arbitrary workspace on a flakey cell connection and never have an outage or data loss? Or would they rapidly shred the remaining competition and become the favored tool of everyone?
In adversarial regimes like trading or drone warfare, you better believe the best hackers have arbitrary assist if you're going to play against them.
I think the thing to be hand-wringing about isn't AI, it's that capitalism no longer seems to be an adversarial regime. The worst software rivalries in the industry look more like a pillow fight than a battle of will.
And if there is any lasting reduction in headcount, it's leaders lacking ambition agreeing tacitly to not play very hard, not AI, that is to blame for that. None of the HFT shops nor amusingly the frontier labs have reduced their hiring or compensation at all. If anything, it's going up!
It's kind of funny that you say this, because I am a frontend developer and I tend to see the state of the art as being very good at doing the boring behind-the-scenes plumbing that I don't care about, and not great at doing the kind of bespoke design work that my day job's clients want.
I'm not saying that either of us are definitively right or wrong, and I agree that having a more generalist skillset is probably the best way to succeed in this new era; I'm just pointing out that LLMs don't really own any part of the stack so thoroughly that specialists in that segment will just go away.
You can't expect that timeline to continue forever. AI is a whole different beast than programming. It is in fact what programming set out to be in the 1940s: A replica of the brain. Automating programming tasks has only gotten us so far, but that timeline is probably ending now. We no longer need to automate programming because we can talk to the machine.
> I don’t see any evidence of this happening, but if somebody wants to make this argument, they should be clear about why this time is different
It does seem to be happening - at least in mobile app stores.
There's some recent analysis that demonstrates how, despite a huge updraft in the quantity of apps released, the aggregate count of reviews and downloads remains static.
In other words, there are now many more apps. But not many (or really any?) more users
> they should be clear about why this time is different to the entire history of the computer industry so far
I can't prove the pie is fixed, but nor can you prove the pie is infinite.
Maybe this comes close to sounding patronising, but I think the key thing people miss, when talking about economic growth of software is, money has to come from somewhere. Someone has to give it to you. So it you want to keep growing, you need someone who isn't paying for software, to start. Who are these people, how much money do they have, and what other costs are you competing against?
>We have been aggressively and enthusiastically automating away software engineering for the entire history of the computer industry. Every time we do so, we are able to build bigger, better things more quickly. When this happens, our work becomes more valuable and expectations rise to match. The world’s appetite for software has been insatiable so far. AI hasn’t replaced software engineers because every time we become more productive, the goalposts move.
Anytime we became more productive in the past we become in a way that didn't remove engineers, just increased the abstraction an engineer would work in. And we did it at times of rapid expansion of computing and internet, meaning way more need for engineers counter-balancing the increased productivity.
>The world’s appetite for software has been insatiable so far.
Has it? The expansion of IT has reached global saturation, we're getting desparate, and try to push shit like Blockchain and IoT, and shoving "smart" features even where people don't want them.
And the world is full of software nobody or very few care for or use/subscribe/buy. App Store have huge "long tails" of stuff nobody cares for.
And with autonomous agents we designed something to replace the engineer altogether. So even if the demand for software increases, that can be like "spawn more agents" not "get more developers".
Some human supervisors per N agents? Sure. Equal human demand as what's now? Unlikely.
In general "we did it 5 times, to we'll surely do it 6" is not a real argument, just a hope that something will never end.
Even if the budget for software development were to stay constant, if an ever increasing part of it is spent on llm usage, it will reduce the money left for developers, resulting in mass layoffs and/or mass salary cuts.
We all tend to assume AI will only ever be good at the execute part, but what if AI will also be good at decide-deliver? What if some day, we could put AI in charge of not only running a company, but coming up with a business idea, getting funded, driving sales, pivoting until product-market fit and then scaling?
> Firstly, we might finally become productive enough to exhaust the world’s appetite for software. I don’t see any evidence of this happening, but if somebody wants to make this argument, they should be clear about why this time is different to the entire history of the computer industry so far.
Not all software but how many movies you can watch, books read? Similarly how many games you can play? How many ads you can watch during a day?
