The right question is how much human code can a human push now vs prior to AI.
Everything we've done in coding has been assisted.
Prior to this current generation of web applications, we had the advent of concepts like Object Orientated Programming and prior to that even C was a massive move up from Assembly and punch cards.
AI has written a lot of code. AI has written very little high velocity production code by itself (ie. for people with no coding background).
1. AI has pushed a lot of code
2. AI has pushed almost no code without the oversight of human software engineers
3. Software engineers are pushing a magnitude more code and producing more functional utility and solving more bugs than ever before
I don't know what the future holds, but I do think that this is not a new trend to use software to help humans build faster, and I don't think software has the ability to fully replace humans (yet).
From my personal account, I started with PHP and Perl (high school and college) and then graduated to Ruby on Rails (early dev career) and now its Python and JS.
I would say Ruby on Rails was a 10x on raw PHP in terms of feature specs per hour and AI is a 10x on Ruby on Rails (and its derivatives).
We're probably 100x the developer productivity on a per developer basis from the early days of Web 2.0 with PHP, just a personal anecdote though.
> We're probably 100x the developer productivity on a per developer basis from the early days of Web 2.0 with PHP, just a personal anecdote though.
Only if you compare create a website in PHP 20 years ago vs using wordpress. But to create a project like wordpress from zero now is as difficult as it was 20 years ago.
> Only if you compare create a website in PHP 20 years ago vs using wordpress. But to create a project like wordpress from zero now is as difficult as it was 20 years ago.
Roll to disbelieve on this one. There's no way that creating something like WordPress 20 years ago, before any modern-day web application frameworks, was anywhere near as easy. I'm not even talking React/Next or whatever - this was before RoR, before Django (or at least very early versions of both).
I think it's pretty impossible to look around and not notice that technical advances have improved a lot of things in the world - why does it sound credible that programming itself hasn't improved at all in twenty years? Even look at languages like C++ today compared to 20 years ago, they're massively different - do you think all those changes to the languages are neutral or negative in terms of productivity?
Quality does not change as fast :)) Actually, most of the times, when things go faster, the quality drops as fast.
> There's no way that creating something like WordPress 20 years ago, before any modern-day web application frameworks, was anywhere near as easy
Wordpress was 10% about coding and 90% about community. This 90% part can't be automated by AI or fast-tracked using RoR, Django or whatever. So much that none of the hundreds of wordpress alternatives created in the past 2 decades got any closer to replace it.
> Wordpress was 10% about coding and 90% about community.
I completely agree. I just didn't think we were talking about that aspect, just how long it takes to actually build the code of WordPress.
You might be saying that making the technical-building aspects of most technical products go faster won't necessarily have as big an impact as many programmers believe. Programming isn't the core of most products to the degree most programmers think, and instead the core of most products/companies is more about product thinking, sales & marketing, etc. I agree with this sentiment a lot.
That aside, I think there's no question it's faster to program in Python or C++ right now, than it was 20 years ago. Things really have improved on the programming language front (and related libraries, and other advances).
Your reply is pseudo-intellectual Slashdot-style Frist Posting, to which I venture:
Can AI FirstPost on HN better than humans? If so, why do they seem to lack the curiosity and gumption to make their own 'Ask HN'? Is there a matching preference to snipe internet points with replies?
Do they also use real names or pseudo real names believing such action will not boost their LinkedIn profile? Have they A/B tested the optimal level of passive aggression in HN replies vis avoidance behaviour of potential future relationships?
Not OpenAI, but Anthropic CPO Mike Krieger said in response to a question of how much of Claude Code is written by Claude Code: "At this point, I would be shocked if it wasn't 95% plus. I'd have to ask Boris and the other tech leads on there."
> During take-home assessments Complete these without Claude unless we indicate otherwise. We’d like to assess your unique skills and strengths. We'll be clear when AI is allowed (example: "You may use Claude for this coding challenge").
> During live interviews This is all you–no AI assistance unless we indicate otherwise. We’re curious to see how you think through problems in real time. If you require any accommodations for your interviews, please let your recruiter know early in the process.
He'd have to ask yet did not ask? A CPO of an AI company?
TFA says "How Anthropic uses AI to write 90-95% of code for some products and the surprising new bottlenecks this creates".
for some products.
If it were 95% of anything useful, Anthropic would not still have >1000 employees, and the rest of the economy would be collapsing, and governments would be taking some kind of action.