Eventually projects will have to be profitable. And even if everybody will make a great game most of them won’t be profitable because you competing for limited eyeballs time
> We have been aggressively and enthusiastically automating away software engineering for the entire history of the computer industry.
I used to work at an overnight NOC many years ago, and I literally learned bash and python just to save time so I could spend more time watching netflix or whatever instead of working. Instead, my scripts handled so many alerts that they laid off someone and gave me a promotion to being a sys admin :(
I've been chasing the dream of automating my job away and collecting a paycheck for doing nothing for decades now, and I keep getting promoted...
Your argument holds across the supply-curve of software-engineering-skills. However, there must be some threshold where a previously employed dev is now sol because AI+Human outcompetes him (ie the supply curve has shifted, and the point they used to occupy is not unprofitable)
I agree with most of what you're saying, but it's also funny that front-end developers are catching strays in your conclusion. As long as human beings are interfacing with software, there's going to be a lot of judgement and nuance necessary for building good UIs. Data structures, back-end infra, can all be alien and still work. But your UI can't be alien.
You’re right that automation has historically expanded what we can build. But the real work of software engineering isn’t coding. It’s architecting solutions: defining the problem, making trade-offs, and designing for maintainability and change. That’s significantly harder to automate, and I explained why in another reply on this thread.
Even if AI becomes competent at engineering (my two-part series covers what “System Definition” would be needed to get there), it still has a deeper limitation: AI is a consensus machine. It excels at recombining what’s already known and rewarded. Paradigm shifts and black swans almost always come from people willing to challenge the current consensus, which is exactly what AI consensus tends to suppress.
That’s the real risk to technological progress: how do we move in new directions when AI defaults to following the majority view from past experience?
https://hackernoon.com/who-dares-to-be-the-first-cassandras-...
The feeling I get is LLM's are the new Excel. I've seen lots of people develop little web based apps that tickle their interest. Things like dashboards for data that would didn't fit on their phone, table tennis scoring (really!), small account keeping apps, plotting calculated GPS path on a map.
These are tiny single use stuff. Exactly the sort of thing the company nerd would create spreadsheets for. The GUI is more advanced, but it is no more maintainable or scalable.
I don’t think pointing at all the corporate coms about ai layoffs as fake invalidates the risk. The corporate stuff can be lies while the tech‘s impact could end up being real. It’s just noise in this context.
Similarly this assumption (the burger diagram in the article) showing execution phase shrinks but somehow everything else expands to keep the burger size the same seems less than plausible.
That said some portions of swe seem like they‘re still very far off from being threatened. Especially the portions where correctness is crucial. With say web dev you’ve got a lot more room to yolo it than say navigation code for rockets. The LLM can likely do both but I don’t think anyone is vibe coding the later any time soon
Things are changing at rapid speed, there is nothing "classic" about any of this, and you should at least be able to understand that much if you want to advise people.
Nah, kids, this is an opinion column. If you can’t tell the difference, then you don’t get to sit at the adults table. I’ve been an opinion writer for most of my life, and dressing up my perspectives in scientific LARP is bullshit. And yes, I do have underlying suspicions why certain cultures feel entitled to get away with taking such a tone in their declarations. This has been formed over decades of observation and I won’t claim it is scientific…unlike these two fellas who enjoy foods I do not.
86 comments
[ 3.5 ms ] story [ 128 ms ] threadI find the problem is we are reaching the top of the slop curve. I will subside because it's impossible to actually do anything useful with all the output. There will just be a ton of half-finished and abandoned projects. Whatever gets into production will require more eyes on it.
I just think a lot of people are still stuck in the "holy f** I'm so productive" and working themselves into the ground being productive pumping out code. I think it's a phase that will pass.
I work as a cloud engineer and have been contacted by multiple non-engineering friends who have now been able to create their pet projects from scratch in different languages and have it running locally, as webapps and native apps. So what they are missing is a platform to easily deploy and maintain their projects, much like a "normal" developer would. Right now it's quite tedious to set up this scaffolding, but it's absolutely possible with AGENTS.md, skills and rigid hollistic tests. Once done, non-technical people can continue developing independently without hiring any software engineers by simply telling claude/codex what they want. Claude/codex will then be able to make judgement calls based on the preset architecture, which will guide the non-technical user.