> If it were 95% of anything useful, Anthropic would not still have >1000 employees
I think firing people does not come as a logical conclusion of 95% of code being written by Claude Code. There is a big difference between AI autonomously writing code and developers just finding it easier to prompt changes rather than typing them manually.
In one case, you have an automated software engineer, and may be able to reduce your headcount. In the other, developers may just be slightly more productive or even just enjoy writing code using AI more, but the coding is still very much driven by the developers themselves. I think right now Claude Code shows signs of (1) for simple cases, but mostly falls into the (2) bucket.
I don't doubt it, especially when you have an organization that is focused on building the most effective tooling possible. I'd imagine that they use AI even when it isn't the most optimal, because they are trying to build experiences that will allow everyone else to do the same.
So let's take it on face value and say 95% is written by AI. When you free one bottleneck you expose the next. You still need developers to review it to make sure it's doing the right thing. You still need developers to be able to translate the business context into instructions that make the right product. You have to engage with the product. You need to architect the system - the context windows mean that the tasks can't just be handed off to AI.
So, The role of the programmer changes - you still need technical competence, but to serve the judgement calls of "what is right for the product?" Perhaps there's a world where developers and product management merges, but I think we will still need the people.
Been using claude code almost daily for over a month. It is the smartest junior developer I've ever seen; it can spew high-quality advanced code and with the same confidence, spew utter garbage or over-engineered crap; it can confidently tell you a task is done and passing tests, with glaring bugs in it; it can happily introduce security bugs if it's a shurtcut to finish something. And sometimes, will just tell you "not gonna do it, it takes too much time, so here's a todo comment". In short, it requires constant supervision and careful code review - you still need experienced developers for this.
Weasel words. No different than Nadella claiming 50%.
When you drill in you find out the real claims distill into something like "95% of the code, in some of the projects, was written by humans who sometimes use AI in their coding tasks."
If they don't produce data, show the study or other compelling examples, don't believe the claims; it's just marketing and marketing can never be trusted because marketing is inherently manipulative.
It could be true, the primary issue here is that it's the wrong metric. I mean you could write 100% of your code with AI if you were basically telling it exactly what to write...
If we assume it isn't a lie, then given current AI capabilities we should assume that AI isn't being used in a maximally efficient way.
However, developer efficiency isn't the only metric a company like Anthropic would care about, after all they're trying to build the best coding assistant with Claude Code. So for them understanding the failure cases, and the prompting need to recover from those failures is likely more important than just lines of code their developers are producing per hour.
So my guess (assuming the claim is true) is that Anthropic are forcing their employees to use Claude Code to write as much code as possible to collect data on how to improve it.
This is classic marketing speak. Plant the idea of 95+% while in actuality this guy doesn't make any hard claims about the percentage. It can just as well be 0 or 5%.
It’s worth pointing out that the statement is about how much of Claude Code is written with it and not how much of the codebase of the whole company. In the more critical parts of the codebase where bugs can cause bigger problems, I expect a lot less code to be fully AI generated.
Standard CxO mentality. “I think the facts about our product might be true but I won’t say it because the shareholders and SEC will hang me when they find out it’s bullshit.” Then defer to next monkey in circus. By which time the tech press, which seems to have a serious problem with literacy and honesty (gotta get those clicks) extrapolates it for them. Then analysts summarise those things as projections. Urgh.
The other tactic is saying two unrelated things in a sentence and hoping you think it’s causal, not a fuck up and some marketing at the same time.
This guy is so full of shit. Anthropic’s leadership are all talk and hype at this point. And they’re not the only ones guilty of this in this hype cycle by far.
I don't disagree it's a naive take, and I would love to read about some examples where this happened.
I haven't seen too many industries be automated away first hand, and I'm sure there are historic examples. I wouldn't expect the lamp lighters to have been championing the rise of electric lamps. Maybe they did though because it meant they could work less hours.
Software developers have been automating away software developers since the beginning of the field. Higher level languages, better+safer languages, improved IDEs. All the ways in which we made developing software easier and more efficient. QA teams investing in automated testing.
One example is probably blacksmiths. Before, each larger village used to have one. Nowadays, this is not the case any more as people order replacement parts for whatever metal piece that broke in your car, house, etc. The metal molding happens in factories instead. But in order you still need to have people who know how to mold metals. The early folks for this profession were blacksmiths. Nowadays you would not call them any more. But it's definitely been "engineers with AI".
Same goes for industrialized food. The huge industrial bakeries still employ people who know how to bake bread, but they can do 1000x more than they could do if they had to knead the dough on their own, separate it, form little pieces of bread, order it nicely on a tray, etc.
Now it's a machine doing most pieces of work, but they still have folks who design new bread products, adjust machines, know what to change in the formula if one particular product from one particular supplier is not available in the quantities needed, etc. A lot of the skills needed aligns enough with the traditional job of a baker that these factories employ actual bakers.
As for software, whenever I use AI tools, I don't really feel that they help me save time much. But it's entirely possible that this is going to change in the future. They are only going to get better. And of course a lot of people say it's useful for them, so it already affects folks. Just because something is not useful to me doesn't mean it's not going to be useful to anybody.
Yeah, this needs to be considered down to the individual level. If your employer not only incentives you to go against your best interests but also threatens the stability of your role, then your choices are either do the job or accept that you might not be reliably employed. This is the culmination of decades of moving power from the employee to the employer.
> If AI is a threat to software engineering, I wouldn't expect many software engineers to actively accelerate that trend. I personally don't view it as a threat, but some people (non engineers?) obviously do.
Software engineers have been automating away jobs for other people for nearly a century. It would be quite rich if the profession suddenly felt qualms about the process! (TBC I think automation is great and should always be pursued. Ofc there are real human concerns when change happens quickly but I am skeptical that smashing the looms is the best response)
Software engineering has also been automating its own jobs for ages. The first thing we engineers do when asked to do a repetitive thing is find ways to automate it. I think the industry had qualms about losing their jobs. But honestly what are the examples of people losing their jobs to software ? Everybody says that this has happened many times yet examples are hard to come by.
> But honestly what are the examples of people losing their jobs to software?
And furthermore, what is the full causality chain that links the precise PR in provided example software to the employment termination decision? Lacking that, can you really assert the software 'automated' a dev out of the job?
Honeslty I would even accept statistical test (but conclusive, no p-hacked bullshit). Yet I just dont see examples of devs being automated out of a job.
Many of these still exist today, though is not as large of numbers. Maybe it's the same for software engineers. I still don't recall hearing of bank tellers accelerating the pace of ATMs or Travel Agents encouraging their clients to use Expedia.
> If AI is a threat to software engineering, I wouldn't expect many software engineers to actively accelerate that trend.
There are two strong forces at play. Employees generally want to put in the least amount of effort possible and go home at 5. Employers want to save money and pay for fewer employees. AI creates a strong symbiosis here and both sides are focused on a short term win.
not just employees, founders as well :) as a serial founder, I find AI exciting because after my 3rd company, thinking of my 4th was quite exhausting but AI has re-invigorated by ambition to start another company.
Sometime very soon we’ll cross a threshold where most people can do most of their coding through a tool like Claude and be more productive. It will feel like coding still, breaking and building things, but they will get more done in the same time. Everyone will switch.
If you work at a startup, sometimes you have to do extremely mundane and boring things unrelated to your expertise (“wear many hats”). AI is especially useful in those cases so you can quickly go back to working on things you enjoy.
I don't work for OpenAI and I doubt some random employee is going to come here and share what is likely a secret. I'm in the industry though so I have some idea of what's going on these days, both where I work and more broadly.
AI is getting better at writing code. However writing code is just some fraction of the work of many software engineers. AI doesn't work independently, it needs to be guided, its work needs to be reviewed, tested etc. There are some domains where it does better and some domains where it doesn't. There's a range of "AI" work between auto-complete style work, assisting in understanding a code base, and writing code from some spec or doing other types work.
All in all I would say it's a decent improvement to productivity for many situations. It's really hard to say how much and it's also not a zero sum game, as productivity improves there's more work.
Something to keep in mind is that if you look at a modern software project likely most of the code executing is not code written by the developers of that project. There's a huge stack of open source bits executing for almost any new project.
Specifically in OpenAI you also need to consider what type of software they are likely writing. Some of it may be more or less "vanilla" code and other is likely very specialized/performance critical. The vanilla code like API wrappers or simple front end pieces is likely more amenable to be written by AI whereas the more cutting edge algorithmic/scheduling/optimization work is almost certainly not done by AI. At least yet.
As software organizations become larger there's a lot of overhead and waste. It is possible that AI can enable smaller teams and that has a multiplicative effect because it lets you reduce that waste/overhead. There are likely also software engineers who will become better/adapt to new workflows and some who will not. It's really hard to say where things are going but overall my sense is that this like many other innovations will lead to more software and more jobs and not the other way around. There are many moving pieces here, not just AI itself but geopolitics, macro-economics, etc. Where are those new jobs going to get created, what new types of software/technology are going to be created etc. etc. History seems to show us that we'll adapt/evolve and grow.