So in my anecdotal case, AI has already replaced several software engineers. Once scaffolding like this is productized, I suspect that greenfield projects can be managed entirely from a product standpoint using agentic coders + platform engineering. And that is today. Imagine in 5 years.
As it stands AIs today are not always great at making decisions (but they're getting much better), and orgs of today still trust people and hold people to account, rather than their AI systems.
Neither of these are strong moats. It's a moat only while AI systems have some limitations vs an expert human, and corporate processes are still extremely human-centric.
> Among the 270 jobs in the 1950 U.S. census, only one job was automated away — elevator operator. But many others were rendered obsolete by new technology, like the job of telegraph operator.
In that same time farm jobs went from 15% of the workforce to 2%.
check Stripe Press latest on indutrial automation: https://press.stripe.com/origins-of-efficiency
The reason humans haven’t been replaced in many areas entirely is because humans like being someone’s.
In theory continuous learning (live weight updates) could help to some degree. But there's essentially no progress towards that because it requires solving a few hard, currently completely unsolved problems. 1. Weights drift over time and there's no way to re-merge them after a few tens of thousands of updates, so when a new model version was released there'd be no way to update existing continuously-learned models to that. 2. It'd allow permanent jailbreaking. And 3. A model can't learn new things without forgetting existing things, unlike humans brains which have hardware plasticity (like London taxi drivers having larger hippocampi due to having to memorize so many streets).
What's the difference between "firing" Claude vs moving to a model from a different provider? The latter seems very analogous to firing an employee for performance and backfilling with someone new.
Re the rest, it's just not my experience that models become incapable of making good decisions in cases where input token count > the context window, but ymmv based on domain.
A very extreme example of this: a couple years ago when GPT 4 was state of the art and the 32K context variant was gated to design partners I worked at an EdTech company in the college admissions space that wanted to produce quarterly reports on student progress for parents. That involved crunching a LOT of data (multiple hours of meeting transcripts per week, very detailed notes about student activities, their general profile - UK and US admissions function very differently!)
It was a difficult problem, but we _did_ manage to produce these reports 4K output tokens at a time at a level of quality that exceeded what humans could do internally, and models+the surrounding tooling have only gotten better since then.
A) diffuse responsibility. nobody is responsible. — this is what typical beruactacyl operates in.
B) everybody is self-responsible. end users use AI, they are responsible. this is typical MIT licenses. "use at own risk", literally first statement in any open source.
Why would anyone think the same thing won't apply here? If you are still a Typescript bunny who fiddles with some newly learned React tidbit -- this won't cut it anymore. The market won't need you. Move up and adapt or move down and become an expert (harder).
I mean, but this is talking about the process as a whole, not individual jobs.
"Farmers won't be replaced by combine harvesters - we still need someone to decide what to plant and to harvest it". Sure, but if you used to have 10 labourers in a field manually ploughing with a pair of oxen and now you have one guy driving the machinery it absolutely has replaced jobs.
Companies are already talking about "1 person teams" to deliver projects. We'll still have _some_ jobs but the ratio will change dramatically and engineering will move a lot closer to "team lead" role (and maybe even Product Manger role to boot)
I really struggle to see why improved capabilities cannot deal with those other layers. I do not believe you have substantiated this claim about not being possible as capabilities improve.
> At one end of the pipeline, development teams need to decide what to build.
Developers are not the ones that do this largely. This role is far more on the side of "Product Owner". Sometimes your job covers both, but this is not the majority of the work and does not mostly require SE knowledge - some input usually.
> This layer is hard to automate because it requires thinking about user needs, market signals, organizational priorities, and in some cases regulatory constraints.
Hmm, these are language models that can talk through much of this already - but more importantly none of what is mentioned there requires software engineering. For parts that do (I'm sure someone would come to correct me if I said that there was none or seemed to suggest it is never ever ever relevant) this is a much smaller slice.
> As AI capabilities improve, the kinds of decisions that can be delegated to AI increase over time. But this does not make the “decide” layer thinner — once a decision can be delegated to AI, it is no longer a source of competitive advantage, and the value of human decision-making migrates upward. Software increases in complexity over time, so there is no ceiling to this process.