I absolutely believe that a large proportion of new code written is at least in-part AI generated, but that doesn't mean a large proportion of new code is 100% soup-to-nuts/pull-request-to-merge the result of decisions made by an agent and not a human. I doubt that very much.
I think the difference between situations where AI-driven development works and doesn't is going to be largely down to the quality of the engineers who are supervising and prompting to generate that code, and the degree to which they manually evaluate it before moving it forward. I think you'll find that good engineers who understand what they're telling an agent to do are still extremely valuable, and are unlikely to go anywhere in the short to mid term. AI tools are not yet at the point where they are reliable on their own, even for systems they helped build, and it's unclear whether they will be any time soon purely through model scaling (though it's possible).
Note that neither of these require direct experience with using AI coding agents, just an interest in the topic! Contrast that with many companies who now demand engineers explain how they are using AI-driven workflows. When they are being serious about getting people to do the work that will make them money, rather than engaging in marketing hype, AI companies are honest: AI agents are tools, just like IDEs, version control systems, etc. It's up to the wise engineer to use them in a valuable way.
Is it possible they're just hiring these folks to try and make their models better to later replace those people? It's possible. But I'm not sure when in time, if ever, they'll reach the point where that was viable.
>> new code is 100% soup-to-nuts/pull-request-to-merge the result of decisions made by an agent
I am beginning to have more success with this in simpler parts of the code. Particularly if you already have a good example of how to do something and you need something very similar. I usually have to do a few tweaks but generally quite useful.
I would argue intellisense is far closer to a select statement than "ai" anything. What could come after myString.s is <<select method_name from all_methods where type_name = 'String' and method_name like '%s%'>> where some IDEs prefer <<like 's%'>> and others the <<contains>> style
IJ does some truly stellar introspection to offer sane defaults in the current completion context, such as offering only variables of the correct type for parameters but I think of that as discipline and not AI. Plus, IJ never once made up an API that didn't exist
I don't work at Open AI but I use Codex as I imagine most people there do to.
I actually use it from the web app not the cli. So far I've run over 100 codex sessions a great percentage of which I turned in to pull requests.
I kick off codex for 1 or more tasks and then review the code later. So they run in the background while I do other things. Occasionally I need to re-prompt if I don't like the results.
If I like the code I create a PR and test it locally. I would say 90% of my PR's are AI generated (with human in the loop).
Since using codex, I very rarely create hand written PR's.
IME I have had to review a lot of code written by AI. At certain almost all of it. And sometimes write the code myself because LLMs just don't get it. AI has written 95% of my code but not without any review.
I think this is the wrong question.
The right question is how much human code can a human push now vs prior to AI.
Everything we've done in coding has been assisted.
Prior to this current generation of web applications, we had the advent of concepts like Object Orientated Programming and prior to that even C was a massive move up from Assembly and punch cards.
AI has written a lot of code. AI has written very little high velocity production code by itself (ie. for people with no coding background).
In Ruby on Rails, the concept of fast coding has been around for over 20 years, look up this concept of Scaffolding: https://www.rubyguides.com/2020/03/rails-scaffolding/
So to answer your question,
1. AI has pushed a lot of code 2. AI has pushed almost no code without the oversight of human software engineers 3. Software engineers are pushing a magnitude more code and producing more functional utility and solving more bugs than ever before
I don't know what the future holds, but I do think that this is not a new trend to use software to help humans build faster, and I don't think software has the ability to fully replace humans (yet).
> Software engineers are pushing a magnitude more code and producing more functional utility and solving more bugs than ever before
Citation needed
More programmers than ever before makes this implicitly true.
It’s not as clever as the author hoped.
From my personal account, I started with PHP and Perl (high school and college) and then graduated to Ruby on Rails (early dev career) and now its Python and JS.
I would say Ruby on Rails was a 10x on raw PHP in terms of feature specs per hour and AI is a 10x on Ruby on Rails (and its derivatives).
We're probably 100x the developer productivity on a per developer basis from the early days of Web 2.0 with PHP, just a personal anecdote though.
> We're probably 100x the developer productivity on a per developer basis from the early days of Web 2.0 with PHP, just a personal anecdote though.
Only if you compare create a website in PHP 20 years ago vs using wordpress. But to create a project like wordpress from zero now is as difficult as it was 20 years ago.
> Only if you compare create a website in PHP 20 years ago vs using wordpress. But to create a project like wordpress from zero now is as difficult as it was 20 years ago.
Roll to disbelieve on this one. There's no way that creating something like WordPress 20 years ago, before any modern-day web application frameworks, was anywhere near as easy. I'm not even talking React/Next or whatever - this was before RoR, before Django (or at least very early versions of both).
I think it's pretty impossible to look around and not notice that technical advances have improved a lot of things in the world - why does it sound credible that programming itself hasn't improved at all in twenty years? Even look at languages like C++ today compared to 20 years ago, they're massively different - do you think all those changes to the languages are neutral or negative in terms of productivity?
Quality does not change as fast :)) Actually, most of the times, when things go faster, the quality drops as fast.
> There's no way that creating something like WordPress 20 years ago, before any modern-day web application frameworks, was anywhere near as easy
Wordpress was 10% about coding and 90% about community. This 90% part can't be automated by AI or fast-tracked using RoR, Django or whatever. So much that none of the hundreds of wordpress alternatives created in the past 2 decades got any closer to replace it.
> Wordpress was 10% about coding and 90% about community.
I completely agree. I just didn't think we were talking about that aspect, just how long it takes to actually build the code of WordPress.
You might be saying that making the technical-building aspects of most technical products go faster won't necessarily have as big an impact as many programmers believe. Programming isn't the core of most products to the degree most programmers think, and instead the core of most products/companies is more about product thinking, sales & marketing, etc. I agree with this sentiment a lot.
That aside, I think there's no question it's faster to program in Python or C++ right now, than it was 20 years ago. Things really have improved on the programming language front (and related libraries, and other advances).
> project like wordpress from zero now is as difficult as it was 20 years ago
This specific point is patently untrue. loveable/v0/etc excel at creating CMSes / UIs for content in hours.
Your examples are a far cry from Wordpress and will probably not exist in 5 years.
I'm pushing like 30-50% more code per week since using AI, and I'm only really using Cursor.
>I think this is the wrong question.
Why? It is a perfectly legitimate question where the answer is some sort of a percentage.
You write an essay in bullet points then you put the essay into ChatGPT to clean up and then you edit the output again before sending in the essay.
Who wrote the essay and in which percentage?
> Software engineers are pushing a magnitude more code and producing more functional utility and solving more bugs than ever before
Maybe, but not at Microsoft, Google or Apple.
That's not a better question but a different one.
Your reply is pseudo-intellectual Slashdot-style Frist Posting, to which I venture:
Can AI FirstPost on HN better than humans? If so, why do they seem to lack the curiosity and gumption to make their own 'Ask HN'? Is there a matching preference to snipe internet points with replies?
Do they also use real names or pseudo real names believing such action will not boost their LinkedIn profile? Have they A/B tested the optimal level of passive aggression in HN replies vis avoidance behaviour of potential future relationships?
>I think this is the wrong question.
Oh, please don't turn this site into StackOverflow 2.0.
OP's question is well defined. Any sentient being can understand what s/he meant.
Not OpenAI, but Anthropic CPO Mike Krieger said in response to a question of how much of Claude Code is written by Claude Code: "At this point, I would be shocked if it wasn't 95% plus. I'd have to ask Boris and the other tech leads on there."
[0] https://www.lennysnewsletter.com/p/anthropics-cpo-heres-what...
They are likely lying:
https://www.anthropic.com/candidate-ai-guidance
> During take-home assessments Complete these without Claude unless we indicate otherwise. We’d like to assess your unique skills and strengths. We'll be clear when AI is allowed (example: "You may use Claude for this coding challenge").
> During live interviews This is all you–no AI assistance unless we indicate otherwise. We’re curious to see how you think through problems in real time. If you require any accommodations for your interviews, please let your recruiter know early in the process.
He'd have to ask yet did not ask? A CPO of an AI company?
That's not evidence of anything to do with their hired developers. Interview practices have never reflected on-the-job practices
95% of Claude Code was actually written on a whiteboard!
95% of Claude Code is actually just inverting a binary tree
TFA says "How Anthropic uses AI to write 90-95% of code for some products and the surprising new bottlenecks this creates".
for some products.
If it were 95% of anything useful, Anthropic would not still have >1000 employees, and the rest of the economy would be collapsing, and governments would be taking some kind of action.
Yet none of that appears to be happening. Why?
> If it were 95% of anything useful, Anthropic would not still have >1000 employees
I think firing people does not come as a logical conclusion of 95% of code being written by Claude Code. There is a big difference between AI autonomously writing code and developers just finding it easier to prompt changes rather than typing them manually.