Now this is rather hidden but a huge leap in logic. The decide layer does get thinner for all the same projects, and then you simply assert that software will get more complex and so this cancels it all out.
A team of 5 may end up being able to ship what a team of 50 used to, and maybe now there are 10 teams outputting more - but is there not a clear limit to this? At some point do we not just need 45 fewer people? That there needs to be some engineers is not the same as needing anywhere near as many as we have.
For a time I think we will see increased output meaning more software, but that tails off as they get better.
> At the other end of the sandwich, human teams need to be accountable for what they deliver.
Why? And if we assume so, why does that need a software engineer?
> It is possible that some day in the future teams will ship mission-critical code without fully testing and understanding it,
You don't need to read code to test it, and people choose to ship products without fully understanding the code all the time. Literally any decision maker who is not a software engineer who knows the entire codebase does this. Companies fully ship systems that are far too complex for any single developer to even understand.
And much of software isn't mission critical. Or at least, if you want to say it is then the mission is low stakes.
> today’s AI is so unreliable that such haphazard practices would represent an existential threat to software teams and their customers.
I'd argue for a bunch of stuff this isn't true, and the whole point of the article is "never even if they get better" which is different.
> A central insight of AI as Normal Technology is that we can collectively choose to keep humans accountable through shared norms, law, and policy.
Sure, we can ban AI writing code, but will we? Is there a huge collective concern for all us high paid engineers being replaced by AI?
There’s two things that could put an end to this. Firstly, we might finally become productive enough to exhaust the world’s appetite for software. I don’t see any evidence of this happening, but if somebody wants to make this argument, they should be clear about why this time is different to the entire history of the computer industry so far.
Secondly, if AI becomes superhuman at software engineering when acting autonomously. Specifically, AI+human developer no longer outperforms AI alone. So far, all the available evidence seems to show AI as a force multiplier for developers and that for good results, at best you can have AI doing 90% of the work as long as an expert developer is driving things.
There isn’t strong evidence that either of these situations is going to happen in the near future, so I think software engineers are safe for now. But if you have a narrow skill set and you are focused in particular areas (e.g. front-end web development), then I would worry more, because even if AI cannot replace software engineers in general, it’s quite likely to be able to completely consume specific domains with generalists holding the reins.
The engineers who can manage large scale projects using agents will, on the other hand, probably get a hefty salary bump.
There will never be a human who can beat Stockfish ever again, digital intelligence has simply accelerated away from human intelligence in that regime.
There is no other human alive who can beat Magnus Carlsen. Stockfish crushes Magnus.
But Magnus and Stockfish playing together crush any conceivable combination of just human or just computer. And no serious chess player would train without a computer or decline the assist if the contest mattered.
And this is in a regime where the dominance of the machine is total, structural, and permanent, far more so than any existing AI's impact on the outcome distribution of any recent development on any white-collar knowledge work include even the most sophisticated software engineering done anywhere. The demonstrated as opposed to completely conjectural lift on SWE outcomes with Claude Code (and even that's controversial, let's take the high end of the claims) is real and changes the geometry of the situation not at all.
Nor is there any apparent limit on how much arbitrarily sophisticated software the world has an appetite for, you could take someone at the absolute top of the field (I'm a big Carmack fan let's go with Carmack), and give him cutting edge AI assist, and the best program a person can write just got better. Sweet!
And this applies anywhere from junior to Carmack: however good they were, they're better now. We can build harder things. We can have extreme quality software where previously we were stuck with some Electron jank, across the board. Does anyone think Slack would lose market share if it went back to its gaming roots and was gorgeously 3D accelerated on every surface against a backend that could instant and perfectly synchronize an arbitrary workspace on a flakey cell connection and never have an outage or data loss? Or would they rapidly shred the remaining competition and become the favored tool of everyone?
In adversarial regimes like trading or drone warfare, you better believe the best hackers have arbitrary assist if you're going to play against them.
I think the thing to be hand-wringing about isn't AI, it's that capitalism no longer seems to be an adversarial regime. The worst software rivalries in the industry look more like a pillow fight than a battle of will.