In one case, you have an automated software engineer, and may be able to reduce your headcount. In the other, developers may just be slightly more productive or even just enjoy writing code using AI more, but the coding is still very much driven by the developers themselves. I think right now Claude Code shows signs of (1) for simple cases, but mostly falls into the (2) bucket.
I don't doubt it, especially when you have an organization that is focused on building the most effective tooling possible. I'd imagine that they use AI even when it isn't the most optimal, because they are trying to build experiences that will allow everyone else to do the same.
So let's take it on face value and say 95% is written by AI. When you free one bottleneck you expose the next. You still need developers to review it to make sure it's doing the right thing. You still need developers to be able to translate the business context into instructions that make the right product. You have to engage with the product. You need to architect the system - the context windows mean that the tasks can't just be handed off to AI.
So, The role of the programmer changes - you still need technical competence, but to serve the judgement calls of "what is right for the product?" Perhaps there's a world where developers and product management merges, but I think we will still need the people.
Been using claude code almost daily for over a month. It is the smartest junior developer I've ever seen; it can spew high-quality advanced code and with the same confidence, spew utter garbage or over-engineered crap; it can confidently tell you a task is done and passing tests, with glaring bugs in it; it can happily introduce security bugs if it's a shurtcut to finish something. And sometimes, will just tell you "not gonna do it, it takes too much time, so here's a todo comment". In short, it requires constant supervision and careful code review - you still need experienced developers for this.
Anthropic should have a check for CC, that runs “rg ‘todo’” after every turn. Then tells Claude to fix it if not.
Weasel words. No different than Nadella claiming 50%.
When you drill in you find out the real claims distill into something like "95% of the code, in some of the projects, was written by humans who sometimes use AI in their coding tasks."
If they don't produce data, show the study or other compelling examples, don't believe the claims; it's just marketing and marketing can never be trusted because marketing is inherently manipulative.
It could be true, the primary issue here is that it's the wrong metric. I mean you could write 100% of your code with AI if you were basically telling it exactly what to write...
If we assume it isn't a lie, then given current AI capabilities we should assume that AI isn't being used in a maximally efficient way.
However, developer efficiency isn't the only metric a company like Anthropic would care about, after all they're trying to build the best coding assistant with Claude Code. So for them understanding the failure cases, and the prompting need to recover from those failures is likely more important than just lines of code their developers are producing per hour.
So my guess (assuming the claim is true) is that Anthropic are forcing their employees to use Claude Code to write as much code as possible to collect data on how to improve it.
> I would be shocked if it wasn't 95% plus.
This is classic marketing speak. Plant the idea of 95+% while in actuality this guy doesn't make any hard claims about the percentage. It can just as well be 0 or 5%.
"You lied" "No I didn't. Look my face is shocked can't you tell?"
Sure but what did the CTO say ? Also was he shocked ? There's no definitive answer, this is an evasive one.
It’s worth pointing out that the statement is about how much of Claude Code is written with it and not how much of the codebase of the whole company. In the more critical parts of the codebase where bugs can cause bigger problems, I expect a lot less code to be fully AI generated.
Standard CxO mentality. “I think the facts about our product might be true but I won’t say it because the shareholders and SEC will hang me when they find out it’s bullshit.” Then defer to next monkey in circus. By which time the tech press, which seems to have a serious problem with literacy and honesty (gotta get those clicks) extrapolates it for them. Then analysts summarise those things as projections. Urgh.
The other tactic is saying two unrelated things in a sentence and hoping you think it’s causal, not a fuck up and some marketing at the same time.
Did he ask Boris?
This guy is so full of shit. Anthropic’s leadership are all talk and hype at this point. And they’re not the only ones guilty of this in this hype cycle by far.
> If AI is a threat to software engineering, I wouldn't expect many software engineers to actively accelerate that trend.
This is a naive take. Throughout history things have been automated with the help of professions who were being automated away.
I don't disagree it's a naive take, and I would love to read about some examples where this happened.
I haven't seen too many industries be automated away first hand, and I'm sure there are historic examples. I wouldn't expect the lamp lighters to have been championing the rise of electric lamps. Maybe they did though because it meant they could work less hours.
Software developers have been automating away software developers since the beginning of the field. Higher level languages, better+safer languages, improved IDEs. All the ways in which we made developing software easier and more efficient. QA teams investing in automated testing.
At some point, if this process continues, it is expected to actually automate software developers, would you agree?
It's everywhere.
One example is probably blacksmiths. Before, each larger village used to have one. Nowadays, this is not the case any more as people order replacement parts for whatever metal piece that broke in your car, house, etc. The metal molding happens in factories instead. But in order you still need to have people who know how to mold metals. The early folks for this profession were blacksmiths. Nowadays you would not call them any more. But it's definitely been "engineers with AI".
Same goes for industrialized food. The huge industrial bakeries still employ people who know how to bake bread, but they can do 1000x more than they could do if they had to knead the dough on their own, separate it, form little pieces of bread, order it nicely on a tray, etc.
Now it's a machine doing most pieces of work, but they still have folks who design new bread products, adjust machines, know what to change in the formula if one particular product from one particular supplier is not available in the quantities needed, etc. A lot of the skills needed aligns enough with the traditional job of a baker that these factories employ actual bakers.
As for software, whenever I use AI tools, I don't really feel that they help me save time much. But it's entirely possible that this is going to change in the future. They are only going to get better. And of course a lot of people say it's useful for them, so it already affects folks. Just because something is not useful to me doesn't mean it's not going to be useful to anybody.
Yeah, this needs to be considered down to the individual level. If your employer not only incentives you to go against your best interests but also threatens the stability of your role, then your choices are either do the job or accept that you might not be reliably employed. This is the culmination of decades of moving power from the employee to the employer.
> If AI is a threat to software engineering, I wouldn't expect many software engineers to actively accelerate that trend. I personally don't view it as a threat, but some people (non engineers?) obviously do.
Software engineers have been automating away jobs for other people for nearly a century. It would be quite rich if the profession suddenly felt qualms about the process! (TBC I think automation is great and should always be pursued. Ofc there are real human concerns when change happens quickly but I am skeptical that smashing the looms is the best response)
Software engineering has also been automating its own jobs for ages. The first thing we engineers do when asked to do a repetitive thing is find ways to automate it. I think the industry had qualms about losing their jobs. But honestly what are the examples of people losing their jobs to software ? Everybody says that this has happened many times yet examples are hard to come by.
> But honestly what are the examples of people losing their jobs to software?
And furthermore, what is the full causality chain that links the precise PR in provided example software to the employment termination decision? Lacking that, can you really assert the software 'automated' a dev out of the job?
Honeslty I would even accept statistical test (but conclusive, no p-hacked bullshit). Yet I just dont see examples of devs being automated out of a job.
> But honestly what are the examples of people losing their jobs to software ?
Bank tellers
Travel agents
Cashiers
Bookkeeping clerks
Typists
Many of these still exist today, though is not as large of numbers. Maybe it's the same for software engineers. I still don't recall hearing of bank tellers accelerating the pace of ATMs or Travel Agents encouraging their clients to use Expedia.
> If AI is a threat to software engineering, I wouldn't expect many software engineers to actively accelerate that trend.
There are two strong forces at play. Employees generally want to put in the least amount of effort possible and go home at 5. Employers want to save money and pay for fewer employees. AI creates a strong symbiosis here and both sides are focused on a short term win.
not just employees, founders as well :) as a serial founder, I find AI exciting because after my 3rd company, thinking of my 4th was quite exhausting but AI has re-invigorated by ambition to start another company.
>Employees generally want to put in the least amount of effort possible and go home at 5.
Really?
In my experience coding engineers are coders because they enjoy the challenge. they like building things, breaking things and rebuilding things.
Its the white-collar 9-5 office workers that churn out power-point and word documents that want to do as little as possible.
Sometime very soon we’ll cross a threshold where most people can do most of their coding through a tool like Claude and be more productive. It will feel like coding still, breaking and building things, but they will get more done in the same time. Everyone will switch.
I'm already building whole projects now with AI -- not a single line of code by myself.
I love it, I've coded for 25 years and I don't need to write another line ever again, I just want to build cool things as fast as possible.
If you work at a startup, sometimes you have to do extremely mundane and boring things unrelated to your expertise (“wear many hats”). AI is especially useful in those cases so you can quickly go back to working on things you enjoy.
This is like asking me "how much of your software is built by the compiler?" -> the answer is 100%.
Ask "how much did you build then?" -> also 100%.
The compiler and I operate on different layers.
I don't work for OpenAI and I doubt some random employee is going to come here and share what is likely a secret. I'm in the industry though so I have some idea of what's going on these days, both where I work and more broadly.