And if there is any lasting reduction in headcount, it's leaders lacking ambition agreeing tacitly to not play very hard, not AI, that is to blame for that. None of the HFT shops nor amusingly the frontier labs have reduced their hiring or compensation at all. If anything, it's going up!
It's kind of funny that you say this, because I am a frontend developer and I tend to see the state of the art as being very good at doing the boring behind-the-scenes plumbing that I don't care about, and not great at doing the kind of bespoke design work that my day job's clients want.
I'm not saying that either of us are definitively right or wrong, and I agree that having a more generalist skillset is probably the best way to succeed in this new era; I'm just pointing out that LLMs don't really own any part of the stack so thoroughly that specialists in that segment will just go away.
It does seem to be happening - at least in mobile app stores.
There's some recent analysis that demonstrates how, despite a huge updraft in the quantity of apps released, the aggregate count of reviews and downloads remains static.
In other words, there are now many more apps. But not many (or really any?) more users
Take a look at p40 / figure 12 of "WRITING CODE VS. SHIPPING CODE: PRODUCTIVITY EFFECTS ACROSS GENERATIONS OF AI CODING TOOLS" (https://www.nber.org/system/files/working_papers/w35275/w352...)
Their analysis is on pg42-43
> they should be clear about why this time is different to the entire history of the computer industry so far
I can't prove the pie is fixed, but nor can you prove the pie is infinite.
Maybe this comes close to sounding patronising, but I think the key thing people miss, when talking about economic growth of software is, money has to come from somewhere. Someone has to give it to you. So it you want to keep growing, you need someone who isn't paying for software, to start. Who are these people, how much money do they have, and what other costs are you competing against?
Anytime we became more productive in the past we become in a way that didn't remove engineers, just increased the abstraction an engineer would work in. And we did it at times of rapid expansion of computing and internet, meaning way more need for engineers counter-balancing the increased productivity.
>The world’s appetite for software has been insatiable so far.
Has it? The expansion of IT has reached global saturation, we're getting desparate, and try to push shit like Blockchain and IoT, and shoving "smart" features even where people don't want them.
And the world is full of software nobody or very few care for or use/subscribe/buy. App Store have huge "long tails" of stuff nobody cares for.
And with autonomous agents we designed something to replace the engineer altogether. So even if the demand for software increases, that can be like "spawn more agents" not "get more developers".
Some human supervisors per N agents? Sure. Equal human demand as what's now? Unlikely.
In general "we did it 5 times, to we'll surely do it 6" is not a real argument, just a hope that something will never end.
Jervons Paradox
https://en.wikipedia.org/wiki/Jevons_paradox
Who would profit from such a company?
Not all software but how many movies you can watch, books read? Similarly how many games you can play? How many ads you can watch during a day?
Eventually projects will have to be profitable. And even if everybody will make a great game most of them won’t be profitable because you competing for limited eyeballs time
I used to work at an overnight NOC many years ago, and I literally learned bash and python just to save time so I could spend more time watching netflix or whatever instead of working. Instead, my scripts handled so many alerts that they laid off someone and gave me a promotion to being a sys admin :(
I've been chasing the dream of automating my job away and collecting a paycheck for doing nothing for decades now, and I keep getting promoted...
These are tiny single use stuff. Exactly the sort of thing the company nerd would create spreadsheets for. The GUI is more advanced, but it is no more maintainable or scalable.
Similarly this assumption (the burger diagram in the article) showing execution phase shrinks but somehow everything else expands to keep the burger size the same seems less than plausible.
That said some portions of swe seem like they‘re still very far off from being threatened. Especially the portions where correctness is crucial. With say web dev you’ve got a lot more room to yolo it than say navigation code for rockets. The LLM can likely do both but I don’t think anyone is vibe coding the later any time soon
It's always the business owner who replaces workers. Let's not anthropomorphize a bunch of graphics cards
Nah, kids, this is an opinion column. If you can’t tell the difference, then you don’t get to sit at the adults table. I’ve been an opinion writer for most of my life, and dressing up my perspectives in scientific LARP is bullshit. And yes, I do have underlying suspicions why certain cultures feel entitled to get away with taking such a tone in their declarations. This has been formed over decades of observation and I won’t claim it is scientific…unlike these two fellas who enjoy foods I do not.