AI is getting better at writing code. However writing code is just some fraction of the work of many software engineers. AI doesn't work independently, it needs to be guided, its work needs to be reviewed, tested etc. There are some domains where it does better and some domains where it doesn't. There's a range of "AI" work between auto-complete style work, assisting in understanding a code base, and writing code from some spec or doing other types work.
All in all I would say it's a decent improvement to productivity for many situations. It's really hard to say how much and it's also not a zero sum game, as productivity improves there's more work.
Something to keep in mind is that if you look at a modern software project likely most of the code executing is not code written by the developers of that project. There's a huge stack of open source bits executing for almost any new project.
Specifically in OpenAI you also need to consider what type of software they are likely writing. Some of it may be more or less "vanilla" code and other is likely very specialized/performance critical. The vanilla code like API wrappers or simple front end pieces is likely more amenable to be written by AI whereas the more cutting edge algorithmic/scheduling/optimization work is almost certainly not done by AI. At least yet.
As software organizations become larger there's a lot of overhead and waste. It is possible that AI can enable smaller teams and that has a multiplicative effect because it lets you reduce that waste/overhead. There are likely also software engineers who will become better/adapt to new workflows and some who will not. It's really hard to say where things are going but overall my sense is that this like many other innovations will lead to more software and more jobs and not the other way around. There are many moving pieces here, not just AI itself but geopolitics, macro-economics, etc. Where are those new jobs going to get created, what new types of software/technology are going to be created etc. etc. History seems to show us that we'll adapt/evolve and grow.
I absolutely believe that a large proportion of new code written is at least in-part AI generated, but that doesn't mean a large proportion of new code is 100% soup-to-nuts/pull-request-to-merge the result of decisions made by an agent and not a human. I doubt that very much.
I think the difference between situations where AI-driven development works and doesn't is going to be largely down to the quality of the engineers who are supervising and prompting to generate that code, and the degree to which they manually evaluate it before moving it forward. I think you'll find that good engineers who understand what they're telling an agent to do are still extremely valuable, and are unlikely to go anywhere in the short to mid term. AI tools are not yet at the point where they are reliable on their own, even for systems they helped build, and it's unclear whether they will be any time soon purely through model scaling (though it's possible).
I think you can see the realities of AI tooling in the fact that the major AI companies are hiring lots and lots of engineers, not just for AI-related positions, but for all sorts of general engineering positions. For example, here's a post for a backend engineer at OpenAI: https://openai.com/careers/backend-software-engineer-leverag... - and one from Anthropic: https://job-boards.greenhouse.io/anthropic/jobs/4561280008.
Note that neither of these require direct experience with using AI coding agents, just an interest in the topic! Contrast that with many companies who now demand engineers explain how they are using AI-driven workflows. When they are being serious about getting people to do the work that will make them money, rather than engaging in marketing hype, AI companies are honest: AI agents are tools, just like IDEs, version control systems, etc. It's up to the wise engineer to use them in a valuable way.
Is it possible they're just hiring these folks to try and make their models better to later replace those people? It's possible. But I'm not sure when in time, if ever, they'll reach the point where that was viable.
>> new code is 100% soup-to-nuts/pull-request-to-merge the result of decisions made by an agent
I am beginning to have more success with this in simpler parts of the code. Particularly if you already have a good example of how to do something and you need something very similar. I usually have to do a few tweaks but generally quite useful.
A large proportion of code written a quarter century ago was also in part AI generated. IntelliSense is AI and it's been around since the 90s.
I would argue intellisense is far closer to a select statement than "ai" anything. What could come after myString.s is <<select method_name from all_methods where type_name = 'String' and method_name like '%s%'>> where some IDEs prefer <<like 's%'>> and others the <<contains>> style
IJ does some truly stellar introspection to offer sane defaults in the current completion context, such as offering only variables of the correct type for parameters but I think of that as discipline and not AI. Plus, IJ never once made up an API that didn't exist
I don't work at Open AI but I use Codex as I imagine most people there do to.
I actually use it from the web app not the cli. So far I've run over 100 codex sessions a great percentage of which I turned in to pull requests.
I kick off codex for 1 or more tasks and then review the code later. So they run in the background while I do other things. Occasionally I need to re-prompt if I don't like the results.
If I like the code I create a PR and test it locally. I would say 90% of my PR's are AI generated (with human in the loop).
Since using codex, I very rarely create hand written PR's.
IME I have had to review a lot of code written by AI. At certain almost all of it. And sometimes write the code myself because LLMs just don't get it. AI has written 95% of my code but not without any review.
Estimates will always be off compared to a plugin like wakatime tracking the real amount of AI generated code vs human written code